SPONSORED CONTENT

Shared by The Ravit Show

A Strategic Playbook for

DATA,
ANALYTICS, and
MACHINE
LEARNING

1 u
Introduction Contents
Organizations that once stored gigabytes
of business data now find themselves swimming in

u The cloud is also the conduit to highly accessible,
pay-as-you-go compute and storage resources,
3 Wdata-driven
anted: A modern,
enterprise
petabytes or even exabytes of information. Harnessing which makes it easy for organizations to scale based
the value of these massive data volumes requires a
modern, cloud-based data infrastructure that unifies
on need while harnessing and analyzing every last bit
of data for critical business insights.
4  imitations of
L
monolithic architectures

5 
disparate information silos. This approach creates a
Organizations now have the opportunity to marshal The benefits of a
universal version of “truth” and empowers everyone in
data, analytics, and machine learning (ML) to promote
cloud-based data foundation
an organization to make more informed decisions and
act with confidence. critical business decision-making. Such data-driven
insights can guide countless business scenarios, 6 How to democratize your data
8 Gwith
A few key trends have brought us to this point: including: et more from your data

u The digitalization of businesses, coupled with a u Launching new product offerings machine learning
borderless Internet that extends application reach

10 Klearning
u Innovating on new business models or revenue
anywhere, has given rise to unprecedented data eys to scaling machine
streams
growth from mobile devices, IoT devices and sensors,

12 5modern
cloud-based apps, social media, log files, click-
 utomating manual processes to improve efficiency
u A

u Improving customer experiences to build trust and

steps to building a
streams, and other sources.
data foundation

u The cloud has driven down the cost of storage,
loyalty

allowing organizations to cost-effectively manage

the deluge without having to decide up front what
 ptimizing interactions with business partners
u O
12 Summary
Regardless of where you are on your data moderniza-
data is worth keeping and what should be discarded.
tion journey, this playbook will help you refine your
strategy to effectively scale data, analytics, and machine
learning across the enterprise, so you can accelerate
innovation and drive your business forward.

A STRATEGIC PLAYBOOK FOR DATA, ANALYTICS, AND MACHINE LEARNING

2 u
t
Wanted: A modern, data-driven enterprise
Becoming a data-driven organization  appos is using cloud-based analytics and machine
u Z
is achievable through a mix of technology, people, and learning to personalize product sizing and search
processes, bolstered by a broad view of data as a stra- results for retail shoppers while preserving a highly
tegic asset. While most organizations have data-driven fluid and responsive user experience. The result:
transformation in their sights, the shift remains elusive Zappos has reduced repeated searches and product
for many: returns, achieved higher search-to-product-click-
through rates, and raised the position of customer
 9% of organizations have invested in data and
u 9
selections in search results.
artificial intelligence (AI) initiatives

 5% have spent more than $50 million on these

u 6
 fter The Pokémon Company International (TPCi)
u A
launched its Pokémon GO mobile game in 2016,
efforts
the number of users requiring access to the system
But… increased to more than 300 million in two years. To
address the massive influx of Pokémon GO users,
u J ust 38% say they are a data-driven organization TPCi migrated from a third-party NoSQL document
database to AWS (Amazon Web Services) fully man-
 nly 27% say they’ve successfully built a data-
u O
aged database services, which allowed it to reduce
driven culture (Source: 2020 NewVantage Partners
the number of database nodes from 300 to 30 and
Big Data and AI Executive Survey)
decrease its monthly database costs by tens of thou-
The numbers show a clear gap between organizations’ sands of dollars.
aspirations for using data strategically and the reality of
These examples underscore the benefits of moving
what they’ve accomplished to date.
away from antiquated, monolithic applications that
Some have already taken steps to close the gap, run on one-size-fits-all relational databases to highly
using the cloud as a foundation for a modern data distributed, microservice-based systems running on
architecture: multiple purpose-built databases. It also means moving
from on-premises and old-guard legacy data ware-
 eorgia-Pacific migrated 50 TB of structured
u G houses to open and flexible data lakes and “lake house”
and unstructured production data from a legacy architectures.
database infrastructure to a cloud-based data lake
in order to cost-effectively ingest, transform, store, “It’s not a one-size-fits-all world anymore,” says Shawn
and analyze that data. Layering analytics tools on Bice, Vice President, Databases, with AWS. “How you
the data has helped Georgia-Pacific optimize key approach data in the cloud is the foundation for future
manufacturing processes, including the ability to business growth. A native cloud application architecture
predict equipment failure 60 to 90 days in advance, allows businesses to innovate faster than ever before, at
which reduces unplanned downtime. lower cost, and faster time to market because you’re not
limited to what one thing can do.”

A STRATEGIC PLAYBOOK FOR DATA, ANALYTICS, AND MACHINE LEARNING

3 u
t
 INNOVATION SPOTLIGHT

Amazon.com
Amazon.com once maintained one of
the biggest Oracle data warehouses in the
world, but it wasn’t enough to keep pace
with Amazon’s growth.

“Five years ago, our ability to grow and

analyze our business was limited by
technology choice,” says Jeff Carter, Vice
President of Data, Analytics, and Machine
Learning at Amazon.

Amazon made the strategic decision to

move off the legacy data warehouse and
onto a cloud-native data lake architecture
comprising a variety of AWS services. The
migration involved moving 50 petabytes of
data, 75,000-plus data warehouse tables,
and 7,500 OLTP databases supporting
the company’s business-critical ordering,
processing, and fulfillment systems.
Limitations of monolithic architectures Carter admits to some initial concerns
about such a massive migration causing
Organizations that continue to struggle with Most CIOs recognize the limits of legacy data infrastruc- disruptions. “We didn’t want to be the team
legacy infrastructure likely face one or more well-known ture—but may hesitate to push their organization out that shut down the business,” he says. “But
challenges involving people, process, and technology. of its technological comfort zone. Such a short-term what we found was in pretty much every
Becoming a data-driven organization requires new perspective could inhibit long-term business growth. instance, availability was better.”
thinking about all three.
“Don’t let familiarity turn into a blind spot that stifles “By migrating to the AWS technologies and
Traditional data analytics architectures rooted in struc- innovation,” says Bice. implementing the data lake, we have been
tured databases and relational data warehouses have able to scale to meet our business needs,”
grown costly to manage, update, and secure. These says Carter, adding that the new environ-
limitations prevent enterprises from fully monetizing ment is about 30-50% less expensive to
valuable data. Legacy architecture wasn’t built to sup- maintain than the previous architecture.
port enterprise-wide data management and analytics at
the scale required to handle petabytes or even exabytes
of data across the enterprise.

A STRATEGIC PLAYBOOK FOR DATA, ANALYTICS, AND MACHINE LEARNING

4 u
t
 INNOVATION SPOTLIGHT

Equinox Media
As COVID-19 bore down in the spring of
2020, fitness brand Equinox recognized
the need to fast-track the launch of its
Variis online fitness streaming platform.
With the company anticipating a surge in
demand for in-home fitness services, the
development team looked to learnings

The benefits of a cloud-based data foundation

from Equinox’s Fitness environment to
develop a modern and scalable data infra-
structure based on an Amazon S3 data
Unlike the rows/columns/table structure of a tradi- lake, and serverless architecture including
A modern, cloud-based data infrastructure
tional data warehouse, a modern data architecture can AWS Lambda, Amazon DynamoDB, and
is essential to gain the flexibility and scalability needed
store all types of semi-structured or unstructured data Amazon Athena.
to react quickly to changing business needs. With a
modern data architecture, there are virtually no limits such as web logs and images, eliminating the need for
By developing Variis from the ground up
to how much and what kind of data an organization can separate data silos. A centralized data lake also enables
with a cloud-native infrastructure, Equinox
store and manage, opening up myriad possibilities for the ability to tag and catalog data to make it discover-
Media was able to accelerate the new
leveraging information in new and better ways across able, ingest and process data in real time via streaming
platform’s launch and quickly scale up to
the business. technologies, and apply security controls and permis-
meet demand, says Elliott Cordo, Equinox
sions to the data to promote governance and maintain
Media’s Vice President of Technology
“Decades ago, databases were designed to be opti- compliance.
Insights. Leveraging the highly scalable,
mized for storage because the costs were so high,”
Elasticity is another advantage of a modern cloud- serverless data platform, “we ramped
explains Herain Oberoi, Director of Product Marketing,
based data infrastructure. A traditional environment from beta users to launch in just a matter
Databases, Analytics, and Blockchain, AWS. “Cloud
requires investment in software licenses, infrastructure, of weeks,” says Cordo.
economics have removed the constraints of having to
decide what data to store and what to discard. Now the and data center horsepower to drive big analytics
“We launched a startup at very low cost
default is, let’s just store everything because we might workloads, even if they are temporary to accommodate
with the confidence that we could scale it
not know what we want to do with the data, what a surge or a specific use case. Not so in the cloud, where
with reliable cost prediction,” he says. “The
questions we want to ask, or what insights we might get you can spin up or scale back infrastructure based on
ability to handle explosive growth clearly
down the road.” the needs of the analytics workload. That provides an
demonstrates the advantages of modern
opportunity to start small and go big or front-load
data engineering and cloud-native design.”
horsepower to handle a temporary surge period and
only pay for the capacity used.

“Fundamentally, cloud is about low-cost and elastic

storage and compute,” says Oberoi. “That makes
analytics the perfect workload for cloud.”

A STRATEGIC PLAYBOOK FOR DATA, ANALYTICS, AND MACHINE LEARNING

5 u
t
 5
characteristics
of a data-driven
organization

1
Go beyond “executive sponsor-
ship” to “executive engagement”
by making visible changes, starting at
the top, to put data to use across the
enterprise.

2
Focus on creating a data-driven
culture and building organiza-
tional capabilities to support the
culture, at a scale beyond just a few
How to democratize your data islands of excellence.

3
Embrace data-driven experi-
Modern data management and analytics tools, Ishit Vachhrajani, Enterprise Strategist at AWS and the mentation to test many ideas and
combined with cloud-based infrastructure, provide the former Global CTO at A+E Networks, prefers the term continually improve the business.
underpinnings of a unified, holistic view of data across “data proficiency” to data literacy, because the goal

4
the enterprise. But tools are only as good as the people goes beyond data awareness to knowing how to put it Use a strong foundation of data
who use them. What most often trips up data-driven to use. “It’s not just about democratizing access to data,” analytics to power transforma-
initiatives are corporate culture challenges and low he says. “It’s about democratizing action using that tion using artificial intelligence
levels of “data literacy,” as New Vantage Partners and data.” and machine learning.
other researchers have noted.

5
Vachhrajani recommends the following steps to create a Eliminate silos by treating data
A critical step toward becoming data-driven involves data-proficient culture: as an organizational asset, not a
better education and understanding of the value of departmental property, and by bring-
u Investigate how data flows in your organization
data, from the CEO on down. The goal is to put data in ing data transparency and account-
and what gatekeeping controls are in place. This
the hands of everyone and enable them to use data for ability to the entire enterprise.
helps uncover data silos and gauge the level of diffi-
everyday choices, not just big decisions.
culty for employees to access the data they need.

A STRATEGIC PLAYBOOK FOR DATA, ANALYTICS, AND MACHINE LEARNING

6 u
t
u Make sure a senior, well-respected, and empow- Cultural change, of course, takes time. But there are INNOVATION SPOTLIGHT
ered leader is driving the cultural initiative to
become a truly data-driven company.
plenty of benefits to capture along the way. “The
steps you take will start to deliver wins early on,” says
Fortnite
Vachhrajani. “And you’ll be able to show people how the Epic Games uses AWS to deliver Fortnite
 reat data as a product, in part by bringing
u T wins fit in the context of the bigger picture.” to more than 250 million players around
application engineers and data engineers together,
the world, along with more than 10 million
and also by very closely aligning data strategy with How do you know when cultural change is taking hold?
concurrent players during in-game events
product and integration strategy. When people start pushing back. “Change is uncomfort-
such as the world’s largest concert.
able and can cause a lot of friction,” Vachhrajani says.
 ake IT a key player. IT has a unique view of the
u M “The stronger that resistance is, especially early on, is a The Fortnite team needed a way to process
end-to-end business cycle, cross-departmental sign that things are working. It’s important to feel that and analyze over 100 petabytes of data
workflows, and transactional systems that hold valley of despair when you’re going through change (125 million events per minute) ingested
useful insights. because that tells you you’re moving the needle.” from game clients and game servers to
 reate a data governance structure that enables
u C understand and adapt to player engage-
employees rather than restricts them. ment. Epic Games turned to AWS for an
Amazon S3 data lake in combination with
Amazon EMR, Amazon EC2, and Amazon
Kinesis to perform real-time analytics that
fuel continuous game improvement.

A STRATEGIC PLAYBOOK FOR DATA, ANALYTICS, AND MACHINE LEARNING

7 u
t
 Machine Learning
and AI

78%

of organizations are

Get more from your data with machine learning

considering or have already
deployed machine learning
technologies as part of a
digital business strategy
With the proliferation of data and virtually Leadership teams should not limit their thinking around
unlimited quantities of computing power available via potential use cases for AI and machine learning. “Every
the cloud, AI and machine learning are poised to have part of the organization should be thinking about
a profound impact on an increasingly broad canvas of machine learning,” says Sri Elaprolu, Senior Leader with 63%
business and technology. the Amazon Machine Learning Solutions Lab.
of companies have invested
The ability to analyze massive amounts of data to For example:
in machine learning to close
derive meaningful insights can help businesses make competitive gaps with rivals
u In sales and marketing, machine learning and AI
faster, more informed decisions. It’s clear that AI and ML
can help analyze activities and behavioral data to
are positioned to give companies a competitive edge:
create personalized experiences for customers.
 8% of organizations are considering or have
u 7
already deployed machine learning technologies as
u In finance, the technologies are empowering more 54%
accurate forecasting models and fraud detection
part of a digital business strategy. (Source: IDG 2019
analysis.
Digital Business Study) of businesses deploying
 perations and logistics are tapping AI/ML to
u O
AI have experienced
u 63% of companies have invested in machine learn- productivity increases
optimize order fulfillment and improve shipping
ing to close competitive gaps with rivals. (Source:
routes for last-mile delivery.
Deloitte, State of AI in the Enterprise report, 2018)
u IT is parlaying tools to aid in security threat
u 54% of businesses deploying AI have experienced
detection and mitigation and to streamline software
productivity increases. (Source:: PwC, AI Predictions
development operations.
for 2018)

A STRATEGIC PLAYBOOK FOR DATA, ANALYTICS, AND MACHINE LEARNING

8 u
t
“The cloud enables extremely low-cost compute and
storage, which opens up opportunities for more mod-
eling,” says Elaprolu. “There’s lots of innovation yet to
happen. We are barely scratching the surface.”

Beyond functional use cases, many sector-specific

examples of AI/ML are emerging as well.

u In healthcare, technology company Cerner

employs machine learning to help predict conges-
tive heart failure up to 15 months before it man-
ifests in clinical tests, improving patient care and
ultimately saving lives.

u In transportation, Convoy disrupted the trucking

industry by introducing a machine learning-powered
model to automate logistics and provide better
matches for shippers and truckers, which drove effi- INNOVATION SPOTLIGHT
ciencies and lowered costs for both parties.
T-Mobile
u In the tech sector, Intuit, creator of TurboTax, lever-
Interestingly, success metrics for the technology
ages computer vision and machine learning to help At T-Mobile, engineering teams are using AI/ML
are the same as for human agents, such as first-
users file their taxes more efficiently. Using Amazon technologies to transform customer support.
call resolution.
Textract, the solution lets users scan pictures of their Models leverage natural language processing to
tax forms, and the service verifies accuracy to iden- extract insights and relationships from unstruc- “We can measure based on just the accuracy
tify any missing data or anomalies using contextual tured text, as well as capabilities such as voice- of the models, but that doesn’t actually help
data from an existing database of tax codes and to-text, text-to-speech, and language transla- the business,” says Nolis. “The things we are
compliance forms. tion, with the goal of improving the customer held accountable for are the same KPIs our care
experience. agents are held accountable for: how satisfied
u In food services, Domino’s Pizza Enterprises
was the customer at the end of the call? Did they
Limited, a Domino’s franchise holder with brands in Machine learning models sit between a care
have to call back again to solve their problem?
Australia and Europe, created a predictive ordering agent and the customer, scanning previous
Because [improving] the way that our customers
solution to help stores anticipate what pizza their customer interactions and serving up relevant
feel about our business and the way they expe-
customers would order. The capability, part of a information to the care agent to help them
rience our business is what we should be aiming
broader initiative to reduce pickup and delivery quickly address the customer’s issue. “We want
for all the time.”
times, has made a difference: Pizzas are now ready to help those experts sitting in the call center
for pickup in about three minutes or delivered within solve your problem as quickly as possible using
10 minutes. all the information that our company has at its
disposal, in a way that is way more scalable than
one human brain,” says Heather Nolis, senior
software engineer at T-Mobile.

A STRATEGIC PLAYBOOK FOR DATA, ANALYTICS, AND MACHINE LEARNING

9 u
t
Keys to scaling machine learning
Machine learning pilots are plentiful,
but the challenge for many organizations is full-scale
deployment. Here are five steps enterprises should
consider to quickly and easily build, train, and deploy
Develop (and augment) your team. Highly spe-
cialized data scientists are critical, but you also need
analysts who can translate business problems into ML
solutions. With talent hard to find, companies need to
“
Working very closely
with the people
who are going to
machine learning at scale: balance recruitment with upskilling existing employees.

Enlist executive buy-in from the business from the

T-Mobile, for example, has had success finding talent
consume what you
in unexpected places—including a marketing manager
start. Too many proof-of-concept (POC) projects fail
because they are started by small data science teams,
in the risk department who writes R code, and a Java make is the only
developer with a master’s degree in linguistics who
who don’t bring the business or executive teams in helped lead a speech transcription project. way you can build
until after the system has been designed, instead of
soliciting input at the onset of the project. A better “Working very closely with the people who are going to something that’s
process is to “work backwards,” where the teams agree consume what you make is the only way you can build
on a mutual set of business outcomes and design the something that’s good and makes sense,” says Nolis. good and makes
solution and approach to achieve them.
It’s also important to augment in-house expertise with sense.”
“This approach provides a clear perspective on how partners. This extends the community of experts you
you can go from Point A to Point B, helps you prioritize can tap into, while simultaneously offloading the “undif- — Heather Nolis,
use cases, and what you should think about near- ferentiated heavy lifting,” so in-house engineers can senior software engineer
term versus recommendations for the long run,” says focus on higher-value work. at T-Mobile
Elaprolu.

A STRATEGIC PLAYBOOK FOR DATA, ANALYTICS, AND MACHINE LEARNING

10 u
t
Make sure you have enough data. The right data—
and plenty of it—is critical to success when you move
beyond a POC, which by definition focuses on testing a
concept with a smaller set of data.

“In the real world, when you take a limited amount of

data and try to deploy it to production, you will run into
use cases and data sets that are not necessarily similar
to what you’ve dealt with in the POC,” says Elaprolu.

Create a process for continuous iteration. Conditions

and business environments frequently change in the
real world, so machine learning can’t be a “set it and
forget it” exercise.

“You need to have that continuous loop of retraining

and redeploying ML models,” says Elaprolu.

Offload undifferentiated heavy lifting. Instead of INNOVATION SPOTLIGHT

expending development resources to build capabilities
from scratch, partner with a cloud services provider that Formula 1
can deliver core ML capabilities so your internal team
At Formula 1, machine learning is front and “There’s a lot of complexity and massive physics
can focus on higher-value activities.
center in improving the experience for racing going on in the background,” he says. “But the
For example, AWS AI helps developers augment fans. Formula 1’s engineering teams have built stuff that pops out on the screen is simple and
applications with capabilities such as image and video machine learning models that help fans better engaging and brings about that excitement and
analysis, natural language processing, personalized understand the split-second decisions made by a that tangible entry point into all of these different
recommendations, and virtual assistants—with no driver or pit crew that can dramatically affect the stories that are going on as a race is unfolding.”
machine learning experience required. In addition, tools outcome of a race.
While the use of analytics in Formula 1 is unique,
like Amazon SageMaker simplify the process of build-
Machine learning dates back several years at the approach that Smedley’s team takes with
ing, training, and deploying models—and lower the bar-
Formula 1, where it was initially used to help data and machine learning is applicable across
rier to entry. Learning tools like AWS DeepRacer, AWS
racing teams better understand car performance, industries.
DeepLens, and AWS DeepComposer can help develop-
according to Rob Smedley, Director of Data
ers learn ML fundamentals. AWS programs including “I think in any business, your first port of call is
Systems, Formula 1, and AWS technical ambassa-
the Machine Learning Solutions Lab and AWS Machine to look at what your problems are,” Smedley
dor. Smedley’s team has since expanded its focus
Learning Embark deliver hands-on onboarding, training, says. “And once you understand your problems,
to fan experience, with many innovative applica-
and implementation support to jump-start the process. then you put in place a strategy for fixing those
tions that leverage analytics and ML, including
F1 Insights powered by AWS and the F1 ProAm problems.”
DeepRacer event.

A STRATEGIC PLAYBOOK FOR DATA, ANALYTICS, AND MACHINE LEARNING

11 u
t
 2 Move
to managed services

3 M
 odernize
1 Break free your data warehouse
from legacy databases 4 B
 uild
modern apps with
purpose-built databases

5 T
 urn
data into insights

5 steps to building a modern data foundation

Without an organizing framework, 3. M
 odernize your data warehouse. Traditional data
it’s easy to be overwhelmed by the opportunities and warehouses don’t have the ability to effectively store
challenges of building a modern, cloud-based data and analyze the growing volume and variety of data,
foundation. To help, AWS created the Data Flywheel, a which leads to data being stored in multiple silos. A
holistic framework that applies the self-reinforcing loop modern “lake house” approach, including a data lake Summary
principles set forth by Jim Collins to a data manage- that can store unlimited volumes of data in various
ment strategy designed to maximize the value of data. structured and unstructured formats, makes it much Data, analytics, and machine learning
easier to catalog data, make it accessible, and ana- have the potential to radically transform
The Data Flywheel outlines five fundamental steps to
lyze it across the business. business processes and revenue models
building a modern, cloud-based data foundation:
as well as shape future innovations. But
4. Build modern apps with purpose-built databases. data is only valuable if you can turn it
�1. Break free from legacy databases. Many organi-
Move from antiquated monolithic apps that run on into action.
zations still have legacy, proprietary databases that
one-size-fits-all relational databases to highly distrib-
are expensive, create lock-in, and carry punitive
uted microservice-based systems running on multiple To become a true data-driven organi-
licensing terms. Moving to open-source databases
purpose-built databases to solve each problem. This zation, leadership teams need to shift
can deliver cost efficiencies without causing a
method frees the application from having to employ culture to view data as a strategic asset.
performance or availability hit.
a single, overburdened database for every use case. A modern, cloud-based infrastructure
2. Move to managed services in the cloud. As provides the scale, flexibility, and intelli-
5. Turn data into insights. Data lakes, analytics, and gence to support this shift and empower
database platforms begin to scale up, IT time and
machine learning help organizations gather smart, your business.
administrative costs can grow as well. Cloud-based,
accurate insights faster and empower end users to see
managed database services reduce time spent on
and visualize their data from any device or application. For more information, click here
undifferentiated heavy lifting so teams can focus on
higher-value activities. The five steps are not linear, which gives organizations
flexibility depending on their current level of data profi-
ciency. “You can start anywhere, and they build on each
other,” says Oberoi.

A STRATEGIC PLAYBOOK FOR DATA, ANALYTICS, AND MACHINE LEARNING IDG Communications, Inc.
12 t