A Recommendation Engine Flop: Scary Recommendations

There’s definitely something wrong with the recommendation engine if Amazon offers you creepy costumes several days after Halloween. But don’t despair: At Dastani Consulting, we know there’s a better way. Predictive Analytics.

When I visited the Amazon site three days after Halloween, I really got a surprise: They were recommending the Grim Reaper and Co. as a party outfit. Why? Because I had already purchased similar products. This had simply been picked up by the recommendation machine as one of my favoured product groups.

It would be more intelligent to predict what I’m going to purchase in the future. Then the recommendation engine would no doubt have deduced that customers who have just bought such an outfit don’t need another one, and that for several days, no more of these outfits had been sold.

This wouldn’t have happened with our predictive analytics software. Our recommendation engine predicts what a customer is going to buy next and presents him or her with a ‘next best offer’. This technique is based on associative predictions and leads to significantly better results. And that is when predictions begin to fulfil their purpose in the sales process – by maximising turnover.