A decade ago I watched vendors of site search solutions teach customers new skills: how to manage synonyms and common misspellings, how to identify poorly performing searches and improve them, how to change their content to produce better search results, how to create and manage rules to control search, how to offer the right navigation facets…
The list was long and daunting, and if you wanted to have great site search, you had to invest in these unappealing skills. Most sites did not invest enough to have great search, but managed to cobble together enough resource to have half-way decent search. Hence the unchanging verdict from users, “Search sucks.”
We’re all pretty excited about recommendations these days. Users like them because they are easy to use and often interesting. Web site managers like them because you can achieve “half-way decent” with very little effort, delivering a nice bump to your bottom line.
But if you’re not content with “half-way decent,” what are the tasks to achieve great results with recommendations? Most likely, you’ll be making a study of best practices, so you can set up recommendations well. Then you are going to test and improve them, iteratively, by finding poor performing segments, pages and categories and trying out different approaches. And this last task will never end, because there are always poor performers; your items change; trends, fashions, and markets change; and your visitors change.
But how much should you have to invest in this improvement program? I think recommendation solutions can do much, much more of the heavy lifting than they do today. Recommendation engines are prediction engines. They predict, using a variety of methods, the items that a person is most likely to find interesting. The vendors tout the expertise of their staff in advising you on how to deploy and improve recommendations. They claim to know best practices.
With all this technology and expertise, shouldn’t the recommendation solution deploy best practices automatically? Tweaking the best-practice default would be a lot easier than learning best practices and applying them. Shouldn’t its analysis and modeling identify poor performing areas, and predict actions that will best fix them?
Here are three questions that you should ask any vendor whose solution you are considering:
1. In what ways does the solution implement best practices as a default, selecting locations, recommendation strategies, and algorithms?
2. In what ways can the solution automatically optimize recommendation approaches to reach client-specified business goals?
3. How does the solution identify and prioritize opportunities for improving recommendations?