Making Big Data Work

"Hmmm.  I see Mrs Mittleshmitz  just bought six more cans of Hint-O'-Bacon spray cheese. Wilhelm, email her a Buy-Eight-Get-One-Free coupon"

I did something last week I haven't done in a long time.  I actually read a whole issue of Marketing Magazine and its twice-daily emails.  On most occasions Marketing only offers stale news or filler "content" or both—its Fall TV preview is an egregious example.  But this time was different.  It featured articles on Big Data that, by an odd coincidence, coat-tailed its Data Driven Marketing event of August 20. 

The main issue with big data is how to make sense of billions of consumer data points in a timely manner and not get it all screwed up.  There's a lot riding on this because many marketers believe that if they torture the data enough with the right analytic it will confess a truth about how and why consumers act the way they do.  But, finding unicorns can be a bitch. 

Data models are only as good as the parameters and the assumptions used to construct them.  And the more data used to create the models, the greater the chance for error or misinterpretation.  Then, there is statistical noise, as well as false positives, that skew results, among other problems.  And just because certain things are detected doesn't mean they are connected.  Correlation is not causation.

Case in point: the lead article of the magazine, Lost in Data Translation, mentions Kevin Keane, who now runs a "neuromarketing" outfit called BrainSights but at one time was a data cruncher for a booze company.  One day, Kevin noticed a huge spike in sales that neither he nor the client could explain. Maybe it was a new ad campaign or perhaps a large shift in consumer preferences.  Unsure of the cause, he decided to sit on the results and dig deeper.  Turns out that was the right thing to do.  Kevin discovered that sales spiked because of something the data could never reveal: because of a threatened LCBO workers strike consumers were stocking up.  Kevin applied a valuable element to his data. 

That element is identified by Matthew Quint, director of global brand leadership at Columbia Business School.  I'm overlooking the fact that he teaches at Columbia for the moment because he uses a quote from Einstein (Princeton) to stress what big data lacks: “Not everything that can be counted counts, and not everything that counts can be counted.”  Quint says, “Sometimes data is only valuable with a human interpretation on top of it – what the data reveals and what insights come from a human analysis of it – but also sometimes we miss things, as humans, with our gut instincts, understanding and anecdotes.” Overlaying the data with some human insight has benefits, but it will provide much more if marketers apply it in a different location.

Right now marketers sift through terabytes of consumer information every day to create models of their ideal loyal customers—the golden grains of profitability.  What they winnow out, however, could be much more valuable.  The chaff—light customers—are separated out because they have no apparent historical worth or loyalty.  But that is dead wrong.  Loyal customers are nice and all, but they are usually too expensive to retain; besides chances are they'll buy your product anyway. 

The biggest opportunity for data marketers comes from finding light customers, those who rarely buy the brand or do so two or three times a year—the segment that provides about 60-70% of a brand's sales.  Applying that layer of human interpretation on top of the data to understand and reach the light buyer is the way to increase penetration (increasing the customer base) instead of increasing market share (getting regular customers to buy more).  This is where big data can pay off.