The Evolution of the Virtual Salesperson: Using Behavioral Context + Rules to Deliver Relevant Product Recommendations

(NOTE: Earlier this year, RichRelevance and CNET announced their strategic partnership to integrate CNET’s rule-based Intelligent Cross Sell recommendations with RichRelevance’s behaviorally-based RichRecs™ offering to offer the industry’s first truly robust personalization solution for the world’s largest online retail sites. Patrick Monasterio is a member of the CNET/{rr} development team and is working with joint customers on integrating this new solution.)

Patrick MonasterioWhen done right, relevant product recommendations can deliver up to a 30 percent lift in conversion and sustain or even increase these levels over the long term.  Therefore, the multi-million dollar question for merchants (literally) is “What recommendations do I want to show? How do I do this?”  The most successful recommendation emulates the logic of a good salesperson by delivering the right product to a specific person for a specific situation.

Once upon a time recommendations were a manually slotted smorgasbord of products for any given item page. The typical presentation of accessories for a laptop computer was dozens of SKUs, with the thought that this shotgun blast of the “best” recommendations would eventually find its mark on the cart page. Obviously, this is a poor practice and online retailers were hungry for a more intelligent approach.

Fast forward a few years to the birth of the black box.  Simply plug in the box and watch the algorithms shape the recommendations. By using the “wisdom of the crowds,” these recommendations seemed more organic than the previous approach and resulted in recommendations that were more rich, relevant, and—from a user’s perspective—more reasonable.  Recommendations grew more specific, so that although the number of recommendations decreased, their stickiness increased.

However, human ingenuity persists and websites soon caught onto the idea of monetization, that is, contracting with Company XYZ to insert their products as recommendations for a fee. How does a pure behavioral recommendation account for this? It can’t.  A pure behavioral engine doesn’t care that you have an agreement with X. If the “wisdom of the crowds” says the best mouse is from Z – so be it.

Enter the introduction of rules. Advanced recommendation engines are able to filter recommendations by targeting attributes.  A rules-based engine can filter recommendations based on price, size, category or anything that identifies the SKU and is communicated to the engine. You want all mice to be from company X? Done. Filter products based on price? Done.

But all that is gold does not glitter. Rules have nuances as well. They can become too stringent or too loose, potentially filtering out compelling products or including everything but the kitchen sink. Often too, rules are based on the consistency of data and catalog format—two requirements that can be hard to come by in the IT world.

In the end, the most impactful recommendation technology is like a sales person who utilizes both qualitative and quantitative information when working with a customer – in this case, employing both behavioral information and rule aspects. The technology gathers behavioral data from the macro-shopping environment, what the person has purchased in the past, and compares that against any promotions. Just as a sales person follows procedures based on the latest retail initiatives, so does the recommendation engine by adhering to intelligent rules. Only then can a recommendation replicate the logic and intuition of a salesperson—a salesperson that never tires, never rests, and is always on its “A” game.

Share :
Related Posts