Civicom: Your project success is our number one priority

Does Machine Learning Signal the End for Marketing Research?

Machine learning’s potential to raise corporate efficiency is huge. But understanding an audience that consists of you and me, each with our own wonderful and unique set of illogical, emotional quirks, is something quite different.

man to machine

By Sinead Hasson

When someone introduced me to the concept of ‘machine learning’ a few weeks ago I caught an unsettling glimpse of the not-too-distant future. For the uninitiated, machine learning is a new field of data analysis, which Stanford University defines as ‘the science of getting computers to learn without being explicitly programmed’. We’re talking seriously smart algo-based software. So smart, in fact, that it can sift through the petabytes without being directed and identify many more trends than a human analyst can, produce more reliable forecasts and as a consequence (gulp) make better decisions too.

It’s news to no-one that a human analyst’s data interpretation is innately biased. That’s human nature; absolute impartiality is impossible for us to achieve. This, in itself, isn’t a problem – we live in a human world where the idea of ‘useful knowledge’ is heavily influenced by our social, economic, historical and cultural situation. The problem here is that we don’t always know the right questions to ask in the first place. By sheer depth and volume of analysis, however, machine learning promises to reveal what those missed lines of enquiry might be, having traced them back from abstract trends that it has already discovered.

All of which begs a question; if researchers are no longer needed to define the questions, discover the trends, make the forecasts or indeed influence the resultant decisions, they might as well all go home, right?

Wrong. Because fortunately, we don’t live in a world governed by machinery, and science fiction aside, it’s pretty unlikely that we ever will. Not all insights can be gained from logic. The ‘right’ decision isn’t always made from an operational analysis, or as a result of a statistical forecast. It’s why Mr. Spock needed Captain Kirk. In fact (humour me here) I think machine learning is Kirk delegating to Spock so he can get the broadest possible picture, upon which he can then impose his ego-centric, ever-so-human judgement.

Around 90% of all decisions made in major commercial organisations are operational in nature, so you can see why people are getting so excited – machine learning’s potential to raise corporate efficiency is huge. But understanding an audience that consists of you and me, each with our own wonderful and unique set of illogical, emotional quirks, is something quite different.

I hear that the Machine to Machine (M2M) age is coming, but you know what? Until we need to understand how machines, and not us fleshy consumers, are influenced and motivated by our clients’ brands, I’m not worried. Neither should market research professionals.

Join the debate @sineadh

Please share...

6 responses to “Does Machine Learning Signal the End for Marketing Research?

  1. But surely the power of machine learning is that it does have the ability to analyze down to you and me. And it only has to learn how that CMO makes decisions to be able to make the same types of decisions. Human market researchers are definitely going to be impacted by this technology. I am going wine drinking!

  2. Machile Learning, in the context of Market Research, will do in large scale what we current do with smaller data sets. At the end of the day, the ML pipeline is Data=>Methods/ Statistical Models=>Intelligence. We will still need humans to 1) ask the right questions (small data training big data)/ 2) create, test, train and validate the methods and models, 3) Interpret the outputs, improve the methods, recommend what to do with the results in a specific business context. So, instead of being a threat, ML will open new horizons to market researchers. The ones who envision that now will be the ones leading the way in the future.

  3. Thanks for your article, very stimulating. Let’s start by saying machine learning is not a NEW field of data analysis. I did my thesis on machine learning – neural networks to predict caffeine content of green coffee beans – and it was some 20 years ago (sigh). And I was all but a pioneer: you can go back to 1968, that’s 40 years ago, or even backwards to the work of Alan Turing. Because of cost, computational capability and better algorithms it has improved and evolved and it is now about to reach a tipping point. When I read “the dawn of” I simply chuckle.
    And the agonizing market research is realizing there is a (yet another) possible new threat, which provides the latest topic for self-proclaimed experts (I am not referring to the author of the article, but – oh dear – read recent posts and blogs …) and for next ESOMAR or local conference papers. Yes, an AI will (or is already?) be able to run a tracking study better+cheaper+faster than a team of researchers. An AI will be able to run a benchmarking study better+cheaper+faster than a team of researchers. Why has Google started its Google rewards program? Potentially they have the best profiled and biggest panel in the world and surely they are developing AI. Bespoke research? That’s where and AI will not replace for now, but rather support the research team (which will be much slimmer).
    Finally, a note of warning on “understanding an audience that consists of you and me, each with our own wonderful and unique set of illogical, emotional quirks, is something quite different”. By definition a AI is a machine which can fool a human passing the Turing test i.e. a machine is able to exhibit intelligent behavior indistinguishable from that of a human. An MIT AI has passed it recently. And yes it can understand an audience which consists of you and me and neither you nor me are able to realize it is a machine.
    Moreover, quantitative market research is based on means (sometimes mistakenly because distributions are not normal, but who cares about non-parametric stats, the graduate who does the SPSS cross tab does not even know what that is), so who cares about uniqueness. Maybe that’s the biggest mistake of MR in the years: as Taleb would say, it has focused on Mediocristan instead of Extremistan, on the white swan instead of the black swan.
    Very exciting times ahead.

Join the conversation

Sinead Hasson

Sinead Hasson

Owner & Managing Director, Hasson Associates Recruitment