Recommendation engines may be brilliant in improving sales for eTailers, but they’re not always perfect. We spoke to Padmanabhan Ramaswamy, former head of business intelligence analytics at online marketplace GEMFIVE, to find out how recommendation engines can work against an eTailer’s interests.
Have you seen recommendation engines work against an eTailer’s best interests?
This is very much much dependent on the algorithm of the recommendation engine and what parameters have gone into consideration. If an eTailer is trying to improve margins by pushing higher-margin products but lower-margin products are more popular, the algorithm would end up recommending more lower-margin products and working against strategy. Another example would be when the eTailer is trying to increase the sales of slow moving product SKUs and hence undertaking various steps in that direction like price discounting, offers, etc. In such a case many recommendation engines are likely to push fast moving SKUs since these tend to be more popular and hence working against the strategy of pushing the slow-moving SKUs.
What practical steps can eTailers take to avoid biasing their recommendation engines towards overly popular products?
This is dependent on the various factors that have been considered while building a recommendation engine. A robust recommendation engine is likely to have a multitude of factors like preferences, attitudes, behaviour of similar people, demographics, etc. The more the number of relevant factors or parameters the lesser the likelihood that additional bias will be assigned to a single parameter.
Are there cases where eTailers can become overly dependent on recommendation engines to push out products?
Not really. I don’t believe the majority of eTailers have reached that stage yet where they are overly dependent on the recommendation engines. In most cases eTailers are trying to improve their recommendation engines and still learning to improve the quality of the recommendations and hence the result.
How do recommendation engines account for human intuition, especially with products where success is not indicated by the data?
One of the key things recommendation engines are supposed to do well is to base their recommendation on underlying data/behaviour of various types. So, in a way, recommendation engines are intended to remove human intuition. Hence, if your intuition suddenly goes totally against your attitudes, underlying values, similarity with the segment your belong to etc. then the recommendation is not likely to be accurate (simply because you are outlier or an exception).
Ramasawamy is set to join more than 200 other e-commerce, retail and payments professionals speaking at Seamless Asia. For more information, download our brochure here, or go right ahead and register here.