Machine Learning depends on Human Learning
As an ex-journalist, I used to make a living at the Financial Times making predictions, which was a relatively easy thing to do because growth was a constant.
Baked bean sales may have stalled during a recession but they always recovered, and there was never a dramatic change that saw beans being replaced with turkey croquettes or wiped out altogether by gym membership.
Now, with news that retail sales have fallen for the third month in a row, against the consensus of a rise of about a 0.9% rise in sales, we have yet more evidence that not only are we not in Kansas anymore, we are not in Oz either.
There is no good Brexit and there are no good US politics, so predict all you like, you cannot predict against bad or missing data. But what you can do, as a handful of retailers are discovering, is predict against good and plentiful data.
Which brings us to Machine Learning, a concept that came out of Big Maths but is now being sold to retailers by tech companies, most of whom are so blinded by the science that they have not yet worked out how to sell it to people who are unimpressed by maths, tech or science.
Machine Learning and its more challenging twin, automated decision making, are the only sensible option for retailers hoping to manage the complexity of multiple locations, formats and customer patterns; massive SKU numbers; and, customer types and behaviours almost as numerous as the customers themselves. Put in enough data from enough sources and retailers should get assortment predictions for replenishment, pricing and promotions that deliver higher margins.
And, they can deliver significant margin gains, as some German grocers have, more or less regardless of the wider economic climate, even allowing for Amazon Fresh targeting Germany.
Two problems. The first is perhaps the easiest to solve; how will vendors market Machine Learning in a way that is directly linked to retailers’ biggest challenges. The second is not so easy, because Machine Learning-based replenishment and pricing force people in different departments to collaborate, which means retail is being asked to undermine its entire business model. That will happen but it will not happen overnight and it will take the CEO to make the decision, not the individual buyers, merchandisers and category managers who have the most to lose or gain, depending on how you look at it.
But there is a third problem, more insidious than the first two and perhaps the hardest to solve. Who will tell the incumbent vendors with point solutions, who have been doing this a very long time, that their solutions will not solve the problem?
Making predictions on bad or missing data delivers almost no value, but convinces both vendors and retailers that the problem has been solved when it hasn’t. But it is time for humans to embrace the power of the machine to deliver better retailing, fit for a world where Trump is not the end, but the beginning of something new.