The Rise of Machine Learning in Marketing: Goal, Process, and Benefit of AI-Driven Marketing

Abstract

This research report describes the goal, process, and benefit of AI-driven marketing. In particular, it explores how marketing leverages machine learning models to automate, optimize, and augment the transformational process of data into actions and interactions with the scope of predicting behaviors, anticipating needs, and hyper-personalizing messages.

Full text

RESEARCH REPORT
THE RISE OF MACHINE
LEARNING IN MARKETING
Goal, process, and benefit of
AI-Driven Marketing
May 7, 2019
by Alex Mari
endorsed by

TABLE OF CONTENTS
2 EXECUTIVE SUMMARY
3 DIFFUSION OF MACHINE LEARNING
Inside marketing technologies
Across marketing functional areas
Along consumer journeys
5 GOAL OF ML FOR MARKETING
Automation
Optimization
Augmentation
8 PROCESS OF ML FOR MARKETING
Data
Action
Interaction
10 BENEFIT OF ML FOR MARKETING
Prediction
Anticipation
Hyper-personalization
1 2 AI-DRIVEN MARKETING MODEL
14 METHODOLOGY
Author
Acknowledgments
Endnotes

EXECUTIVE SUMMARY
Machine Learning (ML) is developing under the great promise that marketing can now be both more
efficient and human. Cognitive systems, embedded or not into marketing software, are powering every
single functional area of marketing and each step of the consumer journey.
AI-driven marketing leverages models to automate, optimize, and augment the transformational process of
data into actions and interactions with the scope of predicting behaviors, anticipating needs, and hyper-
personalizing messages.
Modern marketers utilize user data to deliver hyper-individualized and hyper-contextualized brand
communications, in which each subsequent message builds on previous customer interactions. These
interactions are seen not as a final stage of a consumer journey, but as a way to orchestrate future
experiences in a satisfactory virtuous cycle.
Successful ML-powered companies turn data into seamless interactions with consumers using semi-
automated and real-time processes. These predictive and augmented experiences build deeper one-to-
one relationships with consumers, improve omni-channel customer experience, and drive product
differentiation.
Designing an AI strategy requires managers to systematically evaluate marketing needs in terms of
automation, optimization, and augmentation in relation to the searched benefits of prediction, anticipation,
and personalization. Balancing machine-inspired goals with expected benefits forces managers to
strategically assess their organization to redesign roles and responsibilities while adequately defining the
division of tasks between humans and machines.
This report, developed with the support of 30+ international experts, lays out a model for the definition of
AI-driven strategies within the marketing context. And, it explores the critical elements of what, how, and
why to infuse AI into the sequential steps of a marketing process.
Asking a manager “are you using AI?” in a few years from now will be
like asking “are you using the computer?”
Jim Sterne, Director Emeritus Digital Analytics Association
2

DIFFUSION OF MACHINE LEARNING
Marketing professionals face increasing complexity due
to the explosion of digital and data touchpoints, as well
as unprecedented consumers’ expectations in terms of
interaction, content, and offer personalization. Such a
degree of complexity is driving the adoption of a large
variety of marketing software that marketers require to
turn the vast array of historical data into actionable
insights. Scott Brinker, VP Platform Ecosystem, HubSpot
said, “AI algorithms and technologies are going to be
deeply embedded at every layer of what the marketing
software is.” Definitely, machine learning will leave no
area of marketing untouched. Gianluca Ruggiero, CEO
of Massive, agreed that “We must see AI as an all-
encompassing technology that is applied to every single
field of marketing.” At the same time, machine learning
models are being used to power and shape every step
along the consumer journey.
Inside marketing technologies
According to Gartner, marketing technology represents
the highest marketing expense, even above labor cost,
accounting for 29% of the CMO budget1. The
excitement around artificial intelligence has leaped into
the field of marketing technology (MarTech), with all the
major software providers claiming their ability to make
marketing smarter and faster, but, most of all, more
relevant. More than 7000 software applications
empower every aspect of a consumer journey and
encourage seamless execution2. Besides the number of
solutions, marketing technology is also growing in its
level of sophistication as intelligent algorithms are
becoming core to these services.
In addition to the so-called “AI-first” software, which
natively incorporates cognitive techniques such as
machine learning and deep learning, other “traditional”
enterprise solutions are infusing artificial intelligence as
part of their natural product development (see Sensei
for Adobe or Einstein for Salesforce). IBM’s Watson, for
instance, was initially created for various purposes but it
is increasingly leveraged to fulfill marketing goals. Other
standalone software applications follow a similar path
with players like MailChimp or Hootsuite upgrading
their services with smarter algorithms. Providers, such as
Shopify or Hubspot, have developed app marketplaces
with thousands of vendors offering perfectly integrated
solutions, from SEO to loyalty programs, some of which
incorporate ML models. As Claudio Crivelli, Director
Innovation & Transformation of Salesforce, said, “ML is
a layer embedded in any marketing technologies
offering huge potential to any organizations, making
employees more productive and customers happier.”
An emerging marketing paradigm sees the effect of AI
coming from two parallel exogenous and endogenous
forces (Figure 1). Companies receive a direct benefit
from proactively implementing an AI strategy that
defines how models, whether they are built internally,
bought from third-parties, or rent “as-a-service,”
respond to specific business needs. Firms may also
indirectly benefit from the use of marketing
technologies powered by AI components. In other
words, even if a company is not investing in building AI
capabilities, its marketing department might be already
benefiting from recent AI advancements deployed
throughout its marketing stack.
Figure 1. Marketing and AI
Technologies dependencies

Martin Fabini, Chief Technology Officer at ti&m, argued,
“Not having an AI marketing strategy in place does not
mean that you are not actively using ML. One can have
great benefits from standard applications although their
functioning represents a ‘black box’ for your
organization”. Clearly, marketing and AI technologies
are profoundly affecting each other and, combined, are
contributing to the rise of ML in Marketing.
Across functional marketing areas
AI is affecting every single functional area of digital
marketing, from social media to paid media. Different
functions show, however, different levels of both
technical sophistication and financial results from the
use of ML applications (Figure 2). Today, managers
perceive that infusing AI and ML in the field of paid
media, marketing analytics, and SEO (“performance
marketing") provide the highest return on investment.
Maurizio Miggiano, Head of Data & Analytics at
Mediacom, confirms that the most celebrated case in
ML for marketing is the so-called “programmatic media
buying,” which enables advertisers to reach consumers
with highly-targeted banners based on their past
behaviors. He said, “It is rather common to see a
double-digit optimization of the marketing budget
thanks to in-built algorithms – or even higher with
sophisticated custom algorithms.” Other use cases that
collocate AI in traditionally human-intense activities,
such as content creation, social media, and service
management, characterized by close relationships
between consumers and brands, are perceived to
provide a lower return on investment. According to
Andreina Mandelli, Professor at SDA Bocconi,
“Chatbots are the classic example of a ‘hype’ as they
often don’t deliver what they promise. This creates
disillusion in consumers which, in the short-term, might
go back to speaking with humans rather than
machines.” Being that these dynamics are fast
changing, several experts already predict a broader
usage of machine learning for creative and relational
activities in the near feature.
A BCG study commissioned by Google shows that,
within the same group of applications, the economic
results are extremely variable3. A test-and-control
experiment involving six large brands across Europe
showed the difference in online advertising performance
in terms of cost-per-action (CPA) when ML-driven
bidding, creative optimization, advanced targeting, and
attribution are preferred to the manual process. The
results highlight wide variation not only among single
activities, automated bidding vs. advanced audience
targeting, for instance, but also from campaign to
campaign. For example, audience targeting with
artificial intelligence resulted in a CPA drop, ranging
from a median of 8% to best practice of 27%, and
automated bidding reduced CPAs by a median of 9%
(good) to 44% (exceptional). In other words, studies
show positive results when using ML for performance
marketing. However, it is challenging to establish the
general effectiveness of these technologies as the ROI
largely varies by activity, company, and, ultimately, the
data mindset of the organization.
Figure 2. AI sophistication and
ROI across functional areas
4

Along consumer journeys
AI is infused at different stages of the consumer journey
and requires individual, yet coordinated
implementations. At Amazon, algorithms are shaping
the shopping experience at each step of the journey,
from “passive search” to “re-engagement”. Besides the
classic recommendation and re-targeting algorithms,
Amazon has introduced new ML-driven services such as
“programmatic sampling” to trigger consumer
awareness of new products (Figure 3). This model
enables companies to target a precise audience with a
free product sample. When this feature is used in
combination with retargeting advertising, the average
conversion to sale increases by 15%4.
Although we can expect incremental usage of
consumer-facing applications used simultaneously, like
in the case above, managers need to realize that human
intervention in each workflow largely varies depending
on the defined marketing goals. As Jim Sterne pointed
out, “You cannot write one algorithm that works for the
whole marketing process. Instead, you use a variety of
algorithms to create a variety of models that solve a
specific problem with a specific data set. You solve that
problem for a period of time, and then you need to
change it in order to get even more improvement.”
GOAL OF ML FOR MARKETING
AI enables marketers to amp up automation, optimize
processes, and augment workers in ways that make our
lives as employees, customers, and family members a
whole lot better. Marketing activities are partially
automated for routine tasks, optimized for nonroutine
functions, and augmented for complex decisions that
require employees and machines to build on each
other’s strengths. As managers become more familiar
with cognitive technologies, they increasingly
experiment with business solutions that combine
elements of AI automation, optimization, and
augmentation. According to Jim Sterne, “Definitely
automation, optimization, and augmentation are the
goals of AI. However, automation represents a key
priority for many managers. If there's something I can
automate, that's where I want to start. I need to make
my process as efficient as possible.”
Automation
Every business carries some inefficiencies that can be
replaced by high-performing algorithms. Companies
have traditionally automated activities to reduce the
costs of human labor. Today, marketing automation has
effectively become more than a cost-cutting mechanics,
with the hype shifting towards the automation of
customer experience. Companies have the opportunity
not only to automate internal processes invisible to
consumers, such as segmentation and targeting; they
can also deploy solutions that delight customers with
proactive and automated services.
Several companies are innovating by pushing the
boundaries of customer-facing automation. Vodafone
empowers users with AI-fueled self-service solutions.
Figure 3. AI-powered consumer journey
of Amazon.com (adapted from Hackernoon, 2018)

The chatbot “TOBi” provides personalized
recommendations with a +100% conversion rate
compared to the Vodafone website and answers key
requests with an 80% resolution rate5. Content
automation is increasingly responsible for the creation,
curation, and distribution of brand messages. Bol.com
is using Google’s automated bidding features in display
and video advertising6. The automated bid system
outperformed the manually optimized campaign with a
38% improvement in customer acquisition costs and an
estimated ten hours saved monthly per team member.
Similarly, automated recommendations use predictive
algorithms that account for an estimated 35% of
Amazon’s revenue7 and 80% of the movies watched on
Netflix8.
Because algorithms can automate ongoing decisions at
scale, ML represents a very potent tool for companies
seeking to interact with consumers efficiently. According
to Gartner9, by 2020 customers will manage 25% of
their relationship with an enterprise without interacting
with a human. Duplex, Google’s artificial intelligence
assistant, can independently handle service requests,
such as booking an appointment or managing personal
agendas. Federico Gobbi, Founding Partner at the AI
Marketing Association, noted: “With the deployment of
similar technologies, more customers will be
unknowingly talking to a machine.”
As the understanding of AI advances among managers,
these models become embedded in every business
activity, and the integration with other AI systems
increases the overall level of automation. Companies
like Amazon bring excellent examples of automated AI
strategy at the corporate level. Whenever the sales
forecasting system based on AI detects the growing
popularity of an item, several actions are triggered
automatically. Amazon updates the inventory forecast
and optimizes the supply chain system across
warehouses. As a consequence, the final users will see
more recommendations of the popular item at an
adjusted (and dynamic) price. Based on the outcome of
this “marketing campaign,” the sales forecast will be
updated again.
Optimization
Enterprises are using artificial intelligence algorithms to
optimize processes that reduce overhead, decrease
turnaround time, and improve output. Every marketer
can identify countless opportunities to infuse AI into the
brand building process to maximize practices of
consumer acquisition and retention. An AI-enabled
marketing software allows marketers to engage with
consumers across channels and provide optimized
customer journeys without a substantial increase in
manual work. In fact, by reducing the time spent on
segmenting customers and developing targeted
campaigns, cognitive technologies can dramatically
increase the productivity of entire marketing
departments.
I always say, 80% of the
work can be done by a computer.
The last percent must require human
intervention, and then you create something
extraordinary. I think we are entering a
beautiful world of human-machine interaction.
The machine does the heavy lifting on
identifying the salient elements, and then a
human mind puts the salient elements together
to create a masterpiece.
Dr. A. K. Pradeep, CEO at MachineVantage
An example of AI-optimized experience is offered by
Olay’s Skin Advisor, a deep learning powered app, that
analyzes a woman’s face to determine her “skin age”
and recommend the best product among hundreds of
different variations10. After the introduction of the
personalized Skin Advisor, Olay reported to having
doubled its conversion rate while engaging with 4
million consumers11. Uber predicts the rider’s
destination with over 50 percent accuracy and provides
context-aware suggestions that facilitate frictionless
experiences12. Lowe’s rolled out a retail service robot
called LoweBot to help customers by answering simple
questions in 70 languages while employees focus on
added-value services13. Because of its ability to
effectively navigate the store, LoweBot can scan the
shelves in search of incorrect prices, misplaced
products, and out-of-stock items.
For some marketers, optimization means to focus efforts
on strategic interactions with consumers while reducing
non-strategic ones. Phasing out emails in favor of real-
time customer-agent communications, like chatbots, is a
definite step in that direction. In the optimization
6

process, a machine might help employees to surface
unknown issues and optimize real-time interventions.
German Ramirez, Founding Partner of The Relevance
House, suggests that optimization will be a key driver to
reach a significantly more granular approach to
targeting. He said, “I have seen a fair amount of cat-
food advertising in my life and I do not own a cat or
plan to own one. Every single penny invested in having
me watching an ad for anything related to cat food was
wasted.”
Although AI algorithms are usually automated once in
operation, their development, installation, and training
remain highly technical, research-intensive, and human-
centric activity. Humans play a strategic role in the
ongoing fine-tuning of AI systems that lead to optimized
processes. In fact, AI is not a “set-and-forget”
technology as models are continuously tuned manually,
especially when natural language processing is involved.
No matter how sophisticated algorithms become, there
will always be the need for human-to-human
communication to supplement AI technology.
Augmentation
Algorithms can help teams that operate in a traditional
way to get more out of their marketing effort by adding
layers of intelligence. In some organizations, AI
augments rather than automates activities and
processes. Javier Guillo' Lopez, Digital Business
Development Watson at IBM said, “Augmentation is the
main goal and it's what all our technology is about. AI is
not about replacing people but enabling all marketers
to work better, faster and smarter.”
An effective explorative AI project begins with an
understanding of what human beings and machines do
well. Human beings can usually perform abstract
thinking outside a specific context better than machines.
The ability to deal with ambiguity, reframing research
questions, and applying common sense are skills that
machines are not expected to match, at least in the near
future. On the other hand, algorithms are faster and
more accurate in processing data and provide factual
solutions in a well-defined context.
An increasing number of organizations believe in the
coexistence of machines and humans. Capgemini found
that 86% of those managers implementing AI solutions
at scale firmly believe that machines can greatly
augment human output14. Overall, it is rare to observe
machine learning systems completely replacing human
jobs and processes. The declared business goal of most
companies is often not to reduce headcount but to
handle the explosion of customer interactions without
adding staff. In fact, the current best practice in
marketing organizations is to let AI fulfill basic and
repetitive tasks while employees work on more complex
customer solutions that require empathy and problem-
solving. “Just a few months ago, most of the discussion
around AI was around the concept of automation, but
now managers have come to realize that augmentation
is more powerful and more realistic,” said Thomas
Zweifel, Head of IT Consulting at AdNovum.
In most cases, machines enhance a human’s ability to
draw conclusions. Contact center operations are
adopting AI to streamline inquiry collection and
resolution. The latest AI-fueled platforms can extract
relevant pieces of information from both verbal and
textual conversations in real-time to swiftly capture
popular issues, suggest next best action to agents, and
predict the likelihood of a customer to churn.
Salesforce’s Einstein leverages rules-based and
predictive models to provide agents with contextual
recommendations and offers for customers. These “next
best actions” suggested to employees, such as “give
free shipping” or “offer zero percent financing” lead to
higher customer loyalty and upselling opportunities15.
As Niklas Kolster, CEO of Windsor.ai, noted, “Some
marketing activities are already fully automated like in
the case of recommendation engines, others are
generally optimized like bid management,
segmentation, and targeting, while others like ‘next best
actions’ are clearly augmenting managers.” Algorithms
can balance the amount of time agents spend on more
complex issues versus answering simple questions.
Examples include the service provider Botmind that
helps companies to deliver better customer experience
combining human and artificial intelligence through the
same live chat. Whenever the bot is facing new issues
that require conducting extensive unstructured
dialogues, they will immediately transfer the concern to
an individual. This hybrid process results in higher
consumer satisfaction and significant cost reduction.
Similarly, Userbot built a customer care bot, not with
the intention to bypass human-to-human interactions
entirely, but rather to advise agents on how to improve

their service performance. In these examples,
employees remain in charge while driving higher
efficiency thanks to the machine’s support.
PROCESS OF ML FOR MARKETING
AI contributes to automating, optimizing, and
augmenting three fundamental marketing processes:
data collection, insights gathering through data analysis,
and customer engagement. Modern marketing builds
on intelligence technologies to capture relevant user
data from the interactions with the brand. The benefit
for the user is better assistance on immediately
expressed needs and the anticipation of the
unexpressed ones, from a longer-term perspective.
The process of personalization is a continuous loop16
that offers companies the chance to engage consumers
one-to-one and to build self-reinforcing relationships17.
Companies continuously improve their personalization
processes through an iterative feedback loop, resulting
in the “virtuous cycle of personalization”18. However, Dr.
However, Dr. Christian Spindler, Founder and CEO of
Data Adead Analytics, warns that managers may have
different opinions on what type of personalization is
appropriate at each moment, which opens up for a big
debate. Generally speaking, personalization constitutes
an iterative process that can be defined by the three
stages of the understand-deliver-measure cycle. Rolf
Knöpfel, Migros Bank’s Chief Marketing and Innovation
Officer, said, “We can prove that this process works.
Our marketing spending efficiency for specific
campaigns doubled in the past year by simply shifting
to a data mining mindset and applying rule-based
algorithms.”
A feedback loop incrementally
produces higher personalization.
We all love ‘personalization’ and the
possibilities thanks to ML are limitless.
However, managers need to be ready to
address the question, ‘How much is
too much?’
Andrew McStay, Professor at Bangor University
An “AI-first” strategy needs to consider why and how to
strategically infuse AI in consumer-focused data
collection, promotional campaigns, and customer
interactions. The creation of unique brand experiences
requires marketers to turn any relevant user data into
action (or “campaign”). Every customer reaction to a
predefined campaign, like a commercial product launch
via email, produces a series of interactions on different
touchpoints. For instance, a call-to-action contained in
the message, “book your test drive,” will produce
interactions with users on social media, store, and
contact center. Every single interaction between the
brand and the user represents a new set of individual
data, explicitly or silently collected (see cookies). A
company must capture, structure, and analyze this
information to enrich the individual’s profile. This central
idea of data-driven marketing, captured in Figure 4,
coupled with ML makes the data-to-action-to-interaction
loop shorter and smarter. These three sequential steps
of AI-powered relationships are profoundly interlinked
and require a sound integration of technologies and
processes.
Figure 4. Designing
AI-Driven Experiences
(elaboration from Mari & Rohner, 2016)

According to Jim Sterne, “This figure exactly shows how
the machine learns. You create a model, it runs tests,
and it brings in results from interactions. It constantly
updates its data to action process. However, the rate of
improvement will stop, at some point, and the model
will stay at that plateau until you change something.”
Data
Cognitive technologies require the use of customer-
specific data in order to deliver unique brand
experiences. AI algorithms are not natively “intelligent;”
they learn inductively by analyzing data. Artificial
intelligence helps companies to automate, optimize, or
augment the process of data collection, analysis, and
storage. Sasha Srdanovic, Principal Solution Specialist
Data Platform and AI, said “AI can help to improve the
data quality by automatically screening and checking
data pools and databases. You can, for example, avoid
duplicates or help humans to consider different data
capturing scenarios.”
In the future, no company is expected to succeed
without making strategic use of the collected data.
However, there are still a few myths to bust when it
comes to data. The biggest data-related misconception
says that sophisticated ML models can provide valuable
business solutions even with insufficient data. AI is often
described as a data-hungry discipline because
meaningful data is a prerequisite for its exceptional
performance. “For a long time, data was not being
collected with AI and machine learning in mind. So,
usually, data quantity is not the main problem, but
rather the quality of data,” noted Erik Nygren, AI
Researcher at SBB.
Contrary to what marketers might think, data collection,
integration, and preparation are far more time-
consuming than building a machine learning model
itself. As Scott Brinker explained, “For AI to work, you
have got to have good data sets and clean data, and
now with things like GDPR it is not only good and clean
data, but it is also ‘compliance data’. So, there is a ‘big
data mission' many companies are still getting their
arms around.”
Another common belief says that once a machine-
learning model is automated, it continues to trigger
campaigns over time without human control. The so-
called “garbage in, garbage out” syndrome greatly
affects machine-learning algorithms. Since the
environment external to the model is dynamically
changing, key business users need to review the model
and provide new data sets regularly. This is the only way
for a self-learning system to avoid biased data that may
lead to dangerous outcomes.
Action
Data needs to be envisioned as a source of both action
and self-improvement. Algorithms can absorb live data,
process it, and then deliver real-time actions. Among
the others, artificial intelligence helps companies to
automate, optimize, or augment the process of scoring,
targeting, and campaigning. Maurizio Miggiano from
Mediacom said, “Marketers can apply ML to data in
order to reverse-engineer digital consumer journeys to
determine which tactics and strategies have previously
led to a positive outcome.” Delivering unique
communications at scale requires marketers to move
from classic segmentation to advanced techniques that,
powered by machine learning, leverage smart scoring to
build audiences and predict each customer’s likelihood
to convert. Peter Gassmann, Chief Consulting Officer of
AdNovum warned, “Having data scientists play around
with data may produce good insights but bringing these
into action requires a much broader organizational
effort.”
Turning the collected data into engaging campaigns
produces a positive effect on the overall customer
experience. Predictive campaigns require personal data
in return for a personalized shopping experience and
product recommendations in line with the user’s
expectations. On the one hand, more users understand
the importance of sharing personal information in the
value-exchange process. On the other hand, only 1% of
the gathered consumer data is reported to be analyzed,
i.e., to produce any insight or strategic action19.
If I am a small company selling
online $100 worth of products a
day, I can benefit from the AI system
Shopify ‘Kit’ that recommends ‘next best
actions’ even though I only have 50 daily
visitors. Kit looks at their behavior compared to
other behaviors across thousands of stores.
That's how the algorithm creates value for me.
Jim Sterne, Director Emeritus at DAA

The result of un-analyzed, underleveraged data does
not only represent a missed opportunity to deliver
modern experiences, but it also leads to consumer
dissatisfaction. This is why digitally mature organizations
build data infrastructures capable of gathering
consumer specific insights that generate unique
offerings. Dr. A. K. Pradeep, CEO at MachineVantage,
said, “The only way brands are going to survive is
creating a meaningful and rich consumer experience.
It’s in that creation of consumer experiences that ML
and AI come in.”
Interaction
Interactions are at the heart of meeting the elevated
expectations of today’s connected customers. Users
expect companies to provide intelligent, personal
experiences every time this is needed. In particular,
artificial intelligence helps companies to automate,
optimize, or augment the process of generating
meaningful interactions anytime and anywhere, while
delivering the best next product, content, and offer.
Predicting the evolution of an interaction on countless
channels requires managers to know precisely why,
when, and how to engage with consumers. AI-enabled
marketers are already counting on algorithms to drive
deeper engagement with customers along the value
chain. These managers have the opportunity to offer
unexpected and delightful experiences that consumers
did not have time to desire yet. “Those companies that
are managing this process with success have built real-
time operations,” said MarTech consultant Yusuf
Balcilar.
However, as Sven Blumenstiel, Chief Information Officer
of Sonova Group suggested, “There are limited
examples on how we use machine learning in the
action-to-interaction phase, and these are often not fully
automated.” Brands earn trust from customers every
time they successfully predict wishes and timely deliver
relevant offers. When AI supports interactions, the
ability to build trust in direct-to-consumer relationships
allow companies to be less dependent on centralized
voice platforms like Amazon Alexa or Google Home. In
this respect, Lorenzo Farronato, VP Marketing
Communications at Swarovski, argued, “My problem is
not necessarily to pay Alexa or similar platforms to reach
my consumers but that I do not have individual-level
data needed to develop the relationship with
consumers further.”
BENEFIT OF ML FOR MARKETING
The interest of marketers for machine learning is at least
threefold. In fact, the vast majority of the current use
cases can be classified based on the technology’s ability
to 1) predict consumer behaviors, 2) anticipate
consumer needs, and 3) hyper-personalize messages.
Sasha Srdanovic from Microsoft said, “ML in Marketing
is really about predicting what the next consumer move
might be and anticipating what their current and future
requirements are. The overall objective is having a
personal, individual, and seamless interaction with the
customer you are serving.”
Prediction, anticipation, and
hyper-personalization are an integral
part of what marketing is going to be.
Marketing managers will truly use the tools of
AI and machine learning to understand the
drivers of the non-conscious human mind which
is responsible for 95% of consumer behavior.
Dr. A. K. Pradeep, CEO at MachineVantage
Prediction
Predicting consumer behavior means providing the best
value proposition at the right stage of the consumer
journey. Marketers come to realize that traditional
marketing tools are unable to keep pace with the
velocity, variety, and volume of data. Machines can help
managers in reducing today’s level of complexity in
cross-channel customer engagement and make more
accurate consumer behavior predictions. As an
example, the German e-commerce merchant Otto uses
an AI model that predicts what will be sold within 30
days with 90% accuracy20. This system allows Otto to
automatically purchase more than 2 million items per
year from third-party brands while speeding up
deliveries to customers and reduce returns. “Being able
to find out the need of a customer before he realizes it
by himself is a huge advantage. As a marketer, you can
show potential customers all of your intelligence. This
translates into caring,” said Dalith Steiger-Gablinger,
Founding Partner of SwissCognitive.
10

Anticipation
Anticipating needs affects the offering of the best
products and services at the right price. Managers
exploit intelligent software with the intent of creating
consumer-centric and service-oriented organizations.
ML enhances the business objective of anticipating
consumer need, as required to differentiate services and
re-architect business models from the ground up. For
instance, Netflix21 develops original TV shows
analyzing creative elements of successful movies at a
granular level through the lenses of AI. This practice
doubled the success rate of original shows versus
traditional ones (from 40% to 80%). According to
Cosimo Accoto, Research Affiliate at MIT, “Digitally
mature organizations fulfill consumers needs by using
future consumer data (prediction) to anticipate
consumer behavior in the present time.”
Hyper-personalization
Message hyper-personalization is the delivery of
relevant messages at the right time and channel. AI
marketing enables the collection and analysis of data,
generation of insights, and definition of actions that
more effectively reach the individual.
According to Sasha Srdanovic from Microsoft, “Machine
learning helps to move away from former customer
segmentation and drive real-time automated
segmentation. We understand what the customer is
looking for right now and what he might be interested
in next.” Designing hyper-personalized experiences that
drive relevancy has become a key priority for most
organizations. L’Oréal Paris personalizes videos using
insights on interests and affinities of the audience, as
provided by Google’s AI-powered platforms. Recently,
L’Oréal created twelve versions of a YouTube video to
appeal to each specific segment. This campaign showed
an increase of 109% in brand interest and 30% in
purchase intent22.
AI connects customers in a whole
new way, and the most impressive
breakthroughs are at the dialog level
‘machine-to-human.’ However, while humans
might quickly lose trust in flawed algorithms,
many of us tend to trust machines more
if they have human features.
Claudio Crivelli, Director Innovation &
Transformation, Salesforce

AI-DRIVEN MARKETING MODEL
The conceptual framework presented in Figure 5
defines the critical steps of an AI-driven strategy
throughout its essential questions of what, how, and
why to infuse ML into the sequential steps of a
marketing process. Designing an AI strategy requires
managers to systematically evaluate marketing needs in
terms of automation, optimization, and augmentation in
relation to the searched benefits of prediction,
anticipation, and personalization. According to Cosimo
Accoto, “AI is not only a technology added to the
marketing technology stack. It is a starting point to re-
imagine the nature and the objectives of marketing.”
Seemingly managing relationships throughout a data-
to-action-to-interaction sequence enables companies to
predict behaviors, anticipate needs, and hyper-
personalize messages. Modern marketers utilize user
data to deliver hyper-individualized, -personalized, and -
contextualized brand communications in which each
subsequent message builds on the previous customer
interactions.
These interactions are seen not as a final stage of a
consumer journey, but as a way to orchestrate future
experiences in a satisfactory virtuous cycle. Furthermore,
successful ML-powered companies turn data into
seamless interactions with consumers, in a semi-
automated and real-time fashion. These predictive and
augmented experiences build deeper one-to-one
relationships with consumers, improve omni-channel
customer experience and drive product differentiation.
As Scott Brinker explained, “Overall, AI drives efficiency
and effectiveness of the marketing organization. The
advantage of automation, optimization, and
augmentation is productivity. You can have a small team
who, by leveraging this technology, can serve many
more people, much more quickly, at a much lower cost.
That's a very inward facing benefit, but it's a huge one.
It’s what a lot of companies are expecting from
cognitive systems.” A successful AI strategy can offer
sustainable efficiencies only when built on robust
technical (technology, data, process) and organizational
(people, capability, culture) foundations.
Companies that combine the power
of machines with employees who
possess the right skillsets, to analyze
the data and provide actionable
recommendations, will be able to
differentiate themselves and win customers.
Annamaria Fato, Global Senior Market Development
Manager, Zurich Insurance Company Ltd
Managers need to strategically assess their marketing
organization to redesign roles and responsibilities while
adequately defining the division of tasks between
humans and machines. In doing so, they are required to
strike a balance between the level of human and
machine effort injected into every relevant marketing
step and alongside consumer journeys. Sven
Blumenstiel suggested to, “get away from the mindset
that personal touch is needed everywhere or, on the
other hand, that complete marketing automation is the
Figure 5. AI-driven marketing model
12

ultimate goal. As a consumer, there are things that I
value very much doing myself and others for which I
require personal contact. Managers need to have a
differentiated and selective approach to strategy
design.”
Imagine for a moment the effect of fully automating
customer service, for instance using chatbots, in those
organizations driven by “customer obsession”, like
Zappos.com. Clearly, AI is not always the solution and,
in a future where more organizations become AI-
oriented, the human touch might still guarantee a more
sustainable competitive advantage.
At a strategic level, the ratio between human and
machine-mediated interactions, dynamically affected by
numerous internal and external events, allows
companies to create highly differentiated service
strategies. Like others, Scott Brinker wondered, “In the
future, are businesses going to differentiate themselves
based on the level of human touch they use?”
The human intervention in each workflow of data,
action, and interaction is sometimes close to 100% while
other times it is to zero, largely vary depending on the
defined marketing goals. Experts agree that a balance
between human and machine-driven activities is
required. Online advertising is a canonical example.
Human-driven optimization in terms of bidding,
audience, and budget adjustments can, in fact, add an
additional 15% campaign performance to the 20% AI-
driven improvement3.
In this respect, implementing an AI strategy is less about
developing algorithms and more about building
relationships that balance the strategic goals, processes,
and benefits of AI-driven marketing.
If you don’t care about building a
deep and honest relationship
with consumers, maybe you should
delegate that task to your AI.
German Ramirez, Founding Partner of
The Relevance House

METHODOLOGY
This report captures the insights and experiences of international experts, consultants, and AI-aware executives, as
well as, secondary research. Semi-structured in-depth interviews were conducted both face-to-face and online over a
period of 6 weeks until January 2019. Theoretical perspectives were not employed to facilitate the emergence of
insights. A total of 32 interviews were audio-taped. Transcriptions were analyzed adopting an inductive line-by-line
coding approach. Using NVivo 12, codes were grouped into themes and then re-evaluated to ensure that they reflect
data extracts. At the end of the coding process, 20 main nodes and 76 sub-nodes remained. Key conceptual nodes
were translated into a conceptual framework in Figure 5 that illustrates the strategic areas in the implementation
process of ML in marketing. My deepest gratitude to the following marketing, data, and IT experts:
Executives
Marketing, Digital, IT, Data – Switzerland
Annamaria Fato, Global Senior Market Development Manager, Zurich Insurance Company Ltd
Ciprian Corodan, Head of Digital Marketing Excellence, Sonova Group
Erik Nygren, AI Researcher, SBB
Lorenzo Farronato, VP Marketing Communications, Swarovski
Matthias Rohner, CRM Business Manager, Sonova Group
Maurizio Miggiano, Head of Data & Analytics, Mediacom
Rolf Knöpfel, Chief Marketing and Innovation Officer, Migros Bank
Sven Blumenstiel, Chief Information Officer, Sonova Group
Yusuf Balcilar, Head of Digital Platforms, Sonova Group
Consultants
Service Providers & Technology Vendors - Switzerland & International
Christian Spindler, Founder and CEO, Data Ahead Analytics
Claudio Crivelli, Director Innovation & Transformation, Salesforce
Darko Stanojevski, Senior Account Manager, Forrester
Gianluca Ruggiero, Chief Executive Officer, Massive
Javier Guillo' Lopez, Digital Business Development Watson, IBM
Martin Fabini, Chief Technology Officer, ti&m
Niklas Kolster, Chief Executive Officer, Windsor.ai
Pascal Wyss, Senior Innovation Consultant, ti&m
Peter Gassmann, Chief Consulting Officer, AdNovum
Sasha Srdanovic, Principal Solution Specialist Data Platform and AI, Microsoft
Thomas Zweifel, Head of IT Consulting, AdNovum
Experts
Scholars, Authors, Associations – International
Andreina Mandelli, Marketing Professor, SDA Bocconi, Author ”AI e Marketing”
Andrew McStay, Professor of Digital Life, Bangor University, Author “Emotional AI“
Andy Fitze, Founding Partner, SwissCognitive
Cosimo Accoto, Research Affiliate, MIT, Author “In Data Time & Tide“
Dalith Steiger-Gablinger, Founding Partner, SwissCognitive
Dr. A. K. Pradeep , MachineVantage, CEO, Author “AI for Marketing & Product Innovation“
Federico Gobbi, Founding Partner, AIMA (AI Marketing Association)
German Ramirez, Founding Partner, The Relevance House
Jim Sterne, Director Emeritus Digital Analytics Association, Author “AI for Marketing“
Martin Coul, Entrepreneur Technology Innovation, ETH
Scott Brinker, VP Platform Ecosystem, HubSpot, Author “Hacking Marketing”
14

a) About Alex Mari
Alex Mari is a Research Associate at the Chair for Marketing and Market Research at the University of Zurich where he
studies the impact of machine learning on consumer-brand relationships. He is the former Director of Digital
Marketing at Sonova Group, Brand Manager at Procter & Gamble, AKQA, and others. Alex founded and managed
two technology-driven startups. He is currently teaching digital & AI marketing in business schools and privately
advising companies. Alex holds an MSc in Marketing from the University of Lugano, Switzerland.
University of Zurich
Department of Business Administration
Chair for Marketing and Market Research
Andreasstrasse 15 - 8050 Zurich, Switzerland
+41 44 634 2918
alex.mari@business.uzh.ch
www.linkedin.com/in/alexmari
b) Acknowledgments
I would like to convey my gratitude to Prof. Dr. René Algesheimer, Prof. Dr. Anne Scherer of the University of Zurich,
SwissCognitive, Luca Barbati and Andrea Di Berardino, and all contacts who have made this report possible.
My special appreciation goes to Jim Sterne, Director Emeritus Digital Analytics Association and Author of “Artificial
Intelligence for Marketing: Practical Applications“ (Wiley, 2017) for his wisdom and inspiration.
This independent research report was 100% self-funded. This report is published under the principle of Open
Research and is intended to advance the industry at no cost.
The Creative Commons License is Attribution-Noncommercial Share Alike 4.0 International.
Photo credit: Daniel Kim on Unsplash.
c) Endnotes
1. Gartner “8 Top Findings in Gartner CMO Spend Survey,” Chris Pemberton, November 5, 2018
(https://www.gartner.com/smarterwithgartner/8-top-findings-in-gartner-cmo-spend-survey-2018-19/)
2. ChiefMarTec “Marketing Technology Landscape Supergraphic (2019),” Scott Brinker, April 4, 2019
(https://chiefmartec.com/2019/04/marketing-technology-landscape-supergraphic-2019/)
3. BCG “The dividends of digital marketing maturity,” Dominic Field, Shilpa Patel, and Henry Leon, February 18, 2019
(https://www.bcg.com/publications/2019/dividends-digital-marketing-maturity.aspx)
4. Hackernoon “A Map of Amazon and Modern Marketing,” David J. Carr, September 11, 2018 (https://hackernoon.com/a-map-of-
amazon-and-modern-marketing-372e1590b564ting)
5. Econsultancy, “Vodafone’s chatbot is delivering double the conversion rate of its website,” Ben Davis, October 11, 2018
(https://econsultancy.com/vodafones-chatbot-is-delivering-twice-the-conversion-rate-of-its-website/)
6. Think with Google “bol.com boost CPA 38% with automated bidding on Display & Video 360,” November 2018
(https://www.thinkwithgoogle.com/intl/en-154/insights-inspiration/case-studies/bolcom-boost-cpa-38-automated-bidding-display-
video-360/)

7. McKinsey & Company “How retailers can keep up with consumers,” Ian MacKenzie, Chris Meyer, and Steve Noble, October
2013 (https://www.mckinsey.com/industries/retail/our-insights/how-retailers-can-keep-up-with-consumers)
8. Inside Big Data “How Netflix Uses Big Data to Drive Success,” January 20, 2018 (https://insidebigdata.com/2018/01/20/netflix-
uses-big-data-drive-success/)
9. Gartner “Gartner Says 25 Percent of Customer Service Operations Will Use Virtual Customer Assistants by 2020,” February 19,
2018 (https://www.gartner.com/en/newsroom/press-releases/2018-02-19-gartner-says-25-percent-of-customer-service-operations-
will-use-virtual-customer-assistants-by-2020)
10. P&G “Olay Unveils Global Skin Analysis Platform Olay Skin Advisor – The First-Of-Its-Kind Application of Deep Learning in the
Beauty Industry,” February 27, 2017 (https://news.pg.com/press-release/pg-corporate-announcements/olay-unveils-global-skin-
analysis-platform-olay-skin-adviso)
11. VentureBeat “How Olay Used AI to Double its Conversion Rate,” Matt Marshall, July 19, 2018
(https://venturebeat.com/2018/07/19/how-olay-used-ai-to-double-its-conversion-rate/)
12. Uber Engineering “Engineering More Reliable Transportation with Machine Learning and AI at Uber,” Chintan Turakhia,
November 10, 2017 (https://eng.uber.com/machine-learning/)
13. Lowe’s Innovation Labs (http://www.lowesinnovationlabs.com/lowebot/)
14. Capgemini “Turning AI into concrete value: the successful implementers’ toolkit,” September 2017
(https://www.capgemini.com/consulting-de/wp-content/uploads/sites/32/2017/09/artificial-intelligence-report.pdf)
15. Salesforce “Salesforce Delivers the Next Generation of Service Cloud Einstein,” July 11, 2018
(https://investor.salesforce.com/about-us/investor/investor-news/investor-news-details/2018/Salesforce-Delivers-the-Next-
Generation-of-Service-Cloud-Einstein/default.aspx)
16. BCG “Profiting from Personalization,” Mark Abraham, May 8, 2017 (https://www.bcg.com/publications/2017/retail-marketing-
sales-profiting-personalization.aspx)
17. Vesanen, J., & Raulas, M. (2006). Building bridges for personalization: a process model for marketing. Journal of Interactive
Marketing, 20(1), 5-20.
18. Adomavicius, G., & Tuzhilin, A. (2005). Personalization technologies: a process-oriented perspective. Communications of the
ACM, 48(10), 83-90.
19. The Guardian “Study: less than 1% of the world's data is analysed, over 80% is unprotected,” John Burn-Murdoch, December
19, 2012 (https://www.theguardian.com/news/datablog/2012/dec/19/big-data-study-digital-universe-global-volume)
20. The Economist “How Germany’s Otto uses artificial intelligence,” April 12, 2017
(https://www.economist.com/business/2017/04/12/how-germanys-otto-uses-artificial-intelligence)
21. Forbes “Netflix Used Big Data To Identify The Movies That Are Too Scary To Finish,” Bernard Marr, April 18, 2018
(https://www.forbes.com/sites/bernardmarr/2018/04/18/netflix-used-big-data-to-identify-the-movies-that-are-too-scary-to-
finish/#2b343fdb3990)
22. Think with Google “L’Oréal dare to trust data as they bring the power of customisation to scalable branding campaigns,”
October 2018 (https://www.thinkwithgoogle.com/intl/en-154/insights-inspiration/case-studies/loreal-dare-trust-data-they-bring-
power-customisation-scalable-branding-campaigns/)
16