President Barack Obama’s credit card got declined in a New York restaurant. Because, to quote Obama himself, “…I guess I don’t use it enough, so they thought there was some fraud going on.”
While this is the most high profile example of predictive analytics going awry, it’s not the first, and it certainly won’t be the last.
Many of us have gone through the experience of our credit cards getting declined while we’re traveling, making a large purchase, using the card too frequently – or too infrequently, as in the case of President Obama.
When a genuine cardholder’s transaction is declined, a “false positive” has occurred.
In a more generic scenario, the term refers to the condition in which the Predictive Analytics software raises an alarm which turns out to be false.
False positives have been a major bugbear of Predictive Analytics technology ever since its inception. And, almost right since then, we’ve been hearing assurances from practitioners that False Positives would decline as predictive analytics systems gathered more data, made more predictions, and became more “self-learning”.
If someone believes that claim after the not-so-new technology just bungled with “the most powerful man on the planet”, well, I’d like to sell them my beachfront property in Arizona, London or New Delhi.
False positive is a major reason why I’d root for allocating a part of your analytics budget to Data Mining instead of spending it all on Predictive Analytics.
As we saw in our previous post titled Difference Between Data Mining And Predictive Analytics, Data Mining appears somewhat unscientific.
But I suspect that’s primarily because its practitioners follow the purist approach of mining all kinds of data and letting the “chips fall where they might”. The problem with that approach is that chips often fall wrongly, or on the wrong heads, or at the wrong time – or all three.
Data Mining can overcome its perception problem by focusing on areas that would intuitively make business sense, even if only in hindsight.
Let me illustrate this with the following case study of a data mining exercise carried out by a bank that arrived at the following conclusion (locales and other details changed to protect customer confidentiality):
There’s a higher attach rate of business loans with home loans in Norfolk as compared to other offices.
This finding made immediate business sense to the head of credit of the bank when he found out that, unlike in its other offices, business banking and retail banking sales people sat at the same office in Norfolk, a practice that led to better exchange of information of mortgage buyers being in-market for business loans and vice versa.
The bank’s Data Mining team followed this up with the following recommendation:
“Let’s co-locate our retail banking and business banking teams in all regions. If the Norfolk effect happens everywhere else, we can grow our revenues by x% with no increase in cost or risk.”
As we can see, Data Mining delivered a highly actionable business insight in this instance.
If Data Mining takes a business-oriented approach towards mining data, it can pull analytics out of its current hair-splitting rut and elevate it to a strategic level.
While there are many implementation challenges around process, HR and so on, Data Mining can earn a seat for analytics in the boardroom.