Your recommendation engine is only as smart as your customers’ purchases are.

To set the problem, imagine two companies using recommendation engines: an online music streaming service and a luxury watch retailer. Customers of a music streaming service listen to (acquire) thousands of songs, often on a whim as the product acquisition cost is low (for a subscription based service, the acquisition cost goes down the more you listen to and so customers are encouraged, economically, to acquire many items to “get their money’s worth”). In other words, product acquisition in this business is akin to impulse buying — users make little investment in deciding whether they really want a product as there is little additional cost to them acquiring it (other than the time taken to listen). The luxury watch company, on the other hand, has customers who buy a small number of expensive products. The customers invest heavily in researching what they want and are almost always satisfied with the product they purchase, provided they have done enough research. Customers of the watch company are making an informed purchase. These two business models should be using totally different recommendation engines; here I will explain why.