Machine Learning

Executive Briefing

v 3.0
 WELCOME

Instructor Introduction

Kirill Eremenko wants every person he meets to know the extraordinary potential of data. As
Director of Operations at BlueLife AI, he helps businesses succeed with Artificial Intelligence.
In his role as an online educator and CEO at SuperDataScience, Kirill has helped over 750,000
students master the topics of data science. He co-founded DataScienceGO to give leaders a
forum to discuss strategies for developing data and technology practices.

Machine learning will reshape the way you do business. Let Kirill show you how.

Machine Learning 2
Executive Briefing
 Purpose of the course & overview of learning outcomes

Amazon and Google rank among the world’s most successful corporations. They have reshaped
the way we do business, what we learn, how we read, and how we see the world. They have
the power to know what we are searching for, even if we can’t put it into words. They can tell us
what we want before we know it ourselves.

These prosperous organizations both use a special tool that contributes to their success. It
requires no significant investment of your time and money. It will not ask you to overhaul your
company’s organizational structure or way of working. It is something that you can implement
into your activities, and that will work alongside you and your team.

You might have already guessed what this powerful tool is: Machine Learning.

So many companies fall for the allure of this profitable discipline. But it’s not easy to know how –
or where – to start. Maybe you’ve already tried and failed to implement a data-driven project. You
might have advertised for a Machine Learning engineer, only to be underwhelmed by the results
that they’ve produced.

Implement it in the wrong way, and Machine Learning can lead to snags in your development.
This course will show you how to avoid these traps and instead develop a resilient strategy for
success.

Welcome to Executive Briefing: Machine Learning, a straightforward overview of Machine

Learning packed with real case studies and tutorials. These videos will take less than 90 minutes
of your time, and they could save you – or make you – a fortune.

Executive Briefing: Machine Learning will show you that there is nothing unapproachable about
the discipline. It will:

• ensure you can navigate technical terms and topics related to Machine Learning like
a pro,
• help you to develop a robust strategy for implementing the discipline into your
organization,
• highlight the tools needed to create a stellar Machine Learning project,
• explain the steps to take to build a reliable team of experts and support staff, and
• show you how Machine Learning can supercharge your company’s finances,
innovation and marketing strategies.

After this course, you will see how Machine Learning will quickly become your company’s most
valuable member of the team.

Machine Learning 3
Executive Briefing
 MODULE 1

What, Why & How of Machine Learning

1.1. Purpose of this section

What is Machine Learning? How can it benefit a business?

This section offers a concrete definition of the term and explains its relationship to Artificial
Intelligence and Deep Learning. You will also discover how Machine Learning can help you to:

• innovate with new products,

• market the right services to the right customers, and
• reduce overheads

Machine Learning 4
Executive Briefing
 1.2. What is Machine Learning?

Machine Learning teaches computers to make predictions based on data. The discipline,
therefore, (1) teaches machines and (2) makes predictions. For these two components to function,
you need to have data.

INSIGHTS

PREDICTIONS

Let’s look at how both components use data:

1. Learning & Teaching:

Humans learn through a combination of theory and practice. Computers are no different.
For a machine to understand the benefits and obstacles to your company’s success, you
must first feed it with past experiences. These experiences are units of information that
teach your machines trends and parallels derived from the data, which enable them to
forecast future change.

2. Predictions:
Once the algorithm has seen enough past observations to be able to identify patterns
confidently, you can make accurate predictions on new data. This is where the most
valuable insights are extracted to inform the key business decision makers.

Machine Learning is a discipline based within the

broader field of Artificial Intelligence.

The terms Artificial Intelligence, Deep Learning and Machine Learning are often used
interchangeably. They all refer to automated processes that rely on company data. So, how can
we differentiate between them?

The simplest way to understand how Deep Learning and Artificial Intelligence correlate to our
discipline of interest is to consider each term as one part of a Russian doll. Artificial Intelligence
is the biggest of them all, and therefore the outermost shell. Machine Learning sits inside of it,
and Deep Learning is packed inside both as the smallest ‘doll’. Logically speaking, all Machine
Learning is a type of Artificial Intelligence, but not all Artificial Intelligence is Machine Learning.

Machine Learning 5
Executive Briefing
 DEEP
LEARNING

MACHINE
LEARNING

ARTIFICIAL
LEARNING

Machine Learning has been around for a long time – it is not a new concept. So, why haven’t you
heard of it before now? Because technology needed time to catch up with the idea. Computers
have only recently become fast enough to crunch numbers at the speeds we need.

Think about how long it took you to download films from the Internet in the late 90s, and how
long it takes you now. That’s a world of difference. Towards the end of the 20th century, our
machines either couldn’t manage large datasets or took an agonising length of time to deliver
results. Now, you can throw numbers at a computer, and an algorithm will sort through the
information, all at the touch of a button.

This is why Machine Learning is so important now – because this is the first time in history when
we can work with significant amounts of information within a short timeframe.

Machine Learning 6
Executive Briefing
 1.3. Why do we do Machine Learning?

In the words of the father of business consulting Peter Drucker, “the business enterprise has only
two functions: marketing and innovation. Marketing and innovation produce results; all the rest
are costs.” These functions, he explains, are conducive to what all businesses need: customers.

Both functions – innovation and marketing – can be improved through Machine Learning. As a
bonus, you can also use the discipline to knock down company costs.

Innovation

Google usually crops up as the perfect example of an organization that uses technology to
innovate. One example of this is its Machine Learning algorithm that detects spam emails and
expunges them from your inbox.

This function adds value to their service. Many of us don’t even check our spam inbox – half the
time we don’t even remember it’s there. But imagine how your inbox might look if it had no spam
filter. Google’s function is so useful that it simplifies our life and runs in the background so we can
get on with our day.

Marketing

Amazon uses a powerful recommendation engine based on Machine Learning. You will have
noticed it in the emails the company sends you about products similar to those you’ve already
bought, and in notifications about items frequently bought together with the product that takes
your interest.

By adding a Machine Learning algorithm to online marketing tactics, this little tool uplifts
Amazon’s sales and revenue by 25%. All the algorithm uses are purchase data to target similar
customers. For such a simple idea, Amazon has netted an enormous profit.

Costs

In 2017, investment bank JP Morgan found a way to reduce its overhead significantly. It
introduced a Machine Learning algorithm that enabled it to automate the process of
interpreting commercial loan agreements.

This move saved the bank 360,000 hours of legal practitioners’ work per year – the algorithm
could do the work of a human in seconds. Adding the Machine Learning function saved JP
Morgan both time and those hefty lawyers’ bills.

To help you deepen your knowledge of practical applications of Machine Learning, we have
BONUS
curated an additional 15 case studies of Machine Learning in innovation, marketing and costs. To
MATERIAL
get special access to this insider content, visit www.bluelife.ai/cases

Machine Learning 7
Executive Briefing
 PRACTICAL ACTIVITY 1.3

What are the most significant challenges for your organization right now? Think about
the problems that you want to solve for your customers. Consider the mechanisms that
bring operations grinding to a halt. Perhaps you have access to evaluations forms from
your employees – what, if any, are their recurring complaints? What opportunities do you
envision for your organization, that up until now have been out of reach?

Don’t hold back your pen. The more challenges you can identify, the better. Once you
have a list of at least five issues, enter them into the chart below.

Make sure to use the correct column! The Innovation column should contain problems
related to your company’s future activities. Marketing should only include issues that
concern branding and reaching out to customers. The final Costs column, following Peter
Drucker’s philosophy, is for everything else.

WHAT ARE MAJOR CHALLENGES IN YOUR ORGANIZATION?

INNOVATION MARKETING COSTS

Machine Learning 8
Executive Briefing
 WHAT ARE MAJOR CHALLENGES IN YOUR ORGANIZATION? (CONT’D)

INNOVATION MARKETING COSTS

10

Machine Learning 9
Executive Briefing
 MODULE 2

Machine Learning and Your Business

2.1. Purpose of this section

Strategic thinking is important. Too many executives want to dive straight in, but take a moment
to sit back and consider the big picture. Module 2 will help you to think through your strategy for
implementing Machine Learning into your organization.

Every business professional should recognise the value of strategic thinking – it is the best way
to ensure that you and your team are on the path to success. This module outlines the steps you
should take to develop a well thought-through Machine Learning strategy, to protect you against
project setbacks and snags.

Machine Learning 10
Executive Briefing
 2.2. Machine Learning Strategy

Your Machine Learning strategy must always consist of three components:

• Data
• Tools
• Team

Each one of these is essential to the running of your project. Think of them as cogs in a machine –
if one of them is missing, the others won’t turn.

First, you need data to plug into the tools. Then, the Machine Learning tools must be there to
execute analyses. Finally, you need a successful team by your side to ensure that you collect the
right data and that you have chosen the most suitable algorithm for the job.

TEAM

DATA

MACHINE
TOOL
LEARNING

Machine Learning exists at the intersection of these three components.

Think about the problem you want to solve.

An unclear problem will lead you on the road CLEAR

to failure. Think about a real challenge that GOAL
your company faces. What will its resolution
do – improve revenue? Increase customer
loyalty?

Machine Learning 11
Executive Briefing
 PRACTICAL ACTIVITY 2.2

Using the chart from Practical Activity (1.3), identify which of those challenges is the
most important to you and your company. Which one causes the most day-to-day stress
to your workers? Which one makes a dent in your profit?

Don’t worry at this stage if you are not yet sure how Machine Learning can resolve the
problem. Throughout the course, as you get to know more about Machine Learning, you’ll
come to understand its breadth of possibilities. You can return to this exercise at any time
and change your goal. For now, pick something to use as a case study, so that you have a
tangible example in front of you.

Now, answer the three questions below:

Which challenge have you identified?

How does your organization currently respond to the identified challenge?

Imagine your ideal outcome. What would your organization look like if you could
overcome this challenge?

Machine Learning 12
Executive Briefing
 2.3. Having your Data in Check

Data is generated all the time. It tracks stock on product lines, it notifies recipients when parcels
arrive, it logs customer footfall – there are so many events in our life that generate data. But how
can you capture it?

Every successful Machine Learning project requires an initial three stages related to data. Be
mindful of them when you’re developing your Machine Learning strategy. They are:

• Capture (through sensors, data points on Stage 1

your website, tracking devices on equipment)
CAPTURE THE DATA

• Store (in storage facilities such as SQL or

data lakes) Stage 2
STORE THE DATA
• Process & Analyse (with relevant Machine
Learning algorithms)
Stage 3
PROCESS & ANALYZE THE DATA

There are many potential pitfalls at this early point that are necessary for you to consider, to
ensure the smooth running of the project. Most problems that arise at the third stage are the result
of unresolved issues in steps one and two. They are:

Your data is a mess

MIS ING CORRUPT
Some of the most common
problems for companies that do
not have data science strategies in
place are that the information you
have stored is missing, corrupt, or DATA
just incorrect. Your team might need
time to find, gather, and clean your
prior data, for it to function along-
side your new units of information.
INC0RRECT BIASED

Machine Learning 13
Executive Briefing
 Your data only tells you what you think is true

Biased data is one of the most challenging issues that you might come across. If a mortgage
company is using a Machine Learning algorithm to replace its manual decision-making about
whether or not to offer a loan, it will use the company’s historical data to predict the outcome for
each new client. But the company’s prior data is biased. Although it contains information about
the loans that were and were not successfully repaid it does not offer any detail about the loans
which weren’t approved – but that might have been successful. Thus, the mortgage brokers
continue to make the same errors in judgment because the machine has learned the same bad
habits.

Your data is in the wrong place

Organizations might have grown in such a way that data is kept in different departments,
making them not easily accessible across the company. This disparity leads to data silos –
information archives that exist separate to the rest of the organization. Silos are a common
problem, and it can stop your project dead. Here are four possible reasons for silos:

1. Structural/functional:
A sales division might choose to store their recent months’ sales in one location for easy
access and to archive all previous months in another. This division makes it hard to
analyse all sales data together.

SALES DIVISION
RECENT SALES PAST SALES

2. Political: Data is power, and many managers know it. Some departments become data
hogs, holding onto their information and unwilling to relinquish it to other staff.

3. Growth: Management changes, new tools enter the VENDORS

market, and operational strategies get updated. These
upheavals can lead to disparate, unmanaged data.

4. Software vendor lock-in: This is potentially the

most problematic issue. When you use external
services to log your information, you leave your data
at the mercy of that company. They also understand
that data is power, and many will keep your data in
their cloud, meaning access to it might be
YOUR DATA
temperamental.

Machine Learning 14
Executive Briefing
 Consider each of these issues when building your strategy. Leaving one of these areas
unresolved may cause roadblocks to your project.

Finally, ensure that you leave enough time for quality assurance. Successful Machine Learning
projects take time. As you can see from this section, data preparation, processing, and
compilation take a significant chunk (roughly 80%) of project time. A good rule of thumb is to
quadruple the time requested by your team. If your data scientist or Machine Learning engineer
tells you that a model will take two weeks to build, multiply that amount of time by four to get the
amount of time required for data preparation and quality assurance, and you should expect your
project to be complete in a total of 10 weeks.

PRACTICAL ACTIVITY 2.3

To see if Machine Learning can solve a particular problem, we first need to establish if the
required data is free from any of the issues outlined above. Consider the challenge you
selected in Practical Activity (2.2) and identify:

• the data that might be required to resolve it,

• if there are any possible issues with it, and
• (if ‘yes’ to the above question), how those issues can be fixed.

It takes practice to know what type of data you need. The best way to identify required
information is to remember that any action, any event, generates data – the onus is
on your organization to capture it. Look at the processes that relate to your chosen
challenge. What steps are required to complete those processes? If you’re looking at
online purchases, consider that a transaction first needs a customer to put an item into
their virtual cart. If you want to track employee attendance, remember that workers have
to leave work by clocking out. If you are looking to make production more efficient, take
into account that a factory conveyer belt needs information about each of the points
where a new part is added to a product. Jot down the relevant actions for your selected
challenge.

Consider the problems concerning information that you already learned about, such as
incorrect data and data silos. Always double-check your historical data, or your analysis
will only return biased results.

How can these identified data issues be addressed? If data silos exist, how can they be
broken down? If you need to gather more data, can a budget be allocated for creating
systems to capture it? Can missing data be recovered, or can they be purchased from
external sources?

Use the chart below to record your findings.

Machine Learning 15
Executive Briefing
 IS YOUR DATA IN CHECK?

POSSIBLE ISSUES WITH THE HOW CAN THE ISSUES

REQUIRED DATA
DATA BE RESOLVED?
1

7
3

10

11

12

13

Machine Learning 16
Executive Briefing
 IS YOUR DATA IN CHECK? (CONT’D)

POSSIBLE ISSUES WITH HOW CAN THE ISSUES

REQUIRED DATA
THE DATA BE RESOLVED?
14

15

Machine Learning 17
Executive Briefing
 2.4. The Tools

There are lots of tools available to a data scientist – you might have heard of for-profit software
such as SAS, SPSS, Matlab. These will help you to carry out your Machine Learning experiments.

Many organizations prefer to go with commercial tools because it gives them a feeling of security.
What they’re creating is their intellectual property, and it cannot be easily shared or taken by
other people online.

But there is also a strong case for using open source tools and software.

You might wince at the idea. Open source. Won’t that leave my data accessible to hackers? Will
others be able to steal my information? If it’s free-to-use, surely it can’t be very good?

These are common misconceptions of what open source actually means. You might be surprised
to hear that Facebook and Google both use open source tools for their Machine Learning
practices and even their AI algorithms. These tech giants have plenty of money to buy software if
they wished to do so, but they use open source (specifically, Python). Why?

The value lies in your data, not the tool.

Don’t be misled by the thought of creating the world’s most cutting-edge Machine Learning tools.
Static software is not valuable – the value is sitting in your data. Thousands of researchers at
Google and Facebook are developing open source software and publishing them online. Google
published 454 research papers in 2018 alone – that’s more than one per day. So, why waste time
inventing the wheel?

You are not going to outfox Google in the tech game. Rather than tax your Machine Learning
team with developing software that will go obsolete in months, get them to read Google’s
research papers and consider how to apply their approaches to your business.

Open source tools allow a Machine Learning team to start

from an already developed base, to find solutions online, and
to source talent.

Machine Learning 18
Executive Briefing
 Another benefit to using open source software is that they have massive online networks.
Software that is not open source often doesn’t account for the speed at which technology
develops. With so many people across the world contributing to the development of tools, if your
Machine Learning team has any questions or needs guidance, online help is readily available. The
latter benefit also makes it easier to find talent, directly from the pool of people assisting you.

Two industry-preferred open source tools dominate Machine Learning: R and Python. My
recommendation is Python. This tool allows you not only to carry out Machine Learning but also
develop applications, and it facilitates the productisation of your algorithms. Google uses it for
TensorFlow; Facebook uses it for PyTorch. There are many manuals and resources for both these
tools, to help you in your search for the perfect solution.

Machine Learning 19
Executive Briefing
 2.5. Hiring a Machine Learning Team

When you’re hiring a team of people to help you with Machine Learning, avoid looking for the
‘unicorn’ who will magically solve your problems. Too many organizations want to hire someone
to fix all their problems without really thinking through the specifics of what they need. Career
sites are flooded with completely unrealistic job descriptions, looking to find candidates with
ten years’ experience, a PhD, and knowledge of 50 different tools and algorithms. These are red
flags to the best data scientists – they signal desperate managers who don’t understand what
Machine Learning does and how it can help to solve problems.

UNICORN

HAS NEVER BEEN FOUND

Select and structure your team the smart way.

Tech entrepreneur Lukas Biewald outlines two approaches, one for small companies, and another
for medium-to-large organizations. The former may have budget constraints. If this is the case
for you, look for a Machine Learning engineer who is passionate about your company’s mission.
When they start work, ensure they are not focused on multiple projects – instead, ask them to
prioritise the most critical initiatives. Concentrating on one project at a time will ensure they deliv-
er maximum value to that specific issue.

For medium-to-large enterprises where the budget is not such a constraint, seek to build a team
of Machine Learning engineers. Hire someone who already has experience in the problems you
want to solve. Then surround them with a team of support staff who can support them in pre-
paring the data. These could be analysts or business intelligence experts, or they could be people
already working in the company, who are passionate about the topic and want to learn more.
These support staff should assist the expert with individual components of the project’s lifecycle.

Machine Learning 20
Executive Briefing
 ORGANIZATION TYPE

SMALL MEDIUM/LARGE

IS BUDGET A CONSTRAINT?

YES NO

RECOMMENDATION:

One M.L. Engineer passionate about your One experienced or talented M.L. Engineer
mission

MANAGEMENT KEY:

Not focused on multiple projects Build a M.L. Support Team

Prioritize important projects one by one Team helps out on project lifecycle

Machine Learning 21
Executive Briefing
 2.6. Integrating Machine Learning in the Organization

It can be easy to get ahead of yourself. Blue sky thinking is a large part of AI and Machine
Learning. But when you’re just starting, you also need to keep your feet firmly on the ground.
Prioritise your projects, and focus on specific problems and applications for them.

One useful point of consideration that you should have at this stage is where your team will sit in
the company. Which department will be the ‘parent’ for your team – marketing? IT? Operations?
Sales? There are two approaches for an executive to take. Both methods have their merits, and it
is essential to consider both before settling on one:

1. Create a separate division:

For large companies, many Machine Learning teams begin in IT and eventually branch
out into a division in their own right. When this happens to your company, head up this
division with a Chief Data Scientist. Ask them to sit in on executive meetings and advise
how and where their team can add value to the organization. Enabling the division’s
head to sit in on board meetings will improve communication across departments and
prevent data siloing by giving your Data Science team authority. The only limit here is
the cost of developing a Machine Learning division, but if your scientists are working to
improve company operations, you should consider appraising this value against the staff
costs.

M.L. TEAM

2. Install experts into relevant departments:

Rather than building a standalone Machine Learning team, you could consider dropping
an engineer into every division that could utilise Machine Learning. This approach
enables each scientist to help their target division, not to mention having an expert in
each department will facilitate cross-company communication related to information
archives. One disadvantage of this approach is a lack of assistance. Without support
staff, your experts might become overwhelmed with their projects.

Machine Learning 22
Executive Briefing
 M.L. EXPERT M.L. EXPERT

M.L. EXPERT M.L. EXPERT M.L. EXPERT

M.L. EXPERT M.L. EXPERT

PRACTICAL ACTIVITY 2.6

Which of the outlined approaches will you take? Using the table below, list the benefits
of both options for your organization. If your organization is small and for the time being
you are looking to hire only one Machine Learning expert, then perform this exercise for
a future point in your business. What will your company look like in 5-10 years? How
will it change, and how could a Machine Learning team support or even bring about that
change?

WHICH APPROACH?

BENEFITS OF A STAND-ALONE BENEFITS OF M.L. EXPERTS

MACHINE LEARNING TEAM INTEGRATED WITHIN DIVISIONS
1

Machine Learning 23
Executive Briefing
 WHICH APPROACH? (CONT’D)

BENEFITS OF A STAND-ALONE BENEFITS OF M.L. EXPERTS

MACHINE LEARNING TEAM INTEGRATED WITHIN DIVISIONS
7

10

11

12

13
3

14

15

Machine Learning 24
Executive Briefing
 2.7. Recap

In this unit, you learned that a Machine Learning strategy comprises three components: data,
tools, and team. Identify the questions that you want your Machine Learning team to answer.
Rank them in order of importance. Ensure your team does not disperse its efforts but instead
focuses on one project.

1. Data:
Every business generates data. You need to find a way to collect, store and process it.

2. Tools:
Try out open source software such as R and Python, and leverage these publicly
available tools to gain momentum in your Machine Learning practice.

3. Team:
Build a Machine Learning team that fits your type of organization and the problems that
you want to solve.

SMALL MEDIUM LARGE

ENTERPRISES ENTERPRISES ENTERPRISES

Machine Learning 25
Executive Briefing
 MODULE 3

Technical Case Studies

3.1. Purpose of this section

This module provides real-life business examples of Machine Learning. You won’t find a single
equation here – everything you learn in Module 3 will be from an executive viewpoint.

You will come away from this module with a clear idea of what each algorithm does, and which
Machine Learning algorithms could benefit your business’ individual challenges.

Machine Learning 26
Executive Briefing
 3.2. Regression Models: Intuition + Case Study

Regression is a branch of Machine Learning, and it is used to predict continuous variables: such
values as revenue, expenses and mileage. Continuous variables are numeric variables that have
an infinite number of values in a range between any two values. A value for mileage, for example,
could be anywhere from 5 to 160 mph: it could be 15.2 mph, or 15.21 mph, or 15.217 mph, and
so on.

NOTE Not all numeric variables are continuous. For instance, ‘Birth Year’, while numeric, is actually a categorical variable. We will
touch more on categorical variables in further tutorials.

The Problem

You are the CEO of a luxury travel company and you want to know if the happiness of your
customers affects your revenue.

The Data

You will need to gather your data for revenue and customer satisfaction. Each point on the plot
represents a calendar month and customer satisfaction score, as well as revenue for that month.

The Terms

Revenue is the most important factor for this study because it is the value that you want
to examine. For this reason, it is known as the dependent variable. Customer satisfaction is
therefore the independent variable, because its change is not affected by your revenue.

The Analysis

Let’s say you have been collecting relevant data for many months. By entering this information
into an algorithm for regression, you might see a trend emerge (each point represents a separate
month). This is known as a simple linear regression, and you may remember it from high school.

REVENUE

Linear
Regression

CUSTOMER SATISFACTION

Machine Learning 27
Executive Briefing
 The result in the image is clear: The more the customers are satisfied, the higher the revenue of
the business will be.

Regressions will not always form straight lines – sometimes, the data will show something more
scattered. These are called polynomial regressions.

This is a very basic example of regression, which plots one variable against another. But
regressions don’t have to be restricted to two variables. You could input several variables, if you
feel that they are relevant to answering your project’s question. Using the above example, there
may be factors that affect your revenue additional to customer satisfaction, such as the weather,
or the revenue of your competitor. In these cases, where you use more than one independent
variable, you need a multiple linear regression.

REVENUE

Polynomial
Regression

CUSTOMER SATISFACTION

Machine Learning 28
Executive Briefing
 PRACTICAL ACTIVITY 3.2

The best way to commit something to memory is through practice. Think of a continuous
numeric variable in your organization that you might be interested in analysing with
simple linear regression. This will be the dependent variable for the purposes of this
exercise. Next, think of several independent variables that could relevant to the variable
you identified, for a multiple linear regression.

Brainstorm your thoughts using the chart below

REGRESSION IN YOUR BUSINESS

DEPENDENT VARIABLE INDEPENDENT VARIABLES

Machine Learning 29
Executive Briefing
 3.3. Classification Models: Intuition + Case Study

Classification algorithms group together similar data. Here’s an example of how they might work
for your company:

The Problem

You’re the marketing director of an ecommerce company and you want to analyse which
customers are likely to churn.

The Terms

Here, ‘churn’ is the dependent variable and a categorical variable, the latter because it either
has a value of 1 or 0. It tells us if a customer is likely to leave to a competitor (1) or not (0) within
six months of joining. Your independent variables for this experiment are ‘Age’ and ‘Time on
webpage’ (time spent on your website per month, on average).

NOTE More broadly speaking, categorical variables have one of a set – usually fixed – number of values. These values assign each unit
of information to a group. Categorical variables are the basis of classification because, as the name suggests, you want to classify
your observations into groups.

The Data

Every point represents an existing and/or past customer, their age and the average time they
spend on your website per month. The colour of each point represents your categorical variable
of churn (whether or not a customer left within six months of joining). Since you are training your
algorithm with past data, you can extract this information from your database.

The Analysis

To see if a new customer is at risk of churning within the next six months, you need to first collect
data for their age and average time per month they’ve been spending on your website since
joining, and then plot them onto this historical data.

TIME ON
PAGE

LESS LIKELY
TO CHURN

New
Customer

AGE

Machine Learning 30
Executive Briefing
 In the scatterplot, there is a clearly defined boundary between loyalty and churn. Depending on
which side of the boundary the customer under investigation falls, he or she may or may not be
at risk of leaving. If they are in the ‘remain’ group (on the right) they are less likely to churn in the
first 6 moths. If a different customer were to end up on the other side, then they are at risk and
you could take action (such as notifying your marketing department) to make sure they stay.

This is a trivial example of a basic problem. In this 2D diagram, it is easy to see where the groups
lie. You might think that you don’t need a Machine Learning algorithm to analyse this. However, in
most cases the analysis would involve thousands of customer data points and multiple variables.
It would simply be impossible to plot the results in a 2D diagram. That’s where Machine Learning
comes to the rescue – the algorithm will calculate everything in the background and deliver
actionable insights.

PRACTICAL ACTIVITY 3.3

As with Practical Activity (3.2), think of a categorical variable in your business and up to
five variables that could count as predictors for it. Use the table below to note down your
thoughts. These insights can be used to inform your future Machine Learning projects.

CLASSIFICATION IN YOUR BUSINESS

CATEGORICAL VARIABLE PREDICTOR VARIABLES

Machine Learning 31
Executive Briefing
 3.4. Clustering Models: Intuition + Case Study

Clusters reveal patterns. In the earlier example on classification, you looked for known groups
– customer loyalty vs customer churn. Clustering is similar to classification in the sense that it is
also designed to identify groups within your data. The difference is that, with clustering, you don’t
yet know what you’re looking for. This is called unsupervised learning. Clustering finds groups
for you, but it is your job to interpret why these patterns emerge and how they’re relevant to your
business.

The Problem

You’re the operations director of a mining company, and you want to explore if you can optimise
the servicing needed for your hauling trucks.

The Data

You would need to collect data that you feel is relevant to the organization. For the purpose of
this exercise, either due to your domain knowledge or based on intuition, you have chosen to
analyse data for hauled tonnage and travelled mileage of the trucks.

The Terms

Supervised learning refers to algorithms where you know what categories to expect prior to
the analysis. Classification is an example of supervised learning. Unsupervised learning is
the opposite, where the algorithm returns groups previously unknown to you. Clustering is an
example of unsupervised learning.

TONNAGE

MILEAGE

The Analysis

By plotting these data points on a chart, two clusters emerge. How you choose to interpret
these groups is a question for your team. What do you know about mileage and tonnage and
their effect on hauling trucks? What could be the reason for these clusters? Perhaps the group
containing trucks with high tonnage and low mileage require additional servicing. Use your
domain knowledge to answer this question.

Machine Learning 32
Executive Briefing
 NOTE Once the clusters have been identified and explained, they can be used in a classification model similar to the one in the previous
tutorial. Any new data point (in this instance, a truck) can be analysed across the same parameters (‘Mileage’ and ‘Tonnage’) by
the algorithm, to see which group it falls into. This will then enable you to make appropriate business decisions about that truck.

PRACTICAL ACTIVITY 3.4

Clustering can reveal patterns in your data, which makes it one of the most exciting
branches of Machine Learning. But this also requires your expertise, to interpret the
results.

Using your domain knowledge and business experience, list five variables that might be
of interest to your Machine Learning team when they are searching for hidden clusters
using the organization’s data.

CLUSTERING IN YOUR BUSINESS

VARIABLES

Machine Learning 33
Executive Briefing
 3.5. Association Rule Learning Models: Intuition + Case Study

Associations between different types of

data can give you a deeper understanding
of how customers, processes, products,
or even ads are all likely to behave in the
future. You might have heard the infa-
mous example that the sales of diapers
and beer are strongly correlated. Whether
or not this is true, this example serves a
point, that people tend to make purchas-
es of similar combinations. Crackers, for
example, might be bought in combination
with humous or another dip. Electrical ap-
pliances are likely to be bought together
with batteries.

These trends can be found in any marketplace, from Walmart to Netflix. Organizations make
smart use of association rule learning to better serve their customers and tailor their sales and
marketing. If you watch Stranger Things on Netflix, the platform might later recommend to you a
Stephen King movie, even though you’ve never seen or considered watching one. This is because
Netflix accumulates prior data from other users and the content tags regarding the series to
associate like products.

This is why you may get the impression that conglomerates know you better than you know
yourself. Machine Learning, especially association rule learning, can be used to derive benefits for
customers. Extract these associations with Machine Learning and discover the hidden potential
of recommendations.

PRACTICAL ACTIVITY 3.5

Using the space below, record any associative components that you can find for your
business. What do your customers frequently buy together? Which of your services are
reliant on each other?

Machine Learning 34
Executive Briefing
 3.6. Deep Learning

Deep Learning is a subdiscipline of Machine Learning. Within a Deep Learning tool, a deep neural
network ‘mimics’ the human brain, to learn processes and categories on its own.

Deep Learning has been around since the 1980s. Over this time, its theoretical concepts have
been subject to rigorous attention and adjustment. Now, we can input thousands of data units
and obtain insights within minutes.

What is so attractive about Deep Learning, as opposed to other branches of Machine Learning, is
that it has no predefined framework. There are no groups or structures to fit onto our data. What
we do have, however, is a neural network – a deep neural network comprising multiple layers.

A neural network is an artificial construct designed to mimic

how the human brain works.

A Deep Learning model will learn about your dataset in the same manner as a baby learns a
language. The child has a brain, but it has no prior understanding of vocabulary or the rules of
grammar. It has no predefined framework. By interacting with its parents, it will over time learn to
speak.

A Deep Learning neural network uses the same principles. We feed it the necessary information,
so that it can learn. Let’s say you wanted a machine to distinguish between dogs and cats. By
uploading thousands of labeled photos of these two animals, you can teach a neural network to
distinguish between them. The more photos you have, the more accurate it will become. Again,
the principles are similar to the way that humans process information. The more we know about
an animal’s distinguishing characteristics, the more likely we are to correctly identify it.

HUMAN BRAIN

Machine Learning 35
Executive Briefing
 Although Deep Learning algorithms will increase in accuracy over time, the drawback is that
a great deal of data is needed for it to be effective – far more than other Machine Learning
algorithms. The benefits, however, can be game-changing. A Machine Learning tool that tries to
recognise handwritten text might get an 87% accuracy rate today. With a Deep Learning tool
that has enough data, it’s entirely possible to get 99% accuracy – it’s that powerful.

DEEP LEARNING

PRACTICAL ACTIVITY 3.6

Having lots of data is one of the prerequisites for Deep Learning. Which area of your
business has a surplus of data? Anything to do with rich media (audio/image/video) is a
particularly good option. Write down three potential areas of your business in the table
below.

DEEP LEARNING IN YOUR BUSINESS

AREA OF BUSINESS

Machine Learning 36
Executive Briefing
 MODULE 4

Conclusion

4.1. Review Learning Outcomes

This course advocates taking the time to develop a Machine Learning strategy,
the necessary components of which include data, tools, and a team, to ensure that its
implementation will add value to your business. To recap the process:

• It is essential to identify a question before jumping into a project, as Module 2

highlights. You cannot run a Machine Learning project if there is no data, and
understanding what your question is at the beginning of the process will help you to
identify the data you need to collect.
• Hire the right people in the right number, depending on the size of your business
and what it hopes to do with Machine Learning. Use Module 2 to aid you in making this
decision.
• Select an open source tool such as R or Python to kick-start your first Machine
Learning project. Choose among the common algorithms of Machine Learning –Module
3 tells you which one to use for which project.
.

Machine Learning 37
Executive Briefing
 BONUS

5.1. Next Steps

Now that you’ve completed Executive Briefing: Machine Learning, what can you do next?

Why not get together a team in your company and educate them about Machine Learning as
well? Our comprehensive Machine Learning A-Z course contains 40+ hours of content, 300+
hours of tutorials, and it runs the gamut from R to Python across all branches of the discipline
from regression, association rule learning, clustering and classification, to natural language pro-
cessing and deep learning. This course is the quickest and easiest way to get everyone to under-
stand the business and practice of Machine Learning, and there are plenty of frameworks and
templates for your team to download and use. You’ll even convert the technophobes.

Data scientists are lifelong learners. If you plan to further your education about the dynamic
subject of Machine Learning, as well as Artificial Intelligence, Deep Learning, and countless other
industry-disruptive topics, join our Data Driven Executives forum. We are a growing interna-
tional network of company executives, and we get together several times a year to discuss cut-
ting-edge technologies and how they apply to all areas of business. We are open and welcoming
to new members, and we love to help those who are just starting out in Machine Learning. We
look forward to networking with you and forging lasting, productive bonds.
.

Machine Learning 38
Executive Briefing