Some time ago I saw this interesting little post –Recommendation Engine Secrets We Don’t Want You to Know: It’s not as Complicated as We’d Have You Think – that made the point that:
Most recommendation engines use one of a handful of methods that are well understood
And they are correct, of course. Recommendation engines involve some well understood elements:
- Data mining to determine significant customer segments, based on behavior
- Analytics to predict which products will be attractive to these segments
- Rules to enforce policies or regulations, determine pricing etc
- All packaged up into a Decision Service that takes in context information and returns a recommendation or set of recommendations.
The article went on to say that the issues with adopting recommendation engines were not in the “black box” that makes the recommendation but in cost, ease of setup/integration and customizability. Me I would add organizational change to that list (and put it first) as it is essential that you consider things like bonus or commission structures, marketing campaigns and more as you implement a recommendation engine.
One more thing – while people think of recommendation engines in a consumer-product environment there are, in fact, an almost infinite range of them. An engine can recommend policy coverages in insurance, delivery options in retail, suppliers in manufacturing, carriers in transportation and much more. All “recommendations”, all Decision Services, all useful.