A Demon Under The Streetlight

There's an old joke about a man searching for his car keys at night under a streetlight instead of around the car where his keys were ostensibly lost. When asked by a passerby why he was searching there he replies, “Because the light is better under the streetlight!"

When it comes to searching for insight into the consumer decision-making process, most of what we know is due to research methods that originate from social sciences. We rely on these methods to select a representative sample of consumers to study, to design experimental tests for campaigns and to compensate for gaps in data with statistical inference techniques.

But there is an alternative to dealing with samples and the limitations they carry -- namely the uncertainty as to whether research insights actually correspond to the real world.

With data sets increasing exponentially -- and with significant investments being made to create new platforms or Big Data tools to contain and make sense of this data -- it becomes easy to believe that the answers to all possible marketing effectiveness questions lie somewhere among those petabytes.

But while these developments have moved some to proclaim the end of scientific methods as we know it (that is, if everything is tracked, then there is no need to hypothesize, model and test anymore), time has proven that this is not true.

Today, The Big Data radicals are essentially channeling the French mathematician Laplace, who at the beginning of the 19th century postulated that if all the data about the universe were available, one could predict everything that would happen henceforth.

While this might make perfect sense in the abstract world of mathematics, Laplace's principle, known as "the Laplace Demon," does not hold up in either the world of thermodynamics or quantum physics. Apparently, our physical world is not deterministic. 

But would this principle work in a smaller world, such as the media/marketing ecosystem?

Well, yes, if we had ALL the data possible about the ecosystem. That would significantly change our approach, since the census is always superior to a sample.

But Big Data is not ALL of the data. Collecting infinite data pertaining to the marketing ecosystem would require an infinite investment. So it is only natural for a corporation to default into thinking about itself as a universe, and simply collect all the data to which it has access: for example, all purchase activities customers of company A, all behaviors on the site B, all searches at the search engine C., etc. The resulting Big Data would include all possible performance indicators, but the factors driving them would still remain outside of the framework in someone else's Big -- or not so big -- data set.

Data analysis is only valuable when it shows what causes what. And to connect two data sets, we'd still have to resort to some sort of statistical modeling.

Using traditional research methods, we’re always keenly aware that all measurement tools come with a degree of error -- and we try to account for that error. Big Data, by definition, includes a lot of random, irrelevant data. To separate the signal from the noise, mathematics is not sufficient; we'd still have to resort to the good old statistical method to correct the situation.

And bigger is not necessarily better, either. While this research approach could be compared to going out in the field with a flashlight, the Big Data approach is akin to installing a giant floodlight shining 24/7. But since this light is not easily portable, it might still not cover the entire territory -- where the keys are presumably located. Or it may blind us with its intense light and prevent us from distinguishing the keys from other shining objects lying in the grass.

The main obstacle to complete reliance on Big Data in marketing, however, lies not in technology or methodology, but in marketing and media practitioners who constantly experiment with novel strategies and introduce change, intentionally destabilizing the system to make the future different from the past. In the end, it looks as if we don't really believe in the past determining the future.

So while Big Data can tell us the minute details of what happened at the time it was collected, all the past data available may not be enough to forecast what happens next.

And in the end, we may need to search not just under that streetlight because the light is just better, but closer to our car instead, where they keys to consumer decision-making insight might better be found.

Tags: metrics
Recommend (15) Print RSS