ARTIFICIAL INTELLIGENCE:
IMAGINATION AND REALITY
In our imaginations, artificial intelligence often
comes in blockbuster proportions: the computer’s
takeover in “2001, A Space Odyssey,” battles for
humanity’s fate with multi-limbed robots in
“The Matrix,” the meaning of love in “Her.”
In reality, AI is more pervasive, if less epic. AI
algorithms help IBM’s Deep Blue and Watson win
chess and “Jeopardy!” tournaments. Voice-activated
assistants give information, entertain, and control
devices and homes. Self-driving cars will go on sale
to car services next year, General Motors promises.1
AI, EXPLAINED
At its core, artificial intelligence processes specific
inputs then delivers specific outputs. Asked the
right questions and given the right instructions,
its algorithms will provide solutions and suggest
actions that produce desired results.
The problems addressed by AI are sometimes ones
that people could solve without intervention, on
a smaller scale. AI could, for example, match one
facial image to another, predict next week’s sales,
or say whether someone’s credit history shows
they’re likely to pay off a loan.
By breaking the problem into component parts,
each relevant input can be weighted and a solution
derived. Recognizing a face, for example, is a
combination of evaluating the curve of a jaw, the
angles of a nose, the color of someone’s eyes, and
so on, with some factors being more pertinent to
the solution than others.
“AI is like a spreadsheet on steroids,” says Sara
Robertson, VP of Product Engineering for Xaxis,
and an expert in artificial intelligence.
AI usually works in concert with machine learning
so that algorithms get increasingly better at
increasingly complex tasks like pattern recognition
and predictive analysis. Machine learning takes
the results produced from a round of instructions,
compares them to the predicted outcomes,
evaluates, then sends information back on how to
adjust and optimize.
Further enhancement comes from processes such
as “neural networks” — connected computing
nodes working together to increase processing
Artifcial intelligence (AI) has
changed what we can achieve in
media buying and planning, how
to achieve it, and the metrics
used to understand success.
This paper lays out what AI is,
strategies to apply it to
advertising, and the best
methods for gaining expertise,
evaluating partners, and working
with them to leverage the
opportunities afforded by AI to
achieve the best possible
business outcomes.
EXECUTIVE SUMMARY:
AI is a computer’s
ability to choose and
perform the right
machine learning
techniques at the right
time, successfully,
regularly, and with a
minimum of effort.power — and “deep learning,”
which helps refine the
understanding those neural
networks can produce.
AI algorithms can work
‘supervised’ — with people
indicating which results are
closer to the desired outcomes
— or ‘unsupervised,’ where they
execute and adjust on their own,
usually after a period of human
instruction (see right).
AI today is used in nearly every
industry in ways we access
every day. It helps search
engines find their targets;
executes complex decisions
for financial trading; powers
programming recommendations
for services like Netflix2. It can
also aid in content curation,
enhance cyber security,
improve warehouse inventory
management, assist salespeople
in generating better leads, and
help fly airplanes.
AI FOR ADVERTISING
For advertising, AI is being
used in a variety of ways to
improve effectiveness. It has
been used to find and define
audiences, refine creative
messaging (see graphic
page 5), generate audience
personas, and develop bidding
strategies that optimize for
clients’ stated goals.
“AI has many seemingly small
applications that currently
deliver digital marketing
efficiencies for companies
all over the world — from
advanced consumer targeting
and insights to highly
personalized ad experiences,”
says Adam Grow, SVP of Display
at Rakuten Marketing.3
At present, most advertisers aren’t taking advantage of the full
capabilities of AI. Instead, they deploy it to achieve simple goals. But
it can do much more than elevate discrete performance metrics. When
the many applications of AI are used in concert, they can add up to
a significant transformation in digital advertising strategy that drives
remarkably improved results.
To accomplish that, new campaigns must be conceived and built
around the unique opportunities and strategies afforded by AI. This
requires advertisers to shift their perspective; to refine and expand
their idea of marketing success and approach new campaigns in
cooperation with the way AI works.
The most powerful — and largely unfulfilled — potential of AI lies in the
bigger picture, in its ability to optimize towards business outcomes
rather than simple metrics.
One car maker, for example, sought to increase sales. To do so, it
evaluated the influence of individual performance metrics within
the context of that larger goal. It weighted factors such as website
interactions, brochure downloads, and showroom visits to determine
messages that ultimately lead to that outcome, and used artificial
intelligence with machine learning to continually optimize toward
better scores — thereby making its marketing evermore effective.4
vs
SUPERVISED
Train the
model on data
where the
correct answer
is included.
Train the model
on data with no
answers and
see what it
comes up with.
E.g. What bid
price is likely
to win this
impression
E. g. What do
people who click
my ad have in
common
UNSUPERVISED
SUPERVISED VS. UNSUPERVISED,
APPLIED TO ADVERTISINGAI IN PROGRAMMATIC
Programmatic advertising is exceptionally well-
suited for AI. A world with billions of impressions
auctioned in fractions of a second and always-
changing circumstances creates a scale of
multi-factorial problems that can only be solved
effectively with the help of AI.
“To achieve an objective becomes very hard without
the help of an AI platform that can do a lot of the
heavy lifting,” says Xaxis CEO Nicolas Bidon. “The
amount of data, the combinations that can result, is
growing exponentially to the point that a human will
have trouble determining the right bidding strategy
to buy media for a client.”
DESIGNING MEDIA STRATEGIES
FOR AN AI WORLD
It’s a given advertisers want to reach the right
person at the right time with the right message,
and of course, at the right price.
DIGITAL STRATEGY IS EVOLVING
But, in the digital era, a lot of advertising media
strategies have focused on targeting audience
segments that seem to contain propitious
prospects. Messages are tailored to them via
medium, platform, and screen, and, when possible,
factor in behaviors and locale.
Advertisers run their messages, gather results,
then try to optimize against marketing metrics
such as completed view rates, time of exposure,
clickthrough rates (CTR,) and effective cost per
thousand impressions (eCPM).
Yet, every one of those measurements is an imperfect
proxy for what’s actually desired: sales. By optimizing
toward CTR among a target audience, for example,
a media buyer may be getting a lot of the “right”
people onto a web page, but that’s not necessarily a
measure of sales effciency.
By contrast, artificial intelligence algorithms —
correctly instructed — can help optimize marketing
plans toward better sales metrics.
Audience segments based on demographic,
behavioral, and geographic characteristics can be
highly effective, but they will always miss a large
portion of potential purchase intenders who deviate
from the standard definitions.
Digital media strategies that use AI to identify and
locate prospects without bias or assumptions will
find customers, not just segments.
Take, for example, a high-end home appliances
maker that may be willing to spend $100 to sell one
machine. Articulated in that way — rather than in
terms of metrics like demographics or CPM — data
Determining
the bidding
strategy to
achieve specifc
outcomes
becomes very
hard without
the help of
an AI platform.scientists and engineers can focus on optimizing for
the ratio of the cost to market the appliance against
the number of appliances that marketing sells.
With machine learning, AI may even make some
counter-intuitive leaps, such as finding that a
higher volume of less expensive clickthroughs leads
to better business results than more targeted CTRs.
Plus, a machine-powered bidding strategy can
extract maximum value by conducting large
numbers of real-time tests in very granular
increments well beyond the capabilities of any
human. Where a programmatic specialist might test
bids in five broad increments from $1.00 to $2.00,
AI can run 10,000 tests in that range in increments
of fractions of a penny to hone in on precisely the
most cost-effective bid.
In this process, a feedback loop is created in which
the outcomes continually improve in lightning-fast
increments that add up to big results.
AI, APPLIED TO DYNAMIC
CREATIVE OPTIMIZATION
TEST
Ads run based on models
in small samples
MEASURE
Effectiveness evaluated
against outcomes
LEARN
Inputs refned to make
improved effectiveness
MODEL
Programmatic
specialists say
what should work
VARIABLES
COLOR:GEOGRAPHY:CONTENT:AI can be used
to help with dynamic
creative optimization
(DCO), continually increasing
effectiveness by choosing
combinations of creative components,
testing outcomes, and improving.
Working with skilled professionals,
AI used for DCO will create many
multiples more creative
possibilites than humans
could at a fraction of
the cost.。