T-SHAPED TEAMS

ORGANIZING TO ADOPT AI AND

BIG DATA AT INVESTMENT FIRMS
CFA Institute is a global community of more than 171,000 investment professionals working to build an
investment industry where investors’ interests come first, financial markets function at their best, and
economies grow. In the mainland of China, CFA Institute accepts CFA® charterholders only.

© 2021 CFA Institute. All rights reserved.

No part of this publication may be reproduced or transmitted in any form or by any means, electronic or
mechanical, including photocopy, recording, or any information storage and retrieval system, without permission
of the copyright holder. Requests for permission to make copies of any part of the work should be mailed to:
Copyright Permissions, CFA Institute, 915 East High Street, Charlottesville, Virginia 22902. CFA® and Chartered
Financial Analyst® are trademarks owned by CFA Institute. To view a list of CFA Institute trademarks and the Guide
for the Use of CFA Institute Marks, please visit our website at www.cfainstitute.org.

CFA Institute does not provide investment, financial, tax, legal, or other advice. This report was prepared for
informational purposes only and is not intended to provide, and should not be relied on for, investment, financial,
tax, legal, or other advice. CFA Institute is not responsible for the content of websites and information resources
that may be referenced in the report. Reference to these sites or resources does not constitute an endorsement
by CFA Institute of the information contained therein. Unless expressly stated otherwise, the opinions,
recommendations, findings, interpretations, and conclusions expressed in this report are those of the various
contributors to the report and do not necessarily represent the views of CFA Institute. The inclusion of company
examples does not in any way constitute an endorsement of these organizations by CFA Institute. Although
we have endeavored to ensure that the information contained in this report has been obtained from reliable
and up-to-date sources, the changing nature of statistics, laws, rules, and regulations may result in delays,
omissions, or inaccuracies in information contained in this report.

Photo credit: Getty Images / Cultura / Monty Rakusen

ISBN: 978-1-956563-10-8
T-SHAPED TEAMS
ORGANIZING TO ADOPT AI AND
BIG DATA AT INVESTMENT FIRMS
CONTENTS

Executive Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Chapter 1. T-Shaped Teams and Applying AI and Big Data in Investments . . . . . . . . . . . . . . . . . . 4
From T-Shaped Skills to T-Shaped Teams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
The Innovation Function: The T in T-Shaped Teams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
The Investment Function: Do Future Buffetts Need to Know Python? . . . . . . . . . . . . . . . . . . . 7
The Technology Function: Where to Find Them, and Do They Need to Know DCF? . . . . . . . . . . . . 9
Co-Location, Iterations, and Regular Reviews: Facilitating the Communication between
Investment and Technology Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Team vs. Function and Supporting Business Processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Chapter 2. The Early Stage on the AI and Big Data Adoption Journey . . . . . . . . . . . . . . . . . . . . . . 13
What Defines Early-Stage Financial Institutions on Their AI and Big Data Journey? . . . . . . . . . . . . . 13
Senior Management Support . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Wholistic Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Organizational Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
What Should Early-Stage Firms Focus On? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Build a T-Shaped Team . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Develop Supporting Processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Find and Score Quick Wins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Case Study: UOBAM’s T-Shaped Team at Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Investment Philosophy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Roles and Job Scope of the Team Working on AI/ML and Big Data . . . . . . . . . . . . . . . . . . . . . 15
Workflow in AI/ML Adoption and Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
A Case within the Case: The Multi-Signal Predictive Analytics Project . . . . . . . . . . . . . . . . . . . 18
Engagement of Fintech Companies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Providing a Learning Environment in UOBAM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
Key Takeaways . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
T-Shaped Teams

Chapter 3. The Intermediate Stage on the AI and Big Data Adoption Journey . . . . . . . . . . . . . . . . . 22
What Defines Intermediate-Stage Financial Institutions on Their AI and Big Data Journey? . . . . . . . . 22
What Should Intermediate-Stage Firms Focus On? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
Building Out the Team . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
Honing the Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
Juggling Multiple Projects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
To Centralize or Not to Centralize? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Case Study: NN Investment Partners and the T-Shaped Team at Work . . . . . . . . . . . . . . . . . . . . 23
Company Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Investment Philosophy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
The Organizing Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
The Innovation Team . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
Roles and Responsibilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
The Innovation Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
The Stage-Gate Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
Impact of Innovation Process: Enhancing Existing Processes . . . . . . . . . . . . . . . . . . . . . . . 28
Impact of Innovation Process: Exploring Emerging Technology and Experimenting . . . . . . . . . . . 29
Chapter 4. The Advanced Stage on the AI and Big Data Adoption Journey . . . . . . . . . . . . . . . . . . . 30
What Defines Advanced-Stage Financial Institutions on Their AI and Big Data Journey? . . . . . . . . . . 30
What Should Advanced-Stage Firms Focus On? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
Case Study: T-Shaped Teams at Man Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
Man AHL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
Man GLG . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
The Man Data Science Team and Other Centralized Resources . . . . . . . . . . . . . . . . . . . . . . . 36
Key Takeaways . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

PL Qualified This publication qualifies for 1.25 PL credits under the guidelines
Activity of the CFA Institute Professional Learning Program.

iv | CFA Institute
EXECUTIVE SUMMARY

• The investment industry is increasingly capabilities to improve investment processes

recognizing the benefits of AI and big data and outcomes.
technologies. Cost, talent, and technology are • The role of the T-shaped team evolves through
significant hurdles in AI adoption, as is the time the early, intermediate, and advanced stages of
needed before results can AI and big data adoption. We
materialize. But arguably have included a case study
the most critical factor for Senior executives must for each stage to illustrate
success is leadership vision, develop and implement a the key issues financial
including the supporting comprehensive strategy to institutions face and their
organizational structure and solutions. These also show
introduce AI and big data
culture put in place by senior the real-world application and
into their operations. In this
leadership to enable change. practicality of the T-shaped
process, senior executives
• Senior executives must need to build a T-shaped team. team as a key organizing
develop and implement a feature for firms successfully
comprehensive strategy to embracing AI and big data.
introduce AI and big data into their operations. • The T-shaped team is not a one-size-fits-all
In this process, senior executives need to build concept. On the contrary, we expect firms to
a T-shaped team made up of three functions— develop their own T-shaped team structure
investment, technology, and innovation. and processes that best meet their needs and
• The “T” alludes to (1) the transition from T-shaped capabilities, particularly when they approach the
skills to a T-shaped team, as AI adoption in intermediate and advanced stages.
financial institutions is far more complicated • Whatever the stage of AI and big data adoption,
than an individual can handle, and (2) the fact this report will illuminate for investment industry
that investment skills and technology skills leaders how they can, through leadership and
are “orthogonal” to each other, such that they innovation, put themselves on a path to success
need (3) the innovation function, which sits at as AI and big data become embedded in the
the intersection of investment and technology investment industry.
functions, to bridge the gap in communications
and effectively leverage organizational

© 2021 CFA Institute. All rights reserved. CFA Institute | 1

INTRODUCTION

Financial institutions globally are eagerly looking leading to (3) the innovation function, which sits at
to embrace AI and big data technologies. Not to the intersection of the investment and technology
underestimate the cost, talent, and technology hurdles functions.
in AI adoption, we believe leadership vision is the
The apparent need for the innovation function is in part
single most important factor to kick-start the process.
the result of the fact that AI and data science have
Senior executives must take the lead at this stage
become major disciplines of scientific investigation.
and become champions of AI and data science: Only
They take years of education (perhaps as many as
they can ensure that the organization develops and
10–15 undergraduate to master’s level courses) and
implements a comprehensive AI strategy.
work experience before someone can contribute in
In this process, senior executives need to build what a meaningful way and lead AI and big data adoption
we call a T-shaped team—particularly its technology projects at financial institutions.
and innovation functions—staff the team with proper
The implication is that we cannot expect the majority
talent, allocate enough resources to their projects,
of investment (technology) professionals to know
and ensure alignment and support of the top and
enough about technology (investments) to be a core
mid-level business and technology executives at the
member or leader of the technology (investment)
organization.
function. Training programs will take years to see
Once the team is in place, the key challenge becomes impact, assuming low turnover, staff persistence,
connecting the investment effective training, collaborative
function and technology function work environment, and so on.
effectively so that the organization
Hence the need for the T-shaped
can dedicate its resources to the In AI adoption, we believe
team.
best projects—those that can leadership vision is the single
deliver the largest measurable most important factor to kick- In Chapters 2 through 4, we
results with the least input in proceed to discuss the T-shaped
terms of resources and time.
start the process. teams’ evolution through early,
This challenge remains constant intermediate, and advanced
through all three stages of AI stages of AI and big data adoption.
adoption, hence the need for the innovation function to Included also are three real-world cases—one for
ensure that the team can collaborate effectively. each stage—in which we share the live experiences
of T-shaped teams at three medium-sized and large
In Chapter 1 of this report, we build on the discussion
global/regional asset management firms on their AI
in “Investment Professional of the Future”1 and “AI
and big data adoption journey.
Pioneers in Investment Management”2 and explain in
detail the T-shaped team concept and accompanying Looking through these three stages reveals how the
processes. relative burden across the technology, investment,
and innovation functions shifts over time.
We use the term T-shaped team to specifically
refer to this particular team structure (including In the early stage, the key is to find a project that
investment, technology, and innovation functions) will showcase to the investment function the value
and accompanying processes used by financial and power of the AI and data science tools. After
institutions in the adoption of AI and big data all, the technology is formative and naturally there
technologies. will be doubts from the investment folks about the
effectiveness of AI, as well as some anxiety about how
The T alludes to (1) the transition from T-shaped
this might affect their own career.
skills to a T-shaped team, as AI adoption in financial
institutions is far more complicated than an individual That puts the burden on those in the technology
can handle, and (2) the fact that investment skills function to identify projects that they have a high
and technology skills are “orthogonal” to each other, degree of confidence in delivering in a reasonable

See CFA Institute (2019b).

1

See CFA Institute (2019a).

2

2 | CFA Institute © 2021 CFA Institute. All rights reserved.

 T-Shaped Teams

period of time. Ideally the project does not require too organizations. In the advanced stage, the T-shaped
much input from the investment function but can help team will finally be able to realize its full potential,
the investors do a better job. The innovation function with both investment and technology functions
needs to be able to evaluate whether the projects understanding each other much better, sharing their
the technology function proposed fit that criteria, in opinions openly, and communicating more frequently
particular if the projects can deliver meaningful results as needed. In this ideal stage, the investment function
to the investment function. Ideal projects are usually will become more effective; hence, we would say the
those that provide analytical support and individual relative burden of success is more on the investment
security monitoring. professionals in this stage.
If the technology function successfully pulls off the We hope the likely scenario that we described is
first AI and/or big data application to the satisfaction an important revelation for investment managers.
of the investment function, then the team can expect Technology will play a more important role in our
more input from the investment function down business, and the data science function will become
the road. Of course, sometimes projects fail. Most a permanent part of the investment team. However,
“spontaneous” AI efforts at financial institutions that instead of being apprehensive of their presence, those
lack organizational backing will who embrace the technological
likely not come back from these change will more likely become
early failures. This is where winners in the industry in the long
the T-shaped team will play an Instead of being apprehensive term.
important role in pulling the of their presence, those who
This report is intended to have
team and organization together embrace the technological a “how-to menu” flavor, which
to continue looking for and change will more likely we believe is most beneficial for
implementing the next promising
projects.
become winners in the investment organizations and
industry in the long term. industry leaders looking to build
Few financial institutions in their own T-shaped teams.
the world are operating in the
Several recent industry surveys
intermediate stage today. For
and research reports have come to reveal more
those that do successfully enter the intermediate
supporting evidence for the T-shaped team concept.
stage, the focus will likely be on balancing the input
We have referenced these findings in Chapter 1. Our
from both the investment and technology functions.
conversations with investment professionals around
The focus is on the innovation function leader’s ability
the world have also provided impetus to this work. We
to leverage a small number of initial wins in a small
hope that it will help expedite the industry’s efforts
number of areas to gain credibility within the wider
at applying AI and big data in its daily operations
organization. When more people from the investment
as well as help investment professionals anticipate
function become more open in asking the technology
industry development trends so that they can prepare
function for support in decision making, such as
themselves accordingly.
security selection and portfolio construction, then the
team will really start to deliver its full potential. We welcome feedback from readers about your
experiences. Please email your comments and
It may take a long time for some financial institutions
perspectives to the author at larry.cao@cfainstitute.
to get to the advanced stage, and we believe getting
org or our team at [email protected].
there will be crucial for the long-term survival of

CFA Institute | 3
CHAPTER 1. T-SHAPED TEAMS AND APPLYING AI
AND BIG DATA IN INVESTMENTS

AI is a must-have. This prospect has become ill-suited for investment firms adopting AI and big data.
increasingly evident in investment circles. With Many industry executives have come to the same
growing evidence of AI’s effectiveness in helping realization over the last year.7
investment professionals in their daily work, the
We defined a T-shaped team to be a specific team
focus has now shifted from building conviction
structure that helps investment (or business)
about AI and big data to successful adoption of
organizations in adopting AI and big data technologies
the technologies.
into their core processes (see Figure 1). We initially
The challenges of making AI and big data applications used the concept in the context of discussing what
a reality remain. Our data suggest that investment a future investment team would look like (i.e., what
firms have had mixed success at introducing AI and new roles and skills would be needed and how
big data into their operations.3 Other industry surveys professionals on the team would interact with each
also support our findings.4 other; see Figure 2).
We find that the failed experiences often are related
to the five major hurdles we outlined in our previous FIGURE 1.  T-SHAPED TEAM
research—namely (in increasing order of difficulty
to overcome and/or scarcity of the resource), cost,
talent, technology, leadership vision, and time.5 TECHNOLOGY
Those firms that are successful often appreciate
the strategic impact of AI and big data, understand
the cost implications, and have been able to
recruit the right talent. Above all, however, they INNOVATION
have been successful at addressing the related
organizational issues.
Central to those organizational successes is an
approach that we called the “T-shaped team.”6

FROM T-SHAPED SKILLS

TO T-SHAPED TEAMS INVESTMENT

Suggesting an organizational approach as the

solution to adopting AI and big data at investment
firms may come as a surprise to many. The reason
is that organizational issues are really closely
tied to the hurdles that we have described. Put
another way, many of these hurdles are the result of
organizational structures and processes that were Source: CFA Institute (2019a).

3
For details of the survey and the hurdles to adopting AI and big data, see CFA Institute (2019a).
4
See, for example, the NewVantage Partners report “Big Data and AI Executive Survey 2020,” a survey of C-suite executives of leading global
firms, which indicated that 91.5% of the firms reported ongoing investments in AI but only 14.6% have deployed AI capabilities in widespread
production (NewVantage Partners 2020).
5
See CFA Institute (2019a).
6
For details, see CFA Institute (2019a, 2019b).
7
For example, Gartner’s Fifth Annual Chief Data Officer Survey, as reported by Goasduff (2020), suggested that chief data officers believe
“culture challenges to accept change” is the most critical roadblock to their success. Similarly, the NewVantage Partners Big Data and AI
Executive Survey (see Bean 2020) reported that 90% of executives cited people and process issues as the principal obstacle that they face.

4 | CFA Institute © 2021 CFA Institute. All rights reserved.

 T-Shaped Teams

FIGURE 2. ROLES AND FUNCTIONS OF A T-SHAPED too small and the stage of development too early to
TEAM form any real consensus—this is how innovation tends
to take place: spontaneously.
To develop AI and big data capabilities into a
SIGNIFICANT TYPES OF ROLES AT sustainable competitive edge, however, requires far
INVESTMENT FIRMS OF THE FUTURE more resources and structure than spontaneous
efforts can count on. That is why successful firms
have burdened themselves with the task of building
T-shaped teams to take on the challenge. The complex
INVESTMENT nature of these projects cannot usually be taken on
Investment decision maker by one “wizard.” They require a significant number
Investment researcher
Private wealth manager of professionals with very specific and in-depth
knowledge in their respective area working together.
The T-shaped team structure and process are
designed exactly for such purposes.
TECHNOLOGY
Data scientist
Applications engineer The Innovation Function:
The T in T-Shaped Teams
The link between T-shaped skills and T-shaped
INNOVATION
teams is that in both cases we have different skills
Investment thinking and process innovator
Knowledge engineer represented. The difference is that in the former
Innovation facilitator case, the skills are embedded in the individuals, and
in the latter case, the team. The T means the skills
are different.9,10 The fact that they are so different
Source: CFA Institute (2019b).

A T-shaped team is made up of three functions. In

addition to the traditional investment function, future About the T-Shaped Team
(or, more appropriately, future-proof, as the “future” is T-shaped teams are project teams. We believe,
arguably already upon us) investment teams will also however, that AI and big data integration in financial
have a technology (e.g., data science) function and an institutions is a long-term objective. These projects
innovation function. are different from typical short-term information
In the early days of AI and big data adoption, efforts technology (IT) projects.
are often driven by individuals with T-shaped skills;8 Figure 1 is intended to illustrate the relationships
these individuals have deep subject matter expertise of the three functions: The innovation function
combined with a wide understanding of other domains connects the very different investment and
and an ability to connect the two. Often, these technology functions to form the future investment
individuals are either investment professionals who team. Figure 1 is not an organizational chart in its
happen to have some relevant data science training traditional sense.
or technology professionals with data science training
who happen to work closely with the investment Three cases embedded in Chapters 2 through
function. Early efforts tend to be less “planned” and 4 illustrate the relationships between the T-shaped
have more of a “spontaneous” flavor. teams and the functional departments and their
evolution over time. The functional linkage will
There is nothing wrong with such experiments. With become more superficial as time goes by if the
textbooks or procedural manuals on the subject T-shaped teams’ goals are properly defined and
lacking—since the number of professionals who are staffing is properly managed.
privileged to have such experience in the industry is

8
For more details on T-shaped skills, see the Hansen (2010) interview of IDEO CEO Tim Brown.
9
In fact, in linear algebra, orthogonal vectors are independent of each other, much like the lack of common training and experience of
investment professionals and data science professionals.
10
Another difference is that the T in T-shaped team no longer only implies depth in one dimension and breadth on the other dimension.

CFA Institute | 5
T-Shaped Teams

gives rise to the small T in the T-shaped teams: the at a buy-side firm between investment staff and
innovation function. We illustrated it with a small T marketing staff is far easier to handle by its nature;
joining the two perpendicular functions since this is at least many of them attended business schools,
the function where a firm will absolutely benefit from which offered broadly similar curricula. In comparison,
the staff’s T-shaped skills. The innovation function the collaboration between the investment/business
is also the function whose key responsibility is to function and the data science function is much more
connect the two main functions. challenging and also much more mission critical.
Investors and programmers have little in common in
Our first inspiration for highlighting the importance
terms of skills and culture and, as such, need much
of the “innovation” function (CFA Institute 2019b)
more coordination.
came from a fascinating book titled The Future of
the Professions (Susskind and Susskind 2015) on The old model of collaboration between investment
how technology is reshaping the legal and medical staff and IT departments is unlikely to be able to serve
professions.11 Table 1 summarizes the generic roles the purpose of AI adoption at investment firms.12 Thus,
the innovation function is expected to play. We have we are concerned with whether a business analyst
since come to more fully appreciate the critical can play the role of the innovation function effectively
role this function plays in dialogue with investment since it is reminiscent of an old and failed model.
professionals around the world.
In certain cases the innovation function only needs
In the context of AI and big data adoption at to play the role of a facilitator. However, we suspect
investment firms, the innovation function’s focus is innovation function staff will need to be much more
on facilitating the collaboration between investment than mere “translators.” Their important responsibilities
and technology functions. Thought leaders may be include not only communication between the team
either internal executive sponsors or external experts. and the C-suite project sponsors/initiative leaders
Knowledge engineers usually play the role of both but also setting team goals and firm AI and big data
a strategic planner and project manager. Innovation adoption strategies in consultation with the leadership
facilitators tend to focus on the communication and and the team. More importantly, they need to lead
relationship management aspect of the process. the discussion around what areas of collaboration
to focus on and which projects should get priority.
Although the T-shaped team framework can be
Above all, their role is to get both sides to contribute
applied in general cross-functional collaboration,
to the discussion in a meaningful way. In short, the
it is particularly suitable for the current context
innovation function sets the tempo of the whole team
(i.e., investment firms incorporating AI and big data
and is accountable for the overall success of projects.
into their core processes). The usual collaboration

TABLE 1. INNOVATION FUNCTION AND THE THREE TYPES OF ROLES

Role Generic Description
Innovation function The function that connects investment and technology functions. Focus on
improving the investment process.
Investment thinking Thought leaders are usually found both among senior researchers/executives
and process innovator in the industry and at leading research universities.
Knowledge engineer Subject matter experts identify key industry trends and emerging investment
expertise. They focus on gathering insights from innovators and sharing them
with investment professionals.
Innovation facilitator Generalists who also play a relationship manager role, working with
stakeholders in the sourcing and distribution of insights.
Source: CFA Institute (2019a).

We borrowed the term “knowledge engineer,” for example, from them.

11

See, for example, Squirrel and Fredrick (2020) for discussion on the subject.
12

6 | CFA Institute
 T-Shaped Teams

The ideal skill set for innovation function leaders is reasons. It’s a bit more complicated than simply
a 50–50 split between investment and technology, hiring a group of fresh graduates with degrees in
which allows them to have sufficient grasp of the AI or data science.
subject matters to understand the thought process of
On the positive side, more and more firms in the
both functions and put their opinions into perspective.
business are moving in the right direction, albeit from
For this role to be effective, the leaders need to have
a low base. Various consulting firms working to help
credibility with both investment professionals and
clients incorporate AI and big data into their business
data scientists and engineers. They also need to be
processes have since written about roles very similar
good communicators who listen attentively, speak
to the innovators we describe here. A recent Deloitte
succinctly, and write clearly. The toolkit ideally should
survey suggested that in addition to AI builders,
also include strategic planning skills such that the
executives of early adopters are also seeking “AI
innovation leaders understand both the strategic
translators,” including business leaders, “who bridge
direction of the firm and the strengths and weaknesses
the divide between the business and technical staff”
in the investment and technology functions and are
(Hupfer 2020, p. 7). An article by three business and
able to devise strategic plans accordingly.
data analytics experts at McKinsey referred to “a
These skills can come from classroom training, but at relatively new class of expert, analytics translators,”
the functional leader level and above, they can only and these McKinsey authors expected the translators
be honed from experience. In reality, neither colleges to “play a role in identifying roadblocks” and to “bridge
nor financial institutions have typically offered training the data engineers and scientists from the technical
programs for this role. That gap makes it particularly realm with the people from the business realm”
challenging to find the right candidate—especially the (Fountaine, McCarthy, and Saleh 2019).
first member of the innovation
Note that these “job descriptions”
function, who usually has to play
still fall short of what we expect
a combination of the three roles Our education system and of the innovation function leader.
with likely emphasis on facilitator
financial institutions will have Interestingly, the Deloitte survey
and knowledge engineer.
to change before we can found that the executives’ hiring
At this juncture, the industry have an abundant supply of priorities shift from AI builders in
is largely behind the curve the early stage of AI adoption to,
in appreciating the value of
both capable and successful as the organization gains more AI
the innovation function (and innovation function leaders. experience, business leaders who
particularly the innovation can “[figure] out what results from
function leader). Consequently, AI systems mean, and how those
the industry does not have a career path designed or results should factor into business decisions and
training programs prepared for these types of roles. In actions” (Hupfer 2020, p. 8).
our experience, when early adopters started offering
We have every confidence that the industry will
such roles, successful candidates usually did not
continue to mature in the coming years and move in
specifically train for the positions. Firms come across
the right direction. The challenge in finding competent
this rare species of professional with experience on
innovation function leaders is not surprising in
both sides more out of serendipity than by design.
some ways. AI in investments is very new, after all.
With the majority of the industry still unclear about Our education system and financial institutions will
the (potential) demand for the innovation function, have to change before we can have an abundant
it is hard to imagine that the education system will supply of both capable and successful innovation
supply the right candidates. Fintech programs offered function leaders. It is crucial for the industry’s long-
by the early entrants into this space (i.e., certificates term success that all constituents in the ecosystem
offered by various industry associations and degrees support this development.
offered by universities in recent years) may not have
been designed for the needs of a future member of The Investment Function: Do Future
the innovation function.
Buffetts Need to Know Python?
The nuances around the supply of and demand for
innovation talent offer a good example of the talent The investment function in the T-shaped team is
hurdle mentioned previously. Talent shortages can largely the same as what we have today. At the
take many shapes and forms and can occur for many outset of embarking on the journey of adopting AI in

CFA Institute | 7
T-Shaped Teams

investments, however, it is important to embrace the training and experience. The structure and process
AI + HI (artificial intelligence + human intelligence) will also ensure that these models are well-tested and
philosophy. Investment professionals’ investment reviewed from different perspectives before they are
expertise is valuable, and artificial intelligence will put into production. Not having gone through such a
help them expand their horizons and/or augment rigorous process leaves much room for error.
their capabilities.
Even if the professional is one of the rare “wizards”
Anthony Ledford, chief scientist at Man AHL, shared who can do both investment and data science well,
a story about his firm’s early venture into machine single-handedly launching such an effort will likely
learning (ML). He explained that they set out to develop leave him or her with little organizational support.
a long–short trading signal using support vector So, when they fail, which is likely, there often are no
machines (SVMs), an ML method, and the team ended backup resources to develop an alternative model
up “discovering” relationships that they already knew! or organizational will to continue pursuing such
projects. These pioneers could inadvertently set
Do portfolio managers and analysts need to be able to
back the organization’s AI journey by years, which
use Python?13 The answer to this question illustrates
could mean the difference between the firm’s long-
the fundamental difference between T-shaped skills
term survival or demise. Even when they succeed,
and T-shaped teams. In the T-shaped skill setting,
people often underestimate the commitment required
which worked well in the pre-AI age, the answer is
to regularly update the models. It is certainly not a
obviously “yes.” In the T-shaped team setting, however,
part-time job.
the answer is a much more ambiguous “maybe.”
A stellar staff of the investment function should
The reason is that we do not believe the majority
understand the AI + HI philosophy and as such be
of investment professionals can
interested in participating in AI
operate at the level of a good data
adoption projects. They would
scientist or engineer even if they
choose to devote a significant AI in investments is a also ideally have a strong sense
serious commitment and of curiosity, both about how to
amount of time to it. Data science
improve their own investment
has evolved to become a major not something that one process and how AI can help.
discipline,14 not something that professional with T-shaped Learning Python can certainly help
someone from a completely
skills in investment and data them communicate with the data
different background can easily
pick up without a formal education science can handle. science team or at least appreciate
the hard work that the team does.
and, more importantly, on-the-
To operate at the level of a data
job training to gain the relevant
science professional is, of course,
experience.
a completely different question.
AI in investments is a serious commitment and not
T-shaped teams like other project teams are typically
something that one professional with T-shaped skills
staffed by professionals with different skill levels.
in investment and data science can handle. We have
Three to five courses in computer science and AI
come across many cases where such professionals
may be sufficient for a fresh graduate to start as a
failed to pull off an AI adoption project. Worse, if they
junior data engineer. For a mid-career investment
succeeded in launching a test model, chances are
professional to switch to the technology or innovation
they have not gone through the serious peer review
function at similar seniority, it will likely take five
and cross-functional testing and review processes
to fifteen programming and AI courses plus five
required today.
years of work experience working on data projects.
Not surprisingly, these models are particularly We have come across very few professionals who
susceptible to failure. Part of the benefit of the have made the switch successfully, although we
T-shaped team structure is that professionals in expect the situation to improve in the coming
each function can specialize and aim to be the best decades as the current generation of tech-
in the business. Someone doing model development savvy fresh graduates become investment-savvy
essentially part time can hardly compete in terms of over time.

13
Note that this discussion is in the context of T-shaped teams (i.e., AI and big data adoption). There can certainly be many other reasons
to study Python, which are outside the scope of this report.
14
See, for example, Oh and Maldonado (2021) and Eric (2020).

8 | CFA Institute
 T-Shaped Teams

What specifically should investment professionals lead so that he or she can set the data science team’s
know about how AI can help? That question has driven agenda in line with the firm’s overall strategy.
our research efforts on AI and big data applications in
Firms more experienced in this area tend to have a
investments. Investment professionals should know
larger number of PhDs in STEM (science, technology,
what the main applications are and which AI and big
engineering, and mathematics) from top universities.
data tools are used in those applications.
What matters more than their degrees, though, is their
These training programs can be taught by experienced experience working on data projects—particularly at
staff members at organizations that have pursued AI tech giants and top firms in their respective industry,
adoption for at least a year or two. Online training or where the cutting-edge work in data science is carried
remote learning are also legitimate alternatives, and out. These applications are so new that they are
they would be more effective if taught by professionals usually not in textbooks or taught at even the best
with real-world experience working on AI adoption universities.
projects in investments. We expect business schools
AI and big data applications in finance and other
to offer quality programs for this purpose in the not-
industries are still largely a learning-by-doing process.
too-distant future, and they would serve students well
When we say there is a shortage of data science
by partnering with experienced professionals, at least
professionals, the shortage is particularly true in terms
in the beginning.
of experienced professionals, although universities
around the world are still having a hard time meeting
The Technology Function: Where demand in these areas. As Anthony Ledford, chief
to Find Them, and Do They Need scientist at Man AHL, pointed out to us in our extended
interviews with him, even these experienced data
to Know DCF? science professionals may take a few years after
Hiring data scientists is often the first step when joining investment firms to become productive at a
investment firms embark on the AI journey. The Man higher level, as they still need to learn the peculiarities
Group case in this report begins when Man AHL hired of financial data as well as integrate into the firm.
a machine learning specialist.
The data scientists will be supported by a group of
The qualification and experience of the data engineers who put their models into production. The
scientists vary by firm and its experience with AI. division of labor is that the data scientists are focused
Firms with mostly a fundamental investment function more on developing models and drawing insights out
just starting out on the AI journey may care more of data, whereas the engineers focus more on making
about the data scientists’ understanding of the sure the models operate on the firm’s technology
investment process and be less picky about the platforms. The skills required for data engineers are
science pedigree. For example, one of the first data more transferrable, and these positions may be filled
scientists UOB Asset Management hired was a CFA by some of the existing technology staff.
charterholder with a master’s degree in computer
Do data scientists need to know investments? Again,
science.
the answer is not conclusive. Confucius famously said
Data scientists and engineers clearly play an important more than 2,000 years ago that “the gentleman is no
role in AI adoption projects. Firms will do well to hire [one-purpose] vessel.” Has the world become much
data scientists who are good fits for their overall AI more complicated 2,000 years later? We believe so.
strategy. In this sense, it is advisable to have the three Although we are seeing more young professionals
functions of the T-shaped team in place and start with a STEM background take the CFA Program exams
operating more or less at the same time. If one has the than we are seeing portfolio managers and analysts
luxury of building a firm from scratch, the first hires study Python, this still has not become the norm—and
would likely be the three functional leads. In reality, maybe for a good reason.
however, the investment function is likely already in
The question is, should they try at all? Some think so in
place, and by the time the firm is ready to start building
principle. Without conclusive evidence in our specific
its T-shaped team, there usually has been at least
case one way or the other, we think the best approach
some grassroots effort at applying AI and big data in
may be a compromise. Confucius may well be right,
the investment process (most likely in a limited number
and we should value a jack of all trades if we come
of products). So, some technical talent is already in
across one. We know we may not find that many after
the organization as well. The best bang for the buck at
all, so the practical solution may be to place them,
this juncture is probably hiring the innovation function
if we do find them, in the innovation function or give

CFA Institute | 9
T-Shaped Teams

them room to play the knowledge engineer role. The members they are working with so that the innovation
fluidity of the roles indeed is another important trait platform does not become “an island of isolation.”
of a mature AI adoption team. This practice tends to become more common as
investment organizations gain maturity in their AI
experience.
CO-LOCATION, ITERATIONS, AND
Human communication is imperfect, particularly
REGULAR REVIEWS: FACILITATING between two very different groups. Investment firms
THE COMMUNICATION BETWEEN successful at adopting AI have also found it helpful to
INVESTMENT AND TECHNOLOGY have regularly scheduled project meetings. In addition
to maintaining project momentum, the primary benefit
FUNCTIONS of these meetings is for developers and end users to
Where are data science teams located both physically compare notes and make sure that they are on the
and on the organizational chart? This is another same page.
question that is crucial for the smooth operation of One of the contributors to our report on AI pioneers
AI adoption projects but is often overlooked. (CFA Institute 2019a) described their process: The
Both organizational theory and industry best practice technology function scheduled project deliverables
suggest that the technology function should be on monthly intervals when they would meet with
co-located with the investment and innovation end users from the investment function to go over
functions. Co-location maximizes the formal and their latest additions to the system. These meetings
particularly informal interactions.15 keep everyone’s eyes on the ball: The technology
function can check in to be sure that it is delivering
Hansen, Nohria, and Tierney (1999) argued that something that the investment function needs, and
companies have to choose between two strategies the investment function gets an opportunity to vet the
of knowledge management, codification, and tool before it is formally delivered and provide timely
personalization, depending on their overall business feedback as needed. This check-in avoids the risk of
model. Consulting firms such as Ernst & Young provide longer delivery cycles when programmers could have
“information system implementation by codifying spent months working on something that is different
knowledge,” while in contrast, firms such as McKinsey from what portfolio managers and analysts expected
provide “creative, analytically-rigorous advice on because of miscommunication.
high-end strategic problems by channeling individual
expertise” (p. 111).
AI adoption projects in finance are highly innovative in TEAM VS. FUNCTION AND SUPPORTING
nature, not to mention the communication needed to
take place between two functions with significantly
BUSINESS PROCESSES
different approaches of thinking and working. Such The NN Investment Partners’ beehive map brings up
projects are arguably best suited to the category that another important organizational issue: team versus
requires “personalization.” Direct personal interaction function. Van Nieuwenhuijzen realized early on that the
is very much a necessity in this scenario. ideal organizational structure for investment firms to
In the NN Investment Partners case presented in adopt AI and big data is the (T-shaped) team and not
Chapter 3, our contributors shared their “beehive map,” the functional structure.
an illustration of the team composition that doubles Function versus team as an organizational structure
as a seating chart. This map clearly shows how the has been debated thoroughly. The challenge for
innovation platform is located in the middle of the adopting AI and big data in investments is that
floor with all individual investment product groups professionals in investment and technology functions
sitting around them. Valentijn van Nieuwenhuijzen, CIO share little in common, so the key is to get them to
(chief investment officer) of NN Investment Partners, work together on a shared project with a common goal
explained that some of the technology professionals (i.e., foster collaboration, inspire each other for project
would sit next to the investment function team ideas, discuss the value and priority of a proposed
15
Although a detailed discussion of how organizations should adapt to the post-COVID environment is beyond the scope of this report, we
highlight the need for organizations to take co-location into consideration in their future-of-work designs. We encourage readers to refer
to CFA Institute research on the future of work, which addresses the issues related to hybrid work arrangements, at www.cfainstitute.org/
research/survey-reports/future-of-work.

10 | CFA Institute
 T-Shaped Teams

project, etc.). The functional structure is apparently aligned incentives, for example, has helped in building
less material than the team construct in this context. a collaborative relationship between discretionary
portfolio managers and quantitative analysts.
One of the benefits of the T-shaped team over the
functional structure is that it can help break down Free riding,17 or social loafing,18 is a common issue
silos and align functional objectives.16 From the observed in teams. Will T-shaped teams experience
organization’s point of view, it is not about having the the same problems? It is certainly possible, although
superior functional capabilities; it is about having a most T-shaped teams are operating in their early
coherent team that can work together and develop the stages, where we think free riders will have a harder
best solutions enabled by the necessary input from time to survive. That said, T-shaped teams should
the investment function and technology function. implement the same precautionary measures long
adopted by teams around the world to mitigate this
The budgeting process and incentive structure need to
risk. These measures include clearly defining individual
be in alignment with the T-shaped team organizational
responsibilities, which in themselves should be
structure for it to work. This certainly takes time and
meaningful tasks that contribute to the overall team
planning to get right. Once a firm’s leaders decide
goal. There is also no need to do away with individual
they would like to go down the path of building
objectives, rather than shifting the focus to how
T-shaped teams to adopt AI and big data into their
individuals are contributing to the team objective.
investment process, the firm’s strategic objectives
are then translated into strategic initiatives that will Tracking the success of these AI adoption projects
be undertaken by respective teams. The budgeting seems a natural last step in the project. Everywhere
process needs to reflect that. Resources are allocated we present the AI pioneers report (CFA Institute 2019a)
to each team in accordance with and cases, we are invariably
the strategic importance of the asked this question: How much
initiatives it is tasked with taking From the organization’s did AI and big data applications
on. There obviously will still be point of view, it is not about add to the investment team’s
infrastructure investment required performance? Performance
for each function, depending on
having the superior functional measurement is easier said than
how the transition to the team capabilities; it is about done, though, largely because
structure is going. However, the having a coherent team that these applications are more often
focus will increasingly be on can work together and develop steps in the overall investment
teams. the best solutions. process that contribute to the
final decision. It is not always easy
The incentive structure is also an
to quantify the benefits.
important tool to motivate staff
to focus more on team objectives. In the traditional Our hypothesis is that the majority of these projects
functional setting, employees set their individual at major financial institutions are adding value,
goals and objectives and are evaluated on them. particularly if they have been in production for some
Absent shared goals, this type of arrangement is not time and have not been pulled. In most cases, these
conducive to team collaboration. If the organization are enhancements to an existing process; so, if
expects the employees to “do the right thing” (i.e., they do not work as well as the original process,
collaborate and focus on achieving team objectives), the portfolio managers have no incentive to continue
then incentives should be set up to motivate using them.
individuals accordingly. It is certainly conceivable—and
Still, it would be good practice to measure and
teams should be allowed such leeway—for individuals
document the added value of each incremental
to be rewarded more for achieving team goals than
project—if not in terms of portfolio risk and return,
their own objectives in at least some scenarios.
then maybe in terms of how they have improved
It is clearly also in the organization’s interest for the accuracy of the analysis. For example, if you use
employees to focus more on the team objective an enhanced analysis to predict defaults for credit
so that employees form a collaborative rather than products, you could measure how many of these
competitive relationship. Setting up shared goals and products did have problems as the enhanced system

16
For example, Squirrel and Fredrick (2020) argued that “to break down information barriers and drive transformation, business leaders must
engage directly with technologists and ask the right questions.”
17
See Albanese and Van Fleet (1985).
18
See Karau and Williams (1993).

CFA Institute | 11
T-Shaped Teams

predicted and compare that with the performance of success for an initiative such as the T-shaped team is
the original system. Much of the fundamental analysis to have the culture that will support these changes.
can be described as a “mosaic,” although there should This requires both management commitment and
always be ways to measure how much the enhanced staff support.
process has helped provide a clearer picture.19
In Chapters 2 through 4, we walk through how financial
Since we wrote about the role of T-shaped teams in institutions adopting AI and big data grow and build
investment firms’ AI and big data strategy in 2019, their capabilities through three distinct stages. These
we have continued to see evidence of its real-world case studies are not intended to be templates, but
impact. Some of our conversations with executives their experiences are somewhat typical of firms that
at asset management firms around the world have have embarked on this journey (and prevailed at
been turned into cases in this report. We have also least up to that point). The challenges they face are
seen research and surveys on the subject from a wide common to the experiences of firms at a similar stage.
spectrum of organizations, including some of the most Their solutions are promising for solving the particular
influential management consulting firms, that support problems they face.
our conviction.
We hope that their experiences, illustrated herein, will
Investment firms are slow to change, and industry help you plan your AI journey, anticipate issues, and
norms are hardly pro-innovation. The precondition of develop solutions.

19
For example, “Gartner forecasts that by 2024, 50% of AI investments will be quantified and linked to specific key performance indicators
to measure return on investment” (Goasduff 2019).

12 | CFA Institute
CHAPTER 2. THE EARLY STAGE ON THE AI
AND BIG DATA ADOPTION JOURNEY

The initial efforts to apply AI and big data at financial competitive advantage based on AI and big data
institutions are not always carefully planned applications.
or implemented with well-developed top-down
strategies. Quite the contrary, they typically start in
a spontaneous fashion with limited organizational
support.
WHAT DEFINES EARLY-STAGE FINANCIAL
INSTITUTIONS ON THEIR AI AND BIG
Often the early applications are developed by a
portfolio manager with some background in AI or DATA JOURNEY?
a programmer working closely with the portfolio
In this chapter, we will discuss what it takes to go
management team. They are in many ways “lone
above and beyond these three
wolves” in the organization.
initial types of development.
They are clearly professionals
Senior management plays the
with strong T-shaped skills
in the sense that they have
We believe the following three most important role in propelling
characteristics are what a firm from the pre-T-shaped
in-depth knowledge of
investments (programming) separate early-stage financial team stage into the early
T-shaped team stage. Overall,
and sufficient AI (investment) institutions on their AI and big we believe the following three
knowledge to pull off some data journey from those in characteristics are what separate
of the earliest applications of AI
and big data in investments.
the pre-T-shaped team stage: early-stage financial institutions
senior management support, on their AI and big data journey
Another type is the traditional from those in the pre-T-shaped
wholistic strategy, and
IT development route. Business team stage: senior management
functions develop requests organizational structure. support, wholistic strategy, and
without understanding the organizational structure.
technology required, and
IT departments develop programs without an Senior Management Support
understanding of the business logic and environment
the programs operate in. AI and big data adoption requires major and long-term
commitments in both financial and human resources.
The failure of the traditional IT development It is hard to imagine an organization making such
route is in a way what prompted the fintech commitments without senior management support.
phenomenon: Start-ups believed they could offer A top-down approach is the most efficient approach
better technology solutions with a better user for financial institutions to start thinking about
experience. These start-ups tend to be small, developing an edge in AI and big data applications
and their solutions are usually developed by and then implementing these changes.20 Senior
professionals with T-shaped skills on their staff. management’s marching orders are usually the first
These professionals have profiles similar to the “lone sign that the financial institution they manage has
wolves,” except that they become the heart and soul left the pre-T-shaped team stage and is entering the
of the fintech start-ups. early stage of systematically adopting AI and big data
All three types share one common characteristic: applications in their business.
They depend on professionals with T-shaped skills
in both investment and technology. All three types Wholistic Strategy
are at a pre-T-shaped team stage of AI and big data
adoption. As we have observed so far, all three types Buy-in from the main business and technology
have a rather poor chance of developing a sustainable functions at the financial institution is also needed.

20
A McKinsey survey (McKinsey & Company 2020) finds that respondents at AI high performers are 2.3 times more likely than others
to consider their C-suite leaders very effective.

© 2021 CFA Institute. All rights reserved. CFA Institute | 13

T-Shaped Teams

Developing a wholistic strategy that incorporates AI of both the investment and technology aspects of
and big data adoption helps ensure both long-term the business and can build credibility with both the
business success and the relevance of the AI and big investment and technology functions.
data projects. The strategy will include the financial
Their most important role is to help identify
institution’s long-term business objectives and will
areas of the investment business that need the
identify how AI and big data can help achieve these
most help from data science and help develop
objectives more efficiently. The main business and
the technology roadmap for the AI and big data
technology departments will then collaborate to make
applications that can deliver on those objectives.
these objectives a reality.
For them to do that successfully, they need to
have credibility with both the investment and
Organizational Structure technology functions.
Of course, we are talking about the T-shaped team
here. Transitioning from relying on professionals Develop Supporting Processes
with T-shaped skills to T-shaped teams is another
The T-shaped team structure is new to everybody,
critical sign that a firm has matured and is
and every firm will likely have to develop a structure
entering the next phase of its development on a
that best suits its people and business. For the
higher level.
two distinctly different functions to be able to work
together, building an innovation function alone is not
sufficient. New processes are needed to facilitate the
WHAT SHOULD EARLY-STAGE collaboration between the two functions.
FIRMS FOCUS ON? The objective of these processes is to help the
innovation function leader do
For those who have accepted what he or she needs to do (i.e.,
what AI and big data may do to
The ultimate test for early- to evaluate and prioritize AI and
enhance human intelligence in big data projects that the teams
investments, the natural question stage firms is whether
should work on and eventually
becomes, where should I begin? the team can find and help implement).
We believe that once the AI and score quick wins.
big data strategy is in place, early- Early-stage firms need to be agile
stage firms should focus on the in developing these processes and
following three objectives to begin will likely go through many rounds
its implementation: Build a T-shaped team, develop of trial and error before finding a set of processes
supporting processes, and find and score quick wins. that works for them. What makes a good process? It
needs to be helpful for all participants in achieving the
common objective. We find that effective processes
Build a T-Shaped Team help both investment and technology functions
Firms’ first order of business to implement their AI and engage. Both functions should also have “equal” say
big data strategy is often hiring data scientists and in these processes to achieve optimal outcomes.
engineers to build a data science team. Successful
early-stage firms appreciate the fact that adding an Find and Score Quick Wins
innovation function is equally important, if not more
important, at this stage. With teams and processes in place, the ultimate test
for early-stage firms is whether the team can find and
The head of the innovation function is the heart score quick wins. If the team has the proper skill sets
and soul of this operation. (Those in this role and the processes are balanced and engaging, then
have a variety of titles in today’s industry, which positive results should follow.
are not important. It is who performs these functions
that matters.) Those who play this role likely In selecting these first projects, the team needs to
contribute to the development of the strategy, but more balance the impact on the investment process and
importantly, they are key to implementing the strategy. the technical feasibility and related time/resource
commitment. If a project takes a year or more to
To ensure the successful implementation of their develop and ends up having dubious results, it certainly
AI and big data strategy, early-stage firms need to does not bode well for the overall success of the AI and
identify those who have a sufficient understanding big data strategy. Ideal first projects are those that can

14 | CFA Institute
 T-Shaped Teams

deliver concrete and hopefully measurable results in UOBAM’s efforts at upgrading the digital technology
a reasonable period of time using technology that the available to its portfolio management teams.
team has confidence in developing.
The equities unit believes that with a rigorous and
These early wins are important to reassure senior disciplined investment research process, it can
management that the company is pursuing the right identify high-performing businesses that enable it to
strategy, convince the investment function that AI and deliver superior and consistent long-term performance.
big data can really help, and demonstrate to those in By combining this detailed research effort with a
the technology function how investment works and systematic model portfolio construction process,
how their investment colleagues operate. The early UOBAM believes it will be able to effectively derive
wins will likely help open the floodgate of ideas from sources of alpha to achieve outperformance over time.
both sides and start a virtuous cycle.
The fixed-income unit believes that consistent
The negative consequences of failure can also be performance can be achieved through rigorous and
far reaching, so management needs to have plans independent fundamental research to uncover relative
in place beforehand regarding how much time and value opportunities. By adopting diversified investment
resources it is ready to continue to dedicate to the strategies combined with active risk management, the
AI and big data efforts. team aims to generate sustainable total returns.
The multi-asset team makes recommendations on
strategic asset allocation based on the firm’s view on
CASE STUDY: UOBAM’S T-SHAPED TEAM the economic outlook, and it implements the balanced
AT WORK mandates of retail and institutional clients.

The case was prepared by Chua EnHao and Kimberly Roles and Job Scope of the Team
Yeoh of UOB Asset Management and edited by
Larry Cao, CFA.
Working on AI/ML and Big Data
The T-shaped project team brings together three
Background different business units and their expertise, as shown
in Figure 3. The technology (Tech) function consists
UOB Asset Management Ltd (UOBAM) is an Asian asset
manager offering investment products in fixed income,
equities, and multi-asset solutions that integrate FIGURE 3.  T-SHAPED TEAM AT UOBAM
traditional and alternative capabilities. The company
was established as a wholly owned subsidiary of TECHNOLOGY
United Overseas Bank in 1986 and was headquartered IT Partners, IT Personnel
in Singapore. It has since expanded and now has a
presence in many Asian countries. INNOVATION
UOBAM aims to provide top-class performance,
customer service, and experience by embracing digital Data &
Digitalization
technology to transform the customer experience,
operational processes, and business models.
Through its network of offices, UOBAM offers global
investment management expertise to institutions, INVESTMENT
corporations, and individuals through customized
portfolio management services and unit trusts. As of Portfolio
30 September 2020, UOBAM managed 60 unit trusts Managers,
in Singapore. The company had US$27 billion in total Analysts
assets under management (AUM) as of early May 2021.

Investment Philosophy
UOBAM believes in the use of a systematic portfolio
construction process. By controlling risk, it can better
deliver enhanced returns over the medium to long
term. This philosophy serves as the backdrop for

CFA Institute | 15
T-Shaped Teams

of IT partners, technology personnel, and technology The DD unit is also responsible for driving adoption of
tools. The innovation function is driven by the data and AI/ML at a project level (micro) and at the company/
digitalization (DD) unit, which is responsible for bringing regional level (macro). This is done through changing
the right data and technology to the right people, at the mindset and evangelizing and promoting the
the right time, on the right platform. The investment adoption of technology in the business units. The
function consists of the portfolio manager (PM) unit— DD unit does this by conducting workshops, talks,
including analysts from equity, fixed-income, and multi- and initiatives to increase awareness and by offering
asset units, as well as the newly formed environmental, courses to up-skill the workforce, with the aim of
social, and governance (ESG) unit. driving technological adoption. More of this information
is explained in the “Providing a Learning Environment in
The Investment Function UOBAM” section of this case study.

The UOBAM chief investment officer (CIO) leads the The DD unit also plays the role of governance, where
function in developing the firm’s long-term investment the unit ensures that the technology, the models
strategy and in managing asset allocation with the created, and all data involved are used in an ethical
objective to maximize the value of investments for manner via the data ethics validation procedure (DEVP).
its investors. He has oversight of the PM team, which The DEVP sets out the processes and procedures
includes the units managing equities, fixed income, to ensure that the development and use of artificial
and multi-asset investments and the newly formed intelligence and data analytics (AIDA) technological
ESG unit. solutions comply with the Fairness, Ethics,
Accountability and Transparency (FEAT) principles set
A key consideration that often comes to the CIO’s by the Monetary Authority of Singapore (MAS).
mind is how emerging technologies can be adopted
appropriately and be integrated within current Members of the DD unit have knowledge in both
investment management processes in a sustainable areas of technology and finance and have become
manner. This leads to the formation of the investment the channel that translates between hard code AI
technology unit, with the responsibility to ensure methodology within the technology domain and the
that adoption of the technology achieves the desired business (i.e., UOBAM investment) domain.
outcomes. For example, Ryan has extensive qualifications and
experience. He is a CFA charterholder, has experience
The Innovation Function in finance as an equity and futures market maker, and
holds a master’s degree in information technology.
Under the purview of the chief operating officer (COO) This unique background provides him with an in-depth
at UOB Asset Management, the DD unit serves as the understanding and insight into the best possible
innovation function in the T-shaped team setup. Its solutions available for business problems in the
main responsibilities include the following: financial realm. Another member is Kimberly Yeoh,
1. Serve as a bridge between the Tech function and who has a degree in economics and has prior working
other business units (BUs), and ensure that the experience using AI in a technology company.
needs of the business are fulfilled. The DD unit's knowledge and experience assisted it
2. Drive adoption of emerging technologies, such as in operationalizing business problems and matching
AI/ML, across all BUs. those problems with the best available technological
3. Ensure governance is in place and is adhering to solutions. Apart from the business knowledge they
policies on the usage of these technologies. bring to the project, the DD unit members also have
networking, critical thinking, communication, and
The DD unit, led by Chua EnHao, is the link that collaboration skills, among others. When assessing
bridges the spheres of business and technology in the right solution for the problems, the DD unit has to
an AI/ML project. The work of the DD unit begins by play the role of futurist in ensuring that the solutions
identifying challenges faced by other business units proposed are forward looking and are future proof.
and assessing whether there is an opportunity for
the challenge to be solved by the use of AI/ML. The The DD unit's ability to be an effective bridge between
DD team identifies challenges by sitting in meetings, the business and technological domain, be a driver
through interdepartmental collaborations, and by other of technological change, and be a governor of proper
BUs approaching the DD team with their challenges. and ethical use of technology and data is what gives
UOBAM a distinct competitive advantage.

16 | CFA Institute
 T-Shaped Teams

The Technology Function as various business units may submit overlapping

problem statements.
To accelerate adoption of emerging technologies, such
as artificial intelligence and machine learning, UOBAM Subsequently, when the project receives approval
builds its capabilities on the following fronts: from the working group, the DD unit launches the
project by conducting meetings with the relevant
1. Engaging fintech companies with proven solutions stakeholders to capture detailed business objectives
immediately catalyzed and transformed how and requirements.
UOBAM operates, creating better products and
delighting its customers To encourage and elicit ideas to use AI/ML solutions
from staff, UOBAM’s DD unit also conducts regular
2. Partnering closely with UOB Group Technology,
workshops that involve the heads and staff of
UOBAM has been able to craft technology
various business units. These workshops serve as
roadmaps, execute in an agile manner, and see
a platform for employees to share the challenges
results of quick wins and MVPs (minimum viable
faced by their business units and to brainstorm
products).
possible solutions. Different business units often
3. Consulting services from the UOB Data discover that they face similar challenges, which
Management Office speed up learning curves and can then be fine-tuned and consolidated to find
help shape how UOBAM will need to establish its one comprehensive solution.
own capability.

Workflow in AI/ML Adoption Phase 2: Model Development

and Implementation and Testing
The approved projects then progress to Phase 2 for
UOBAM has put in place an AI/ML development life
model development and testing. The first step is to
cycle, which defines the framework to develop AI/
ensure all requirements—including business objectives
ML and advanced analytics solutions to enhance
and functional and nonfunctional requirements—are
business. The four major phases of the life cycle,
properly captured. Resources, such as required data
shown in Figure 4, describe the major activities and
features, are also identified by the stakeholders and
interactions between stakeholders (including business
data scientists.
users) to develop practical AI/ML solutions and bring
them to production. Subsequently, the data scientists begin the
development of the AI/ML solution in an agile
Phase 1: Ideation manner, delivering incremental features of the
solution to the stakeholders for evaluation at
At the ideation phase, employees from business the end of each sprint.
units identify processes within the business that can
potentially be enhanced through the adoption of AI/ After the prototype solution is completed, the
ML solutions. Once these challenges are crystallized stakeholders of the project come together to
into problem statements, these are brought forward determine whether the prototype solution can be
to an innovation working group, comprising DD piloted for wider testing. The pilot testing will enable
staff and business representatives who have an stakeholders to validate whether the completed
in-depth understanding of both the business and solution can meet the business needs or if it requires
the applications of technology. The group evaluates additional development.
and prioritizes various problem statements based Finally, when the proposed solution has been
on the impact to the business and the feasibility of sufficiently tested and is able to meet all the business
possible solutions. The group also serves to maximize needs, the stakeholders can elect to move the solution
intra-project collaboration and minimize duplication to Phase 3 to operationalize it.

FIGURE 4.  FOUR MAJOR PHASES OF AI/ML LIFE CYCLE

Phase 2
Phase 4
Phase 1 Model Phase 3
Maintenance &
Ideation Development & Operationalization
Monitoring
Testing

CFA Institute | 17
T-Shaped Teams

Phase 3: Operationalization As UOBAM continued to expand its regional footprint

within the ASEAN region, the portfolio of the team
Before the solution is operationalized or deployed expanded faster than the size of the team. Analysts
in a production environment, it has to be checked were also required to evaluate a large number of
and validated by an independent internal team. Also, factors, and the team would have to rely on the
the model must adhere to the MAS FEAT principles. experience and expertise of its portfolio managers. In
For example, the individuals or groups of individuals order for UOBAM to maintain its standing as a regional
should not be systematically disadvantaged by asset manager with top-class performance, it would
the solution decisions unless the decisions can need to ensure that its investment team worked
be justified. effectively to cover a larger portfolio and make more
In the last step before operationalization, stakeholders data-driven investment decisions.
must present the solution before a management-led With the advancement of AI/ML, the investment team
technology committee for approval. This ensures that decided to collaborate with the DD unit to harness
the solution is aligned with the goals and philosophy the power of AI/ML to cover approximately 120 factors
of the business. and a larger portfolio of equities. After incorporating
After final approval is obtained from the management- UOBAM’s investment philosophy, the project team
led technology committee, the Tech function will developed a model that aimed to predict equity
proceed to operationalize the solution and deploy it performance and portfolio allocation.
into the production environment for the business.
The Process
Phase 4: Maintenance and Monitoring The project followed UOBAM’s best practice model,
After the project is operationalized, the solution enters the AI/ML life cycle shown in Figure 4, which begins
a maintenance and monitoring phase. The solution will with ideation (Phase 1) by the PM team through the
be reviewed periodically by its users and stakeholders digital pipeline. The PM team was searching for a
to ensure it remains relevant to the business. more efficient way of analyzing the vast amount of
equity stocks in the APAC region (as mentioned in the
The model also undergoes regular testing to ensure “Project Background” section of this case study). The
it is reliable. For example, if the solution includes a PM team consulted with the DD unit to find the best
forecasting model, the model drift (the difference technological solution that could assist the PM team in
between the model forecast and the actual data) will its business problem, and both teams concluded that
be measured and reported. applying an AI/ML solution would be the most effective
In addition to tracking the solution performance, approach.
business users will also be required to incorporate The DD unit then assessed the viability of using AI/ML
such metrics as the cost-to-benefit ratio, productivity, to solve the PM team’s problem (Phase 2). After the
and return on investment of the solution, where teams assessed that there was a high probability
applicable. Solutions that do not meet business needs of success in applying AI/ML to the problem, the
are either decommissioned or improved through a new functions conducted a more formal discussion to fine-
AI/ML life cycle. tune the scope of the project. Subsequently, it was
determined that the goal of this AI/ML project would be
A Case within the Case: The Multi- to harness the capabilities of AI/ML to create a portfolio
of Asian (ex Japan, ex China) stocks that would be able
Signal Predictive Analytics Project to exceed the benchmark.
In order to better understand how the T-shaped team
The PM unit set the business objective, while the
has benefitted UOBAM, a case study of the T-shaped
DD team translated the objective into an actionable
team at work is provided next.
development workflow for the Tech team to work
on. The project was presented to the technology
Project Background committee, and the green light was given for
In 2019, the equities team, consisting of several development of a pilot. During the model development
portfolio managers and analysts, managed over stage (Phase 2), the team faced several challenges
120 funds, which required analysts to cover over (e.g., the model generated portfolios that exhibited
3,500 equities within the Asia-Pacific (APAC) region. high turnover rates, which would not be practicable
for the business). This was where the DD unit was able

18 | CFA Institute
 T-Shaped Teams

to step in to advise the Tech team on the tweaks that The UOBAM DD unit (i.e., innovation function) serves as
should be made to ensure the model was aligned with the bridge between the domains of technology (Tech
UOBAM’s business goals and investment philosophy. team) and investment (PM team), playing the role of
The pilot went through several iterations before it digital translator to ensure that business needs are
could go into production to be operationalized met in the model development. From a governance
(Phase 3) in the business. perspective, the DD team will ensure that the data
are used responsibly and will work with Value3 to
This project served as an eye opener for the PM team
embed ESG considerations on the platform to facilitate
as the model was able to uncover hidden gems that
responsible investing efforts. Figure 5 depicts this
would usually fly under the radar. Subsequently, the
model for a fixed-income ratings project with Value3
PM team understood the added value of introducing
and UOBAM.
AI/ML into its business model and took the initiative to
up-skill in said areas. Quantitative success measures In “AI Pioneers in Investment Management” (CFA
include increasing alpha and efficiency of the analysts Institute 2019a), Larry Cao identified five major hurdles
working on the resultant portfolio. The knowledge investment firms face in applying AI/ML and big data
and insights the PM team gained allowed it to better solutions: cost, talent, technology, vision, and time.
assess the merits of technology partners that UOBAM UOBAM’s collaboration with Value3 helps counter these
may engage. challenges.
At present, a majority of the investment team The cost of developing an integrated system—such
have taken courses on artificial intelligence for as the one Value3 has—is generally regarded as a big-
investment professionals and have learned Python ticket item. UOBAM was fortunate to be able to tap a
through DataCamp. Some have also gone on to government grant provided by MAS in the adoption
attain professional qualifications (e.g., a professional of the system. Talent is spread across the T-shaped
certificate in Python programming and machine team, with each domain providing input in its area
learning from SMU Academy). of expertise. Technology is ever evolving, with new
players entering the market providing niche solutions,
Engagement of Fintech Companies and it is UOBAM’s prerogative to be aware of these

The asset management industry in APAC will continue

to grow rapidly and will provide asset managers with FIGURE 5. T-SHAPED TEAM FOR THE FIXED-INCOME
vast opportunities. To maximize these opportunities, RATINGS PROJECT
asset managers will have to leverage technology, at
speed and scale, to give themselves a competitive TECHNOLOGY
advantage. Value3
Technology solution
UOBAM has recently partnered with Value3 Advisory, a
fintech company that offers a capital market AI platform INNOVATION
for independent, predictive, and automated credit
ratings, research, and analytics (UOBAM 2019). Value3’s Data &
platform, which uses a combination of AI, ML, and NLP Digitalization
(natural language processing), empowers investment Team
and risk managers in financial institutions to make Digital translator
better decisions in portfolio selection, risk monitoring, and governance
benchmarking, and early warning indicators. The
platform combines financial data with unstructured INVESTMENT
online digital footprints, news, events, trends, and
patterns of the companies from diverse sources to
PM/Fixed-Income
transform data overload into actionable insights. and Risk
Analysts
This strategic partnership serves as a win–win for
both parties. In the development of the model, Value3
Market expertise
creates the prototype with its technological expertise
and UOBAM refines the model with market intelligence
gathered from its years of experience in asset
management and bond valuation in the ASEAN region.

CFA Institute | 19
T-Shaped Teams

niche solutions and evaluate whether they could users (power users) to leverage on data from the data
provide value to the operationalization of the firm. lake for analysis and generating insights.
UOBAM’s evaluation requires forward-thinking vision
to ensure that these niche solutions, if adopted, are Strengthen Skill Set
viable to the firm’s business needs, adaptable to
future advances in technology, and, more important, UOBAM realizes that to have a data-driven culture,
in line with the mission, goals, and values of the employees will need data-related skills and knowledge.
firm. In this project, UOBAM has found that its time is UOBAM has several training initiatives to ensure its
predominantly spent on the refining of the model to employees have avenues to gain knowledge, which
suit business needs rather than in the development are illustrated next, in the fourth element of digital
phase, which can be time consuming. transformation.

From a technological perspective, in UOBAM there has

been a gradual shift from doing everything in-house Shift Mindset
to working with partners that are able to enhance The shifting of UOBAM’s employees’ mindsets is a
and complement UOBAM’s core business, which will critical component in the drive to be a digital-first
provide the firm with a competitive edge in the future. company. The firm has initiated several programs to
help its employees make that transition, including
Providing a Learning Environment a series of training initiatives and resources for
employees to tap into.
in UOBAM
Training for the acquisition of various data-related
UOBAM places a strong emphasis on employee
skills can be found in UOBAM’s in-house learning
learning and up-skilling of staff so that it can better
portal, iAMdigital. The portal has various classes on
meet the rigours and demands of an ever-evolving
programming languages, such as R, Python, and SQL,
technological work landscape. UOBAM’s digital strategy
as well as tutorials for AI and ML. Employees also are
is based on the two pillars of people and technology.
equipped with access to a DataCamp and LinkedIn
These two pillars form a symbiotic relationship in the
Learning account, which offer courses at various
firm’s quest to be a digital-first enterprise.
proficiency levels so employees can improve their data
This relationship is further illustrated in the four and digital skills.
elements of digital transformation UOBAM has in place:
These training programs not only have facilitated the
solidify dataset, sharpen tool set, strengthen skill set,
strengthening of UOBAM employees’ skill sets but
and shift mindset.
also more importantly have shifted the mindset of its
staff in confidently adopting technology in their daily
Solidify Dataset working lives.
Data are the fuel that provides a desired result.
The quality and reliability of data will determine the Key Takeaways
accuracy of results in any model. Solidifying the
Initiating change is never an easy process, and
dataset will include areas of data governance, strategy
that process is exponentially harder with a large
alignment, and data privacy and security.
organization.

Sharpen Tool Set Leadership style within UOBAM has also evolved
into a transformational style of leadership. There is a
Technology tools greatly assist any AI/ML project, and strong mandate from upper management to integrate
the better the tool, the better the results. UOBAM has technology in the work processes.
made improvements to its enterprise data architecture
and governance (EDAG), which include a data lake Management has also realized that while work
and enterprise data warehouse (DLEDW) and a data processes can be automated, human intelligence is
discovery platform (DDT). The DDT was developed as still needed to ask questions, identify problems, think
a “sandbox” that allows businesses to perform data of possible answers, and create solutions. Training is
analysis and develop analytical prototypes. These are constantly being developed to ensure that staff are
cyber-secured environments that enable pre-endorsed up to date with technological knowledge. UOBAM’s

20 | CFA Institute
 T-Shaped Teams

CEO, Thio Boon Kiat, has been a champion of the of job losses, companies winding down, and many
organization adopting and utilizing technology to bring more facing financial crises. The normal we once knew
better value to customers. was obliterated in a short period of time.
Technology is becoming more ubiquitous and will UOBAM recognizes the need to embrace technology
permeate almost every industry in the near future. This and use it to assist employees in being more efficient
coming decade will see AI and ML play a significant in their job scope and better decision makers.
and fundamental role in finance, particularly in areas of The recent successes of the projects involving AI
asset allocation, risk management, fraud prevention, and ML serve as a testament to this notion. Being
and process automation, among others. a company that is entrenched in technology will
allow UOBAM to be agile and act swiftly to counter
The COVID-19 pandemic has illustrated that no industry
uncertainty in whatever form it may come.
is safe from uncertainty. The pandemic saw multitudes

CFA Institute | 21
CHAPTER 3. THE INTERMEDIATE STAGE ON THE AI
AND BIG DATA ADOPTION JOURNEY

There are clear distinctions between firms in the be the highest level they reach—more on that in the
early and intermediate stages on the AI and big data next chapter). It is in this period that the T-shaped
adoption journey. teams truly shine. They grow together with the
organization’s efforts and, if they are successful,
the organization’s performance.
WHAT DEFINES INTERMEDIATE-STAGE We find that successful intermediate-stage firms tend
FINANCIAL INSTITUTIONS ON THEIR AI to focus on the following three aspects.

AND BIG DATA JOURNEY?

Building Out the Team
If senior management plays the most important part Intermediate-stage firms tend to have a more
in propelling a firm into the early stage, then both elaborate T-shaped team structure compared with
senior management and the innovation function leader early-stage firms. As the number of projects and areas
are important for the successful transition to the that adopt AI and big data increases, the demands on
intermediate stage. the T-shaped team increase as well.
For a firm entering the The more elaborate team is usually
intermediate stage, the innovation reflected in two aspects: roles and
function head, with the support of Both senior management caliber of staff. More people will
senior management, has usually and the innovation function join the team, and with increasing
assembled a group of capable scale, team members will play a
team members, developed leader are important for the
more defined role. This growth
some processes that formalize successful transition to the
process is very similar to what
the selection of projects with intermediate stage. start-ups go through following
input from both investment and the successful launch of their
technology functions and, most products.
importantly, delivered a number of successful projects
that have improved some of the firm’s investment and/
or business processes. Honing the Process
In the intermediate stage, firms will continue to build Teams at intermediate-stage firms generally have
on their organization structure and processes so that developed more elaborate and better-defined
they can continue to drive AI and big data adoption processes in selecting projects. There is far greater
across their organizations. demand on the efficacy and efficiency of processes
at this stage, as AI and big data applications will start
to expand to more parts of the investment process
and more teams across the organization. They will also
WHAT SHOULD INTERMEDIATE-STAGE start to expand into the middle-office and back-office
FIRMS FOCUS ON? operations. Hence, the demand on the process also
multiplies.
We do not get asked this question a lot, because the
majority of firms in the industry have not gotten to Juggling Multiple Projects
this stage yet. Still, we feel it is important to describe
what the future (i.e., intermediate stage and beyond) The most obvious sign that a firm is in its intermediate
may look like so firms can develop their own strategy stage of AI and big data adoption is that it has
accordingly. multiple projects across the organization. In this
stage, there is increased organizational buy-in for
Most firms will spend an extended period in the AI and big data technologies. More departments will
intermediate stage (and for many of them, this could open themselves up and welcome the opportunity to

22 | CFA Institute © 2021 CFA Institute. All rights reserved.

 T-Shaped Teams

collaborate with the new talents from the innovation CASE STUDY: NN INVESTMENT
and technology functions. There will also be more
leeway to take on higher-impact but also higher- PARTNERS AND THE T-SHAPED
commitment projects. TEAM AT WORK
The case was prepared by X. (Tracy) Zhang
TO CENTRALIZE OR NOT TO CENTRALIZE? and P. (Prasenjeet) Bhattacharya of NN Investment
Partners and edited by Larry Cao, CFA.
Should data scientists be centralized or join individual
portfolio management teams? This question similarly
applies to the innovation function. So, should financial Company Background
institutions build one T-shaped team at the firm level, NN Investment Partners (also known as NN IP) is
or should they build multiple T-shaped teams, each a Netherlands-based asset manager operating in
at the level of the smallest investment unit? A related 15 countries in Europe, Asia, and the Americas. A global
question is whether financial institutions should build active asset manager for institutional and wholesale
one T-shaped team at the headquarter level or at each investors, it manages €300 billion in assets for pension
division/regional subsidiary. funds, insurers, family offices, independent financial
advisers, banks, and private individuals.
There are two ways to look at this question of
centralization. The simple way is to look at the number NN IP offers a diverse range of funds within fixed-
of projects the firm is taking on at any one time. In the income, equity, multi-asset, and alternative investment
early stages, when the number is low, it is difficult to strategies. Fixed-income strategies account for almost
justify having a T-shaped team for every investment two-thirds of the firm’s assets under management.
unit. It is also important to centralize the knowledge
building at that point. Investment Philosophy
As the operation matures and the number of projects NN IP believes that markets are complex and not
increases, it will start to make sense to add a limited fully rational. The market ecology has shown to be
number of individual technologists in investment more complex in nature, and while the efficient
units, as needed. This will most likely take place in market hypothesis (EMH) remains vital, there are
the intermediate stage. The decision rule that we pockets of opportunities for active asset managers
recommend using is to see whether the individual or to exploit inefficiencies driven by the complex and
the function in question has more interactions with evolving environment. NN IP has adopted tenets of the
others in the central team on a daily basis or with the adaptive market hypothesis (AMH) as a foundation
individual team. of its investment philosophy. The AMH not only offers
The situation may change from time to time, and an extension of the EMH that integrates the benefits
the staffing arrangement can change, too. In and tools of modern finance theory, but it also adds
general, as operations continue to mature, we do insights that allow one to look at the market as an
expect to see multiple T-shaped teams operating evolving entity within a complex and adaptive system.
at different divisions/subsidiaries/regions of NN IP complements and enriches its investment
a large financial institution, particularly in the philosophy with the application of artificial intelligence,
advanced stage. such as machine learning and NLP, as well as
traditional and alternative structured and unstructured
In addition to the T-shaped teams in different datasets. This helps the firm capture the fast and slow
subsidiaries/departments that operate in parallel, regime shifts, new drivers of factor premiums, and
a more complicated structure is multiple T-shaped timely insights on economic activities and nonfinancial
teams at different levels of the corporate structure factors, such as ESG factors.
with their own hierarchical relationships.21 There is a
central team at the corporate level that supports the
main business lines with a core technology function The Organizing Framework
and innovation function. And the “mothership” also The information and technological age has begun to
arranges satellite teams to support some of the dominate parts of NN IP’s communication. This age,
business units, as needed. along with improved data storage, computing power,

One of our contributors referred to this as the Russian doll of T-shaped teams.
21

CFA Institute | 23
T-Shaped Teams

and the rise of open-source licensing, is causing behavioral science. Moreover, experimenting with
exponential advancement in the area of big data applications of emerging technology and working with
and AI technologies. new behavioral science hypotheses within investing
are also crucial aspects to build organizational
Using data in new ways is essentially about
knowledge.
augmenting the firm’s experience with data. It is
achieved by using advanced algorithms to unearth To keep up with the pace of advancement and leverage
new patterns and nonlinearities and by implementing upon it, an effective change management process is
behavioral insights when working with information. also crucial for organizations to experiment with and
In order to identify and democratize access to the adopt AI and big data in their investment decision-
sources of alpha derived from ESG factors, big data, making processes.
AI, and behavioral science, NN IP brings people with
NN IP describes its way of working as inspired by
domain knowledge and technical expertise together to
the idea of working on the edge of chaos. That is, by
work around three core objectives.
combining the right level of chaos with structured
The first core objective is related to data. Getting checkpoints and methods, NN IP aims to control its
access to datasets, including new and alternative data innovation projects in the most efficient manner.
sources that have potential to bring fresh insights to
As part of NN IP’s strategic direction, it reoriented the
the investment function, is the crucial first step. This
structure of its investment organization by placing
also means making sure that internally generated data
ESG specialists and technologists (data scientists
are more readily available and are in an accessible
and engineers) in each investment team, as opposed
format. Accessibility and usability of data allow for
to an earlier structure of a centralized team. This
faster cross-fertilization of ideas and insights.
decentralization helped create both technological
The second objective relates to the technological focus and ESG expertise within each investment
environment. Advancing technologies, such as AI, team. It also accelerated the application of new data,
will be disruptive to the asset management industry. technology, and research in investment processes.
To that end, the innovation platform and investment Another strategic step was to establish a centralized
team work together to build new data engineering innovation and responsible investing (RI) platform. This
and computing capabilities and new analytical way of structuring helped the firm build a T-shaped
competencies. A scalable tech environment that organization in its true sense. At the organization
can handle structured and unstructured data is very level, the innovation and RI platform is the linking
important for AI/big data–related work. pin between the technology and operations (Tech
& Ops) department and the investment department
The third objective is related to research and
(see Figure 6).
experimentation. To make sure NN IP is continuously
learning and challenging existing processes, models,
and techniques, it has taken a more strategic The Innovation Team
perspective on all the research efforts within the NN IP’s innovation and RI platform sits in the middle
company. This allows the firm to focus and unlock of a beehive structure, which enables the investment
knowledge in the areas of data, technology, and teams to embrace responsible investing and new and

FIGURE 6. NN IP INVESTMENT TEAM STRUCTURE AND EMBODIMENT OF THE T-SHAPED STRUCTURE

TECHNOLOGY
Chief
Investment
Officer
Specialized Specialized Innovation
Fixed Equity & RI
Tech- Income Platform
Innovation
Embedded
Platform
Partner
Fixed-
Income Multi-Asset
Solutions
Adaptive
Investing INVESTMENT

24 | CFA Institute
 T-Shaped Teams

emerging technology, understand the latest big data (e.g., data engineers, data scientists, software
and alternative data trends, and think critically about engineers) and investments (e.g., portfolio managers,
their own investment decisions. The innovation team analysts). The solution team works on various data,
was created to bring people and technology together data science, and tooling-related projects.
by inspiring, facilitating, and creating smart investment
Upon setting up the innovation team, it has been
solutions for its clients and colleagues.
agreed with the CIO and the COO (chief operations
To provide and maintain an appropriate level officer) that the innovation team should consist of
of support across all investment functional colleagues from both technology and operations,
areas, the innovation function is divided in two as well as investments, to bring combined domain
layers—the management team and the solution knowledge to the table to develop the best solutions.
team—as shown in Figure 7. In the management team, the head of innovation
and innovation area leads come from the investment
The management team consists of senior managers
department, while the tech lead and program leads
from both the investment department and the Tech
come from the Tech & Ops department. Similarly, in the
& Ops department, program managers, and subject
solution team, innovation drivers and tech members
matter experts of various investment strategies. This
come from Tech & Ops, while senior users are portfolio
group of people operates across the functional area
managers and analysts of different investment
within the organization and serves as a center of
teams, who are responsible for bringing the business
excellence with a focus on the following:
perspective to the projects.
1. Providing thought leadership, vision, and direction
The innovation team in itself is a T-shaped team that
2. Establishing and maintaining the NN IP innovation combines different skills, including domain, innovation,
ecosystem and technology expertise. Table 2 highlights how
3. Establishing best practices to adopt AI and big the team members of the innovation and RI platform
data represent the skills in accordance with the T-shaped
4. Overseeing all innovation programs/projects skills spectrum. The innovation team brings in deep
executed and assisted by the innovation platform knowledge about the investment domain, data
science, and technology infrastructure to build
5. Supporting colleagues to gain either technical
scalable solutions. Combined with NN IP’s way of
skills or investment skills through education and
working based on design thinking, lean start-up, and
training
agile methods, this provides just enough structure
The solution team consists of innovation drivers to be focused on business value while allowing for
and subject matter experts from both technology creativity and adaptiveness. This combination of

FIGURE 7. INNOVATION FUNCTION

Reporting to Tech & Ops Reporting to Investments

The Teams
The Management Team The Solution Team

Head of Innovation Innovation Driver/

Product Owner &
Scrum Master
Technical Lead
NN Investment
Partners
Technical Members
Program Leads

Innovation Area Leads Senior User

Business Sponsor

Source: NN Investment Partners.

CFA Institute | 25
T-Shaped Teams

TABLE 2.  THE T-SHAPED INNOVATION AND RI PLATFORM

Technical Skills Innovation Skills Investment Skills
Tech & Ops Department • Software Engineers
• Data Engineers &
Scientists
• Scrum Masters
Innovation Platform • Technical Lead • Program Leads • Head of Innovation
• Technical Members • Innovation Drivers • Innovation Area Leads
Investment Team • Quantitative • Head of Investment
Researchers Team
• Portfolio Managers
• Analysts & Strategists
• Economists

skills helps connect the investment and technology leads, program leads, and innovation drivers play a key
functions of the company efficiently. The T-shaped role. An investment process has many components
innovation and RI platform focuses on inspiring new that are needed for an investment strategy to be
ideas, facilitating cross-team projects, and creating successful. However, if one looks at a meta level, it
innovative solutions to enable NN IP to adopt AI and takes only two inputs: information and experience
big data in its investment processes at scale. with the information. The information could be about
traditional indicators and financial factors or even big
Roles and Responsibilities data factors. Experience refers to the understanding of
the information and its application in making informed
The head of innovation and RI is part of the investment decisions. Here, AI and advanced analytics can help in
management team and is responsible for the overall augmenting human experience when engaging with
innovation portfolio and responsible investing. The traditional or new datasets. AI and machine learning
technical lead brings a wealth of technical knowledge techniques can identify the nonlinear relationships
and expertise in managing the technical team. The between the datasets and provide additional insights
technical lead has the technical knowledge of the on the characteristics of the datasets. In order to
solution and is able to identify potential technological integrate big data and advanced technology in
gaps and risks in the solution. Along with the investment processes, program leads and innovation
innovation area leads, the technical lead guides the area leads, such as the strategic R&D (research
product backlog prioritization and actively participates and development) lead and the investment science
in the definition of the new innovation product strategy lead, bring their cross-functional skill in investment,
with the product owner. technology, and innovation methods to identify
In order to enable efficient collaboration and creation the right approach to achieve investment process
of practical yet innovative solutions, the innovation innovation with AI and big data.
team works closely with an internal business sponsor. The strategic R&D lead owns the research agenda of the
The business sponsor reports to the CIO and is a senior investment engine, thereby having a complete overview
investment professional and head of an investment of all the research activities. This helps in identifying
team. The business sponsor brings additional thought areas of synergy within all the hives where big data
leadership and business perspective to the innovation and AI are being investigated. Since the collaboration
function by covering the innovation project cost, between investment and data science functions is
bringing in insight specific to the investment process, challenging, the R&D lead also leads the community
and often providing resources from his or her team of data scientists embedded in the hives to facilitate
to jointly work on the project. Priorities within the knowledge exchange in the areas of AI, machine
innovation project are set by the executive sponsor. learning, and big data and their applications in investing.
To bring process innovation and practice leadership The R&D lead’s responsibility is not only coordinating the
to the investment processes, the innovation area research agenda but also communicating the progress

26 | CFA Institute
 T-Shaped Teams

back to the C-suite and the management team of the on the emerging industry trends and translate these
investment function. More importantly, he or she leads into action points that enable adoption of AI, machine
the discussion around which competencies within AI learning, and big data in the investment processes.
adoption the company should focus on and prioritize. He
Program leads are T-shaped professionals who have
or she is also responsible for getting the management
sufficient knowledge and understanding of both
team, data scientists, and data engineers to contribute
technology and investment. They are also experts in
to the discussion in a productive way. Thereafter, the
innovation methods and project management and
R&D lead incorporates a cross-functional AI competency
work closely with the innovation area leads to ensure
and is responsible for its overall success. In short, the
the function operates in an efficient and structured
R&D lead sets the strategic direction of the research
manner. They are responsible for managing and
activities around AI, machine learning, and big data
executing the specific elements within the innovation
adoption in investment processes, acts as a translator,
process. The program leads and the innovation
and leads new AI competency development within the
area leads also actively engage in stakeholder
company.
management.
The investment science lead manages the technical
members, such as the data scientists and data The Innovation Process
engineers, of the innovation team. Jointly they
focus on using AI, machine learning, and big data to All big data and AI projects at NN IP follow the
develop novel investment solutions that investment innovation process, which typically includes the
professionals in the rest of the organization can following steps:
use to improve their investment processes. These • Ideate
improvements typically come in the form of efficiency
• Explore
gains or enhanced alpha potential. Being part of the
innovation and RI platform, special attention is given • Experiment
to responsible investing-related opportunities, which • Execute
account for more than half of the investment science • Business-as-usual (BAU)
portfolio of innovation initiatives. Furthermore, he or
she plays a crucial role in further scaling up the AI Figure 8 depicts this process.
competencies prioritized as the strategic direction.
He or she is also responsible for actively gathering Ideate
insights from internal experiments and sharing these
broadly with investment professionals, thereby The amount of data and tooling in the industry is huge,
inspiring them to adopt emerging technologies in growing, and mostly too immature to directly embed in
their investment processes. More importantly, he investment processes. At the same time, many internal
or she also plays an important role as innovation ideas and initiatives are somewhat fragmented and
ambassador, working together with other senior not yet aligned with the strategy. Ideation happens
investment professionals in the sourcing of new mostly within the investment hives themselves,
ideas and distribution of the insights. The actionable where portfolio managers and analysts, together with
and innovative solutions developed by the technical technologists in the teams, sketch an investment idea
members of the team under the direction of the or define an investment challenge. Such an idea or
investment science lead enhance the investment challenge will be brought to and centralized with the
processes of the teams by providing fresh insights on program leads of the innovation platform.
efficiency gains and alpha opportunities.
Both the innovation area leads are subject matter
Explore
experts and thus function as knowledge engineers The idea/challenge will be explored, a team formed, the
who bring in their investment expertise and insights vision and mission of the team defined, and the scope

FIGURE 8.  THE NN IP INNOVATION PROCESS

BUSINESS INNOVATION

Ideate Explore Experiment Execute BAU

CFA Institute | 27
T-Shaped Teams

set. These decisions are made within the management The Stage-Gate Process
team of the innovation platform. Further, this phase will
result in a set of assumptions about the investment To keep a fast pace and a short time to market, NN
idea/challenge faced by investment professionals IP works with a stage-gate process. A stage-gate is
and the opportunities for the innovation platform a meeting with the team and decision makers and is
to address this challenge. It is then determined used to “unlock” the gate to the next stage. The stage-
whether resources and expertise can be allocated to gate is used for three purposes:
experiment with the idea. 1. Fail fast, following the agile innovation approach.
This way, the team keeps up its fast pace.
Experiment 2. Create transparent decision making by clear
The goal is to find a solution that fits the problem. agreement of “definition of done” with the
When developing new ideas, the solution team innovation platform and users.
tries to find, via experiments, whether the solutions 3. Prioritize through scarcity; this will breed creativity
are feasible, desirable, and viable. Continuously and better results. This means that along the way,
checking with senior users whether the solution can the firm explores/experiments with many ideas
actually solve their problem is key. Using the build– while only the strongest will follow through to the
measure–learn approach and allowing to tweak, phase in/out stages.
pivot, and kill solutions, the team can easily adapt During a stage-gate, an innovation team presents
the solution as necessary. its innovative project to the leadership team and the
In this phase, the solution team starts with a business sponsor. After the presentation, the panel
proposition in the form of a prototype and will seek is able to ask questions about the project and decide
to further develop it into an MVP (minimum viable whether the project receives a “go” to the next stage
product). or a “no go.” Stage-gates are organized with a certain
frequency.
Execute (Pilot, Implementation)
Impact of Innovation Process:
The pilot phase is a test phase and delivers a lot of
lessons that are essential to implement innovation Enhancing Existing Processes
in practice. The solution team is running the MVP in The team has identified opportunities across the
a pilot environment to collect feedback from the organization where AI and big data technologies can
investment professionals. Before starting the pilot, be used to improve current investment processes or
the team creates success criteria, which need to be develop new methods. Below are a couple of examples
met in order to develop the product/algorithm further where the objective was primarily to enhance existing
after the pilot. After the pilot, the team can further processes.
develop the project based on the feedback of the
users. During the implementation phase, the product/
algorithm is integrated into the business operations of Scout
NN IP. By making products fit the requirements of the The idea originated out of a separate study of
ongoing business, the implementation is streamlined. anomalies. Whenever an algorithm points to anomalies,
During this phase, the innovation platform is still in the it is not clear what exactly the firm is dealing with. It
lead of the project. can be a data problem, something naturally fat-tailed,
or it can be an interesting investment case. To figure
Business-as-Usual this out, the firm needs input from an expert. On top
of the algorithm on anomaly detection, the innovation
After the solution has been implemented, the project platform developed a front-end process that enables
can no longer be seen as an innovation platform analysts to inspect the latest batch of (securities)
project. Depending on the nature of the project, it is candidates and supply their opinion. To decrease
handed over to either the investment or technology overload, security characteristics that may be driving
function. the candidacy are highlighted. In addition, analysts

28 | CFA Institute
 T-Shaped Teams

do not need to go through thousands of securities. reach their decisions. This was mainly due to the
Selection logic by the analysts is then imprinted in the difficulty in processing unstructured data to derive
model to improve the candidate generation process. meaningful insights. With the recent acceleration
in new technology and processing capacity, most
ImpuNet organizations are embracing advanced technologies
to engage with 80% of this bulk. And those who fail to
In the world of financial data science, a good data invest in such tools and technologies will face the 80%
source without missing values is an exception rather blind spot while making crucial decisions.
than a rule. This is especially the case within alternative
data and unstructured data. The NN IP team proposed NN IP is building its competency in NLP technology to
a new algorithm for estimating missing values, which manage and analyze such datasets in efficient and
is based on the idea of denoising autoencoders. scalable ways. Such competency will transform NN
Autoencoders are a family of unsupervised techniques IP’s capabilities to translate the 80% blind spot to 80%
in machine learning based on neural networks. In fresh insight. The NLP competency is being executed
general, an autoencoder aims to learn the underlying by a cross-functional team involving technologists,
structure of the dataset (e.g., relationship between subject matter experts, and five different investment
features) without assuming simple linearity. Depending teams.
on the exact architecture, an autoencoder can, for
instance, convert a big set of (potentially correlated) Super Forecaster App
characteristics into a smaller representation that still
Human predictions are inherently noisy.22 Some
retains the bulk of information about the observation
people overestimate while others underestimate the
(think, for instance, of a data compression task).
probability that some future event will happen. When
Another popular flavor of autoencoder, called a
aggregating a sufficiently large number of individual
denoising autoencoder, can reconstruct the missing
forecasts, these errors tend to cancel out. This
data with imputed values. The approach can benefit
phenomenon is called “the wisdom of the crowd.”
from accelerated computing and therefore scale better
to large datasets with a lot of samples and/or features. By tracking individual forecasts, it is possible to
Second, by applying a generative model to produce identify the “superforecasters” in the group. By giving
realistic missing value patterns during training, we can a greater weight to their forecasts, collective forecasts
relax the assumption of randomness in data gaps and can be made even more accurate. Moreover, through
increase out-of-sample accuracy. training and feedback, individuals can improve their
forecasting skills over time.
Impact of Innovation Process: Applying these insights, NN IP has developed a
Exploring Emerging Technology novel “Super Forecasting” app. Using this app, the
investment professionals at NN IP will be making
and Experimenting regular forecasts about various market-relevant
Below are a couple of examples where the objective topics. Furthermore, users will receive feedback about
was primarily to develop capabilities that previously their performance and learn how to become better
did not exist at NN IP. forecasters over time. The forecasts that are harvested
with this app will be used as inputs in the investment
Building NLP Competency to Unlock process of various NN IP teams.

Value from Unstructured Text

Until recently, most decision makers primarily
relied on structured and organized datasets to

See, for example, Mellers, Stone, Murray, Minster, Rohrbaugh, Bishop, Chen, Baker, Hou, Horowitz, Ungar, and Tetlock (2015).
22

CFA Institute | 29
CHAPTER 4. THE ADVANCED STAGE ON THE AI
AND BIG DATA ADOPTION JOURNEY

Understandably, the advanced stage of the journey is significant number of projects right demands a level
where we know the least, so we will be brief. Readers of sophistication far beyond serendipity. It is thus
can probably find answers to their questions in the interesting that in this “final” stage, structure and
accompanying Man Group case. process are no longer the focus.
There are two reasons for this change. One is that
the structure and process by now are ingrained in
WHAT DEFINES ADVANCED-STAGE everyone at the organization. The project teams have
FINANCIAL INSTITUTIONS ON THEIR become the organization because this is how they
operate. The other reason is that in this stage, talent
AI AND BIG DATA JOURNEY? has finally become more critical than the structure
and process. This is competition at the highest level.
The natural evolution from
A team cannot expect to win in
the intermediate stage to the
the industry simply because it has
advanced stage is that AI and
set up the right T-shaped team
big data technologies are truly It may feel like you have gone and processes. All members of the
integrated into the business full circle and have ended up team need to be the best at what
processes throughout the
exactly where you started. they do and deliver solid results.
organization, from front to middle
to back office and from global But if you look under the hood, It may feel like you have gone full
headquarters to remote outposts all the machinery is now new circle and have ended up exactly
where you started. But if you look
in emerging markets. Strictly and ready to move forward under the hood, all the machinery
speaking, hardly any firm is in that at a much faster pace and in is now new and ready to move
“final” stage today and may not be
a more steady fashion. It’s a forward at a much faster pace and
any time soon. Still, some firms are
closer to that stage than others. brand new world. in a more steady fashion. It’s a
brand new world.
Another characteristic that
defines advanced-stage firms is
advanced technology. We suspect only the winners
of the industry can “advance” to this stage, which
CASE STUDY: T-SHAPED TEAMS
requires that they use technology that is superior to AT MAN GROUP
that of their peers.
Larry Cao, CFA, prepared this case. It was reviewed and
approved by Man Group.

WHAT SHOULD ADVANCED-STAGE In Spring 2009, Anthony Ledford, then head of research
at Man AHL, hired a person with a PhD in machine
FIRMS FOCUS ON? learning to join the team.
In the advanced stage, the T-shaped team structure Hiring another person with a PhD was not unusual.
and processes will become business-as-usual. Many employees in the firm’s core investment,
research, and technology functions had received
The evolution of a T-shaped team over its life cycle
PhDs in STEM (science, technology, engineering,
is interesting. It was designed to tackle a particular
and mathematics) disciplines. A few others had
and critical problem. In the early stage, the structure
backgrounds in economics or finance. But none of
and process remain simple as the team tries to get
them, at that time, came from ML.
the hang of it. The focus is on getting a small number
of quick wins, where luck could be as important as The ML specialist hire was a significant event in
the process. Both the structure and process play an hindsight: 2009 was several years before the branch
important role in the intermediate stage as getting a of ML called “deep learning” revolutionized scientific

30 | CFA Institute © 2021 CFA Institute. All rights reserved.

 T-Shaped Teams

applications (and the game of Go),23 having been live with client assets in 2014. Over time, its ML team
shunned by most ML researchers for at least a decade. has expanded, and a platform has been built that
Once that had happened, tech giants, such as Google, today not only supports Man AHL but also extends
Facebook, and Microsoft, were all over deep learning across the entire Man Group. Here is their story.
and pretty much hired every ML specialist they could
find. Still, the investment profession had yet to show Introduction
any real interest in the subject at the time.
Man Group is a global investment management
A mathematician by training who previously worked firm headquartered in London with more than
as an academic statistician, Ledford was intrigued by US$135 billion in assets under management as of
machine learning. He wanted to see whether these 30 June 2021. Its various investment subsidiaries,
new techniques could be applied to financial data to including Man AHL and Man GLG, manage investments
enhance Man AHL’s systematic investment programs. separately but share centralized infrastructure,
What he and his team found out remains relevant for including a common technology platform.
professionals today looking to apply AI and big data in
their respective investment processes. In this case, we examine the T-shaped teams at Man
AHL, a systematic investment manager, and Man GLG,
One of the first ML projects Ledford’s team worked on, a discretionary investment manager. We include how
in part to gain experience in applying ML to financial they collaborate across functions to apply various AI
data, was building a prototype long/short trading and big data techniques in their respective investment
signal using an established machine learning tool processes and how the centralized Man data science
called a support vector machine (SVM). Although the team supports both operations.
team knew this problem domain well, they tasked the
SVM with discovering the relationship between a set
of data inputs (or features, in ML parlance) and a set of Man AHL
look-ahead price changes (the targets). Other choices As noted above, Man AHL has been experimenting with
of ML tool were possible, of course, but they had to AI/ML techniques since as early as 2009. It currently
start somewhere. applies AI/ML and big data techniques—including deep
The result? The SVM “discovered” predictive learning, Bayesian machine learning, pattern matching,
relationships correlated with ones the Man AHL team reinforcement learning, and natural language processing
already knew! (NLP)—in its client investment programs. It exploits a
wide range of data sources and data types as inputs to
With the benefit of hindsight, this is not surprising. these investment models, everything from traditional
The SVM was simply capturing the relatively strong market-originating price and volume data to mall traffic,
data patterns their non-ML models already exploited. the latest consumer spending and web traffic trends,
The lesson learned: If you want to discover new weather forecasts, and text-based articles, including
relationships with ML, then you need to be able to company filings/announcements, commodity reports,
either “steer” the search away from relationships you and transcripts of interviews.
already know about or focus on different data.
Man AHL’s initial foray into ML applications provided an
Undeterred but with this acquired knowledge now part extremely important lesson: The tools themselves were
of the mix, the team later applied a different suite of rather powerful, but careful deployment was required if
ML tools to develop new long/short trading signals. new effects were to be captured. While rediscovering
This time, the results looked different from the signals what you already know provides reassurance, it does
Man AHL already traded, potentially providing useful not provide diversification—the prize all quantitative
diversification. (quant) trading researchers seek. Armed with that
Looking back at that experience today, Ledford thought understanding, when Ledford and the team decided to
the team learned a lot from the process, developing give ML models another shot, they knew they had to
a comfort with ML models and invaluable know- find something genuinely diversifying. Either you tune
how about applying them. These were just the first the ML tools to avoid things you already know (e.g., by
pieces of the puzzle, with vastly more research and penalizing similarity to existing signals or neutralizing
experimentation undertaken in the ensuing years as these in the underlying data) or you extend the range
the team expanded. Man AHL’s first ML models went of features and targets that are searched over.

See https://deepmind.com/research/case-studies/alphago-the-story-so-far#what_is_go_.
23

CFA Institute | 31
T-Shaped Teams

The ML model that later entered the client investment takes a few years of on-the-job training at Man
program went through the same rigorous testing AHL before they become efficient in working with
process that all of Man AHL’s prototype models pass financial data. Ledford, now chief scientist at Man
through on their way to client investment, from internal AHL, believes ML techniques are often misaligned
peer review to test trading with the firm’s own capital. with financial data due to such issues as low
The only difference was the number of checks. The signal-to-noise ratios and the nonstationary
ML model and the signals were significantly different relationships that are frequently found. To make
from those Man AHL already had, so more checks were ML work in investment typically requires much
required. The model performed similarly in live trading know-how and judgment that is not found in
as it did in paper trading, confirming the accuracy of textbooks. This is consistent with what we
the backtest assumptions, and it was eventually given discussed in the second hurdle of applying AI
the green light to enter client trading. and big data in investments in “AI Pioneers in
Investment Management” (CFA Institute 2019a,
In retrospect, Ledford believed the original ML research
p. 14).
modelling and the strong validation and test-trading
process combined to produce a robust strategy. The • Technology: responsible for maintaining the
firm was also well positioned when the ML model technology platform the team uses both to
became eligible for the client portfolio in 2014, as Man develop models and to manage investments.
AHL already had a client investment program with a The technology function staff typically have
mandate that sought diversifying strategies with low advanced degrees in computer science,
correlation with mainstream quant signals, which often from a research-leading university.
made it particularly suitable for ML strategies. The firm Man AHL’s technologists work closely with the
also had some good luck with timing, as the strategy research function to develop and maintain
generated positive returns soon after entering the trading strategies in production, as well as
client portfolio. Whatever the reason, the experience the entire production estate, which allows
gave senior management the confidence to approve efficient implementation and reliable execution
building out the ML team at Man AHL. of investment strategies. They also create
research tools and frameworks to increase
Building out the ML team started in earnest in 2015
research velocity.
and put Man AHL head to head with the tech giants in
recruiting ML talent. Man AHL switched to using Python across the
investment and technology functions in 2011,
which significantly simplified the workflow
Team across the firm. Prior to that, researchers used
Currently, the Man AHL core investment team is programming languages such as R and MATLAB
structured around three functions: research and while technologists implemented production
investment, technology, and data implementation. systems in Java and C++. Ledford thought
switching to Python was a fortunate choice, with
• Research and investment: responsible for benefits that were not anticipated or even on the
investing and monitoring client portfolios as well radar back then. For example, the research and
as developing the models used in that process. technology groups today work much more closely
This unit houses a number of separate groups compared with the days of separate computing
alongside the ML team, including those focusing languages, and together they make up a more
on volatility, credit, liquid strategies, macro, continuous research-to-technology spectrum
equities, and fast trading. It is very similar to what than the distinct groups that existed before the
you see at a systematic investment operation. move to Python.
The “quants” (quantitative trading analysts) at • Data implementation: responsible for ensuring
Man AHL usually have advanced degrees or have that the existing and new pipeline datasets
conducted postdoctoral research in diverse areas, needed are available and can be integrated into
such as econometrics, mathematics, statistics, the platforms and workflows supporting the
computer science/AI, or engineering. investment process.
The vast majority of these team members came The function performs similar roles as the Man
to Man AHL with practical experience in working Group data science team, with a focus on
with data. The skills are often acquired in the field, ensuring compatibility with the Man AHL platform.
although outside of investments. Still, it usually For more detailed discussion on the specific roles

32 | CFA Institute
 T-Shaped Teams

and skills required for this function, please refer to principles—namely, a tested process for generating
the section titled “The Man Data Science Team and ideas and the ability to vet and prioritize them. In this
Other Centralized Resources.” section, we highlight how the research machinery at
From a standard T-shaped team perspective, the Man AHL operates.
research and investment function just described Table 3 shows how the research process at Man AHL
matches closely with the standard investment can be divided into four stages, with three milestones
function; the technology function and the data in between that separate them.
implementation function combined play the role of the
standard technology function. It will become obvious In Stage 1, researchers, including ML specialists,
from the research process discussion that follows how generate a myriad of ideas inspired by academic
the role of the innovation function can be played by and industry research across different fields. The
any team lead and researcher. data science team will work with researchers in
data acquisition where needs arise. Accurate and
Ledford maintains that the distinction between the comprehensive data collection, usually performed
aforementioned functions is rather fluid at Man AHL automatically, is essential. The hypotheses are then
and that all staff work closely, with some staff playing put through paper trading/backtesting, a simulation
interchangeable roles across teams. process that helps reveal how each strategy would
have performed through multiple market cycles.
Research Process To ensure that models are robust, researchers also
“We start with a large number of ideas, and perform sensitivity analysis and examine performance
we need to work out which of these has the results in different historical environments. Milestone 1
potential to produce sustainable returns for reviewers help vet the strategy by looking through
our investors.” possible issues—for example, too much exposure to a
—Man AHL region or industry or drawdowns or transaction costs
that may be too high. If the risk-adjusted returns look
When it comes to investing, the devil is in the details. promising (i.e., the signal is strong and consistent
Successful investment teams largely follow the same across in-sample and out-of-sample periods, risk

TABLE 3.  THE RESEARCH PROCESS AT MAN AHL

Stage Oversight
Stage 1: Idea generation and paper
trading (backtesting)
Milestone 1: Research Review Two independent reviewers with appropriate expertise. Are research
results strong enough to embark on production version?
Stage 2: Test trading—Implementation
Milestone 2: Test-In Review Same people from Research Review.
Representatives from technology, risk, compliance, and trade
execution.
All checks to be passed before test trading starts.
Stage 3: Test trading—Conclusion
Milestone 3: Test-Out Review CIO, CRO/risk officer, CCO/compliance officer, technology, execution.
Has the strategy performed as expected?
Have all issues been resolved?
Stage 4: Client trading eligibility Allocation of AUM within a suitable client program (typically relatively
small proportion of risk budget initially).

CFA Institute | 33
T-Shaped Teams

maintains appropriate levels, and correlation with In Stage 4, portfolio managers and the CIO gradually
existing strategies suggests meaningful potential allocate funds to the strategy while continuing to
diversification), then a production version of the monitor its risk, return, and diversification performance.
strategy is constructed and a review is organized to Responsibility for the ongoing monitoring of the
seek approval for test trading to begin. strategy and its validation against paper trading is
retained by the original research team, ensuring that
Milestone 2, the test-in review, is an important
the people who know the strategy the best keep their
milestone as it separates an investment strategy with
eyes on it. Periodic reviews of all strategies across the
production-standard implementation code from a
entire client trading portfolio are also undertaken.
research idea with backtest simulations. Because of
its importance, this is where the entire multi-function That, in a nutshell, is the process of how an individual
team is set in motion. The strategy is reviewed from strategy goes from an idea to becoming part of the
the validity of the hypothesis to the efficiency of its investment process. Although the CIO oversees
implementation. The technology team will review the overall process supported by the investment,
the code thoroughly to ensure it reflects what the execution, risk, and compliance teams, there is
researchers intended, is robustly implemented, and a high degree of autonomy within the individual
replicates the original research results. The chief risk investment strategy R&D teams, each having its own
officer (or risk team representative) will determine specialist research and technology resources. And
whether the strategy adheres to acceptable risk most initiatives remain “bottom-up” in their origin,
parameters. Compliance will make sure the strategy with C-suites identifying longer-term opportunities
is consistent with all relevant rules. The execution and themes. Usually tens of strategies are being
team will offer its opinion on whether the strategy researched and tested at any point in time, and regular
can potentially achieve results in live trading similar firm meetings are held in support of these processes.
to those in paper trading. If the strategy receives Project meetings, in contrast, take place frequently
everyone’s endorsement, test trading can commence. but only as needed; they do not follow a particular
At this stage, the firm’s own funds are used (i.e., no schedule. It usually takes several months for a strategy
client capital is placed at risk). to migrate from initial research through test trading
and into the client portfolio.
Once a suitable history of live trading experience
has been built up, another review is called in order to Ledford believes that a collaborative culture is also
establish whether the strategy can leave test trading about carefully documenting research results, both
(Milestone 3, the “test-out review”) and should be those that work and those that do not, so that all
made eligible for client investment. The duration of teams can benefit from the research conducted
test trading is flexible; for example, if a strategy is previously and the lessons thereby learned.
notably different from existing strategies, then a longer
period is merited, as was the case for the previously Oxford-Man Institute of Quantitative
discussed ML strategy. Issues that arose during the
test-trading period and how they were addressed Finance
are reviewed, and the realized profit and loss (P&L) is At the Oxford-Man Institute (OMI) we aim
validated against the paper trading research P&L over to do academically outstanding research
the test-trading period. that addresses the key problems facing the
Test-out review is the most important milestone in financial industry. We create new tools and
the chain because a strategy can benefit investors methods that can give deeper insight into
in Man AHL’s programs only once it enters client financial markets—how they behave, how they
trading. Participants for the test-out review are the become stable or unstable and how to extract
same as those in the test-in review plus C-suite value from diverse data at scales beyond
representatives. They need to be fully authorized to human [powers]. . . . We achieve this through
greenlight the strategy for client trading, a significant a unique combination of academic innovation
responsibility. The participants closely review the live and external engagement.
trading results and make sure that they perform in line —Professor Stephen Roberts, OMI Director
with those in paper trading. Inputs from the operation and Chair of Machine Learning, Department
and execution teams are both essential. If all checks of Engineering Science, University of Oxford
are met, a strategy will have finally reached the end
zone, becoming eligible for client trading. In the spirit of collaborating with academia to generate
interesting investment research ideas and techniques

34 | CFA Institute
 T-Shaped Teams

and to ensure a pipeline of suitably qualified PhD-level GLG was a discretionary firm set up in 1995 and
quantitative finance researchers for both the industry acquired by Man Group in 2010. The discretionary
and academia, Man Group and the University of Oxford team’s composition and background have not changed
jointly founded the academically focused Oxford-Man very much. Roles are not much different from a typical
Institute of Quantitative Finance in 2007. Since 2016, discretionary investment manager: Analysts research
OMI’s focus has been academic ML and data analytics companies, and portfolio managers make portfolio
research with applications to quantitative finance. decisions.
Academic research undertaken at OMI can include The quant team’s goal is to help discretionary
areas that at first may seem unrelated to investment— managers make more money. In addition, quants trade
for example, techniques for supernova identification. some portfolios directly using their models.
However, even there Ledford and his team saw
In an interview with Chambers, he asserted that a
applications for stock picking.24 The key, he said, is to
ratio of 1:1 is an ideal mix of quant researchers and
identify problems of the same shape. The supernova
technologists, although, like Ledford, he maintained
application involved combining many conflicting
that the boundary between a quant researcher and
views about whether an astronomical photograph
a technologist is increasingly blurred. They share a
showed a supernova or not, while the stock picking
similar interest in coding and generally come from
application involved combining many conflicting
a scientific field of investigation. Their industry
analyst recommendations about whether the price of a
backgrounds and experience levels may vary, from a
stock would go up or not. It is essentially the same ML
freshly minted physics PhD to a veteran trader in exotic
problem but with different features and targets.
options, but by and large their purpose is to distil
Indeed, understanding where and how ML has been actionable insights from data.
successful in other application areas is something
Ledford believes to be crucial for deploying ML in Research Process
financial applications, as virtually no ML technique has
its origins in finance. The other ingredients he noted How do the two functions work together? There is
that are needed for success are diversification within no standard process for the teams to follow, but it
the areas of alpha drivers, modelling techniques, generally takes place in one of two ways.
trading frequencies, and tradable instruments/ Portfolio managers often have specific hypotheses/
markets. intuitions that have not otherwise been scientifically
researched. The quant team can potentially back
Man GLG up the theory, enhance it, or reject it—all with data.
Portfolio managers at Man GLG now have access
Man GLG is the discretionary investment engine of
to a dashboard of insights as their theories get
Man Group. It has a long history of using AI and big
investigated so that they can put them into use.
data techniques in equity and credit strategies. The
efforts became more deliberate after Paul Chambers, Alternatively, quants regularly propose their own ideas
a nine-year veteran of the Man AHL equities team, to portfolio managers. The data they rely on can come
was chosen to head the quant group at Man GLG in from many sources, including external sources and
September 2019. internal data on the portfolio managers’ behaviour.
Chambers put the mix of the these two approaches
Team at about 50–50. He helps prioritize the quant projects,
The investment team at Man GLG is made up of (1) where necessary. Portfolio managers, in contrast,
the discretionary portfolio managers (PMs) and their prioritize their own requests to the quants.
respective analyst team and (2) the quant team, which Both functions sit on the same floor; in fact, all staff
includes quant researchers and technologists. The on the Man AHL investment team sit on the same
quant team structure is similar to the Man AHL team floor as well. Chambers stated that sharing research
described earlier. The discretionary PM teams each work and collaborating on research efforts across the
have a unique focus by region, industry, or style; each engines are encouraged. The insights can usually be
has its own dedicated analysts. incorporated into the portfolios in different ways given

See Bew, Harvey, Ledford, Radnor, and Sinclair (2019).

24

CFA Institute | 35
T-Shaped Teams

their different investment processes, but transparency it became evident to Man Group’s executive committee
of the research agenda helps everyone. that a central function responsible for dealing with
big data would be required—becoming a shared
For example, a discretionary investor might use a
resource for the entire group. Theoretically, the signal
signal based on real-time news to inform a decision
generated with big data should be able to add value
on an individual stock. Conversely, a quant team
to investment processes at large, be it a systematic
might use the same data to trade all 3,000 stocks in
strategy or a discretionary strategy.
an investment universe and profit from a very small
expected payoff across many names.
Team
The Man GLG team also works closely with the data
science team. The Man GLG team may go to the data The central function was created in early 2020, with
science team with a specific data source, where the a total of 8 staff members, and has now grown to an
latter will simply operate as a project manager working operation of 15 between Man Group’s London and
with all internal and external parties involved to Boston offices. Its structure has now expanded to
acquire the data, bringing it in-house quickly. The Man include the following four groups.
GLG team may also request specific data that the data • Data sourcing and strategy
science team then needs to locate and vet the vendor.
Often called “data scouts,” this sub-team manages
The most complicated request comes in the form of a
the full data acquisition life cycle, which typically
question for which the data science team will need to
starts from a request by the investment managers
research the issue involved and find the right data to
to legal and compliance review. A typical day’s
explain it.
schedule involves scouting for new data vendors,
Chambers emphasized that how quickly an investment maintaining vendor relationships, and attending
team can bring data on board matters in this day and weekly and monthly meetings of the investment
age. Project turnaround time at Man GLG can now be managers so as to understand their needs and
as short as a matter of weeks rather than months. help resolve data-related issues. This group is
So, working closely with the data science team and constantly plugged into the data community,
collaborating with other investment engines all help in bringing new and interesting datasets to the
that regard. investment teams for evaluation.
When interviewed for this case, he commented, The background required varies, from business to
technology and from finance to data science. The
My view is that innovation—incorporating key is knowledge about the big data landscape,
the latest data and analysis—is and always including data points and vendors out in the
has been crucial to the success of any market, and an appreciation of how big data fits
discretionary investment manager. That said, into the investment puzzle.
it requires a very significant investment and
• Data scientist
a particular set of expertise. The same can
be said of systematic investors, who need to This is where the magic happens. The data
work harder and harder to maintain returns. scientists aim to turn raw, unstructured data
We’re fortunate that we have both disciplines into insights for investment teams. Big data is
under one roof at Man Group so we can prone to having much noise and is challenging
leverage each other’s skill sets. I don’t believe to handle. This sub-team’s main challenge is to
incorporating the latest data techniques distil insights from the seemingly random piles of
is optional; those firms that are not able to data that can be used by investment teams. The
keep up are at risk of diminishing returns and process is not much different from what the Man
eventual closure. I’d say that this view is fairly AHL researchers do, except the end result here is
consistent across our teams and our clients. not to create a stock forecast or an investment
decision but to communicate insights from the
The Man Data Science Team data that can be applicable for a quantitative
researcher or portfolio manager. As such, the
and Other Centralized Resources data scientists also need to work closely with the
As different investment engines within Man Group investment managers who are end users of their
started increasing their exploration and application insights. In addition, the data scientists work with
of big data in their respective investment processes, technology to introduce various analytical tools for
the benefit of Man Group’s investment teams.

36 | CFA Institute
 T-Shaped Teams

Data scientists usually hold computer science Critical factors for the success of a central function
PhDs or advanced degrees in similar subjects, are agility and visibility. Four cross-functional teams
such as statistics, science, or finance. Some require constant communication. They engage daily
are CFA charterholders. The key is their ability to in “cross-team standups,” data discovery blogs, and
program, preferably with Python, and the ability monthly “data demos” to ensure transparency across
to work with “big data” (i.e., to be able to process the entire staff. Visibility is key to prioritization and
unstructured and/or large volumes of data and be agility. The central function continuously engages
good “storytellers” of the data). with investment teams to ensure they are aware of
• Data science engineering the successes and failures around the application of
big data. In other words, “Everyone is in the know,” as
These technologists are often trained in computer
Hinesh Kalian, who leads the central Man data science
science, although a PhD is not always required.
team, explained in an interview.
They are embedded within the Man data science
function. They are responsible for supporting
data scientists, and their main role is to build and Centralized Execution and
maintain the data platform infrastructure so as Technology Functions
to speed up the process of getting data to the
investment managers. In addition to the data science team, Man Group also
has centralized execution and technology functions.
They are usually paired up with data scientists at As is the usual practice in the investment industry,
a ratio of two data scientists to one data engineer. risk, compliance, legal, and other corporate functions
This is higher than otherwise seen in the industry are also centralized at Man Group. We provide here
because some of Man Group’s data engineers some important highlights of the centralized execution
work in the centralized group operations, which is and technology functions.
outside of this team.
• Data management In 2017, Man Group combined execution research,
execution technology, and the trading teams from
This sub-team is responsible for the production Man AHL, Man GLG, and Man Numeric. The combined
aspects of data that feed Man’s investment team now serves the trade execution needs of all Man
models and portfolios. The team manages Man Group investment engines, encompassing its own
Group’s central security master, responsible for research, technology, and trade execution functions.
data tagging and metadata, quality assurance, The execution researchers are similar to the quant
and data management tools. The background researchers at Man AHL and Man GLG, but they also have
required for this group also varies; however, their an intimate knowledge of trading and exchanges. The
ability to program, preferably with Python, is a key majority of trading is fully automated today; however,
requirement. the desk traders provide an indispensable role in cases
where human skill still offers a material advantage
Process (e.g., in markets with limited liquidity or where electronic
The data sourcing and strategy team meets monthly execution is not available or well developed). The
with all investment teams to review investment trading execution team ensures all trading is carried out
themes and related data requests. A governance efficiently and with adherence to best execution.
committee manages the governance framework and In addition to the technologists at each Man Group
ensures data compliance throughout the new data subsidiary and the data science team that manages
production pipeline. the specific technology needs of the unit, a platform
The typical process begins with a proof of concept team focuses on the quant computing needs of
during which a trial agreement is generally required. the front office. Its remit covers the whole data
During the trial phase, the data scientists work with engineering function and includes
the vendor to ingest, transform, curate, and visualize • maintaining the databases and data platform tools
insights for the investment teams. Once proven, the used in research and trading;
team secures sign off from legal and compliance
• the computer infrastructure underpinning the
teams and signs commercial data license agreements
trading systems;
with the vendor. Costs are shared among investment
teams based on data usage. • the high-performance computing clusters used for
research and, in particular, a large Spark cluster used
for data science and machine learning jobs; and

CFA Institute | 37
T-Shaped Teams

• the research and development tools used by within an investment engine can all easily become
quants and engineers across the Man Group their own silos that put their own interests first.
business.
Man Group has made a conscious effort to avoid
In addition, a centralized Man Group technology the downside of silos by building teams. Putting
function houses teams that look after corporate professionals from multiple functions with distinct
IT needs. These needs include desktops; laptops; cultures and skill sets together in one team helps them
phones; all of the finance, project, and reporting tools; collaborate more smoothly and effectively. The fluidity
the order management and settlement systems; in the process helps enforce the team’s interests and
the investment book of record; databases; the data objectives.
centres; and cloud-hosted services. An NLP platform
was also made centrally available and is now managed Align Incentives
by the Man Group technology team.
Teun Johnston, CEO of Man GLG, made a conscious
decision to embed a quant team in the firm so
Key Takeaways that the incentives are aligned across the Man
Design Your Own T-Shaped Team GLG team. Evaluation of both the discretionary
portfolio managers and the quants is based more
The T-shaped team structure and process is a general on the overall investment team performance and
model of organizational structure and process that how each professional on the team aligns with the
emphasizes team objectives over function or individual team objective.
objectives. To make it work in the most efficient
manner, Man Group operates a somewhat different It is not unusual for discretionary managers to have
form from what is seen at a typical investment firm. doubts about using AI and big data in their process,
It could do so for the following three reasons: despite the fact that these techniques tend to add
to their performance rather than detract from it
1. Historically, the Man AHL staff across the core (CFA Institute 2019b, p. 16). Overcoming these initial
investment, research, and technology functions hurdles is obviously important. Shared goals and
have rather similar backgrounds, so the need for a aligned incentives will go a long way in easing the
rigorous T-shaped structure with distinct elements portfolio managers’ concerns.
is not as strong when there is fluidity across
functions with shared knowledge and skills. Similarly, this speaks clearly to the notion that the
quant team is part of the alpha generation process and
2. The staff have become familiar with the research an equal partner in this journey.
process, which has become more or less part of
everyone’s habit and flows very smoothly.
3. Most importantly, complementary to the first two
Hire ML Specialists
reasons, Man Group has nurtured a culture of This is an important piece of the puzzle that may get
collaboration, transparency, and accountability overlooked. Many quants can do ML, but to operate at
that is ensured by the peer review process. the highest level, the team still needs ML specialists.
ML has become such a distinct discipline that it is
Break Down Silos difficult for people without dedicated training and
Silos can come in many shapes and forms, but they work experience to be able to deliver the best results.
share one common unpleasant trait that does not help Despite the fluidity in its processes, Man Group
the team and, ultimately, the organization achieve their placed high importance in the special skills that each
objectives. Different investment engines within the professional brings to the team.
corporate structure or different functional departments

38 | CFA Institute
REFERENCES

Albanese, Robert, and David D. Van Fleet. 1985. www2.deloitte.com/content/dam/insights/us/

“Rational Behavior in Groups: The Free-Riding articles/6546_talent-and-workforce-effects-in-the-
Tendency.” Academy of Management Review age-of-ai/DI_Talent-and-workforce-effects-in-the-
10 (2): 244–55. age-of-AI.pdf.
Bean, Randy. 2020. “Why Culture Is the Biggest Barrier Karau, Steven J., and Kipling D. Williams. 1993. “Social
to Data Success.” MIT Sloan Management Review Loafing: A Meta-Analytic Review and Theoretical
(30 September). Integration.” Journal of Personality and Social
Psychology 65 (4): 681–706.
Bew, David, Campbell R. Harvey, Anthony Ledford, Sam
Radnor, and Andrew Sinclair. 2019. “Modeling Analysts’ McKinsey & Company. 2020. “The State of AI in 2020”
Recommendations via Bayesian Machine Learning.” (17 November). www.mckinsey.com/business-
Journal of Financial Data Science 1 (1): 75–98. functions/mckinsey-analytics/our-insights/
global-survey-the-state-of-ai-in-2020.
CFA Institute. 2019a. “AI Pioneers in Investment
Management” (September). www.cfainstitute.org/-/ Mellers, Barbara, Eric Stone, Terry Murray, Angela
media/documents/survey/AI-Pioneers-in-Investment- Minster, Nick Rohrbaugh, Michael Bishop, Eva
Management.ashx. Chen, Joshua Baker, Yuan Hou, Michael Horowitz,
Lyle Ungar, and Philip Tetlock. 2015. “Identifying
———. 2019b. “Investment Professional of the
and Cultivating Superforecasters as a Method of
Future” (May). www.cfainstitute.org/en/research/
Improving Probabilistic Predictions.” Perspectives on
survey-reports/investment-professional-of-the-future.
Psychological Science 10 (May): 267–81.
Eric, Mihail. 2020. “A Complete 4-Year Course Plan
NewVantage Partners. 2020. “NewVantage Partners
for an Artificial Intelligence Undergraduate Degree”
Releases 2020 Big Data and AI Executive Survey.”
(May). www.mihaileric.com/posts/complete-artificial-
Business Wire (6 January). www.businesswire.com/
intelligence-undergraduate-course-plan/.
news/home/20200106005280/en/NewVantage-
Fountaine, Tim, Brian McCarthy, and Tamim Saleh. Partners-Releases-2020-Big-Data-and-AI-Executive-
2019. “Building the AI-Powered Organization.” Survey.
Harvard Business Review (July–August). https://hbr.
Oh, Rachel, and Peter Maldonado. 2021. “The Rise
org/2019/07/building-the-ai-powered-organization.
of AI in Stanford’s CS Curricula.” Stanford Daily
Goasduff, Laurence. 2019. “3 Barriers to AI Adoption.” (4 February). www.stanforddaily.com/2021/02/04/
Smarter with Gartner (18 September). www.gartner. the-rise-of-ai-in-stanfords-cs-curricula/.
com/smarterwithgartner/3-barriers-to-ai-adoption/.
Squirrel, Douglas, and Jeffrey Fredrick.
———. 2020. “Avoid 5 Pitfalls When Building Data and 2020. “Solving the Problem of Siloed IT in
Analytics Teams.” Smarter with Gartner (9 July). www. Organizations.” MIT Sloan Management Review
gartner.com/smarterwithgartner/avoid-5-pitfalls- (7 July). https://sloanreview.mit.edu/article/
when-building-data-and-analytics-teams/. solving-the-problem-of-siloed-it-in-organizations/.
Hansen, Morten T. 2010. “IDEO CEO Tim Brown: Susskind, Richard, and Daniel Susskind. 2015.
T-Shaped Stars: The Backbone of IDEO’s Collaborative The Future of the Professions: How Technology
Culture.” Chief Executive (January 21). Will Transform the Work of Human Experts. New York:
Oxford University Press.
Hansen, Morten T., Nitin Nohria, and Thomas J. Tierney.
1999. “What’s Your Strategy for Managing Knowledge?” UOBAM. 2019. “UOB Asset Management and Value3
Harvard Business Review (March–April). Advisory to Launch AI-Enabled Credit Rating Platform
in ASEAN for More Effective Bond Investments.”
Hupfer, Susanne. 2020. “Talent and Workforce Effects
Press release (13 November). www.uobam.com.sg/
in the Age of AI: Insights from Deloitte’s State of
web-resources/uobam/pdf/uobam/about-us/media-
AI in the Enterprise, 2nd Edition Survey.” Deloitte
releases/news-rls-uobam-MOU-with-Value3.pdf.
Center for Technology, Media & Telecommunications.

© 2021 CFA Institute. All rights reserved. CFA Institute | 39

ACKNOWLEDGEMENTS

We gratefully acknowledge the helpful comments, greatly from our interactions with speakers and
suggestions, and reviews by Anthony Ledford (Man participants at CFA Institute, CFA Society, and industry
Group and Oxford-Man Institute), Bilian Ni Sullivan events organized in Frankfurt, Hong Kong, London,
(Hong Kong University of Science and Technology), Montreal, Shanghai, Shenzhen, Singapore, Sydney,
and Giorgio Valente (Hong Kong Institute for Financial Tokyo, and Toronto.
and Monetary Research). The project also benefitted

40 | CFA Institute © 2021 CFA Institute. All rights reserved.

CFA INSTITUTE STAFF

Larry Cao, CFA, Senior Director, Industry Research

Author
Rhodri Preece, CFA, Senior Head, Industry Research
Editor
Paul Andrews, Managing Director, Research,
Advocacy, and Standards
www.cfainstitute.org
[email protected]