And he
should know. He is in the business of monitoring “the shifting dynamics of the
marketplace context within which consumers live and shop.” Now, with a media measurement
industry that is increasingly moving towards artificial intelligence and
algorithms, his work is involving more of these technologies.
Charlene Weisler: How
prevalent is the impact of algorithms and AI on the marketplace?
J. Walker Smith: It is important to distinguish algorithms
and AI. Algorithms are structured
decision rules that, nowadays, are executed through software used to run
automated marketing programs or consumer apps.
AI consists of learning systems that modify and (hopefully) improve
algorithms or decision rules over time. AI
is still being tested and assessed, although we hear a lot about many of its
big early successes. Algorithms are the
beating heart of programmatic marketing and real-time targeting, and they are
the way in which all kinds of apps work.
Algorithms make the matches for dating software, figure out which cars
are available that meet your preferences and which new music your streaming
service pushes at you. They determine
which movies to suggest you watch or additional products you might want to add
to your basket. Looking ahead,
predictive software systems will get better at profiling consumers (within
regulatory restrictions about data) and matching inventory to individual
profiles. Personalization or
customization is a process driven by algorithms.
Weisler: What are the
positive and negative aspects of this?
Smith: An enormous amount of time and money is wasted on bad
matches. Nobody wants to spend money on a
product that isn't a good fit or waste time learning something that is a bad
fit. Nobody wants to get entangled in a
bad relationship that could have been avoided with an algorithm. Economists care about this because they study
efficiencies and bad matches are a drag on the economy. It's hard to measure, but economists can
point to things like people finding jobs faster and being better matched as
direct improvements to macroeconomic welfare.
The negative aspects are two-fold.
First, people like to choose and to be involved in deciding. An algorithm that does everything for you
takes away the pleasure of discovery.
However, this can be solved with future software that builds in
discovery. People can also control
themselves by using algorithms to aid decisions rather than to make
decisions. We refer to this as the
"algorithmically enabled consumer" — using algorithms to make better,
smarter, more satisfying decisions. The
second thing to protect against is data security. Algorithms require data about consumers, so
considerations of data privacy are front and center. Activists and regulators are more worried
about this than the typical consumer, though.
I believe we will eventually reach an equilibrium point, and then from
there, consumers will still want algorithms that can predict or aid in
decision-making.
Weisler: How fast
will it grow?
Smith: It's growing fast and is very prevalent already. There are no good metrics to track algorithms
per se, but the take-off rate of the Internet of Things is the best proxy,
because algorithms will be embedded in all of these devices. Thye are in every consumer app too and that
has become the new interface for content and products. So, the question of how fast it will grow is
a bit of looking backwards because algorithms are already present en masse.
Weisler: Can
algorithms be "gamed" and if so, how?
Smith: Everything can be gamed. You game algorithms by figuring out the
inputs and the relative value an equation places on each input. This is how people have tried to game Google,
and it's why Google keeps upgrading its algorithm. Indeed, gaming algorithms will become a large
part of tomorrow's marketing landscape — not to do anything inappropriate or
illegal, but to increase exposure to consumers and to boost the likelihood of
consumers choosing one brand over another.
If this sounds familiar, it's because this is what marketers have always
tried to do — not necessarily by gaming something, but just by putting more
effort and money into the things that are known to work best in getting
consumers to choose one brand over another.
Weisler: Algorithms
can be wrong - what can a consumer do in that case?
Smith: In fact, algorithms are wrong a lot. Marketers play the odds, looking to increase
the likelihood of consumers choosing their brand—and they want to do so
affordably. Structured decision rules to
accomplish this have always been embedded in media buying equations and
marketing rules-of-thumb. Algorithms are
just the 21st century version of that.
So being wrong per se is not bad for marketers, as long as they are
being wrong less frequently by using an algorithm. For consumers, they must apply nothing more
than normal diligence. Consumers also
know that many decisions will be less than optimal, so a good algorithm is one
that minimizes not eliminates bad choices.
Algorithmic mistakes are not life-threatening, at least not yet, so for
consumers, wrong outcomes are just an inconvenience.
Weisler: How do they
evolve?
Smith: Algorithms evolve through experience that improves
the models embedded in the algorithmic system.
This can be done periodically over time or it can be done in
real-time. Validations are run to assess
predictions and then once enough data has been accumulated, updates are
made. The latter is the AI future that
is getting so much press. By the way, the
next frontier of evolution is voice assistants.
Algorithms will be a big part of voice-based systems, but this is just
in its initial stages right now.
Weisler: What can
marketers do to not only prepare but excel?
Smith: Marketers need to learn to operate and change their
own systems more quickly. The
battleground in digital is better data and better models, which is to say,
better algorithms. For the algorithms
that consumers use, marketers must learn how to be responsive to their systems
and how to build these systems into their understanding of the consumer
decision journey. Right now, marketers
implicitly assume that it's the same ol' consumer taking in information and
making decisions. Yet increasingly, it
is consumers using algorithms to do that, so algorithms are the audience not
consumers. Marketers must learn how to
"advertise to algorithms," as we like to say.
Weisler: What types
of datasets are needed to get started? To progress?
Smith: We need better databases of consumer profile data,
better databases of in-market response to advertising and promotions, real-time
databases of actual choices or searching or consideration or queries, better
validation databases of actual outcomes and more servers and processing power
to use algorithms that can perform and take actions in real-time.
Weisler: Doesn’t this
type of marketing lead to "fishing in the same pool" for consumers.
What about potential customers who may not know about your product or service?
Smith: There is nothing new about this with algorithms. In fact, algorithms are about standing out in
this crowded pool, and so are a new source of competitive advantage and a new
kind of barrier to entry. Algorithms are
not simply about marketing to past consumers or modeling past behavior. Going after existing customers is profitable when
done well, but marketers are aware that they have to grow their franchise and
the category itself.
