Top 5 Factors in Choosing a Recommendation System
I have published detailed requirements for recommendations systems, with at least a hundred questions you should consider when evaluating a solution. But a hundred details is not where you want to start your technology selection process. You start by narrowing your focus to find your most likely partners. I think there are just five factors to consider initially:
1. Vendor’s target market
2. Algorithms and data sources
3. Interface you’ll use to manage recommendations
4. Automation and self-optimization
5. APIs and data access
Why are these elements important, and what are the considerations?
Vendor’s target market. You want to be the vendor’s target market, because then the solution is designed for you. A features checklist might make it seem that a solution is a match – then you find out that “shirt” meant button-down to you, but it meant t-shirt to them. There’s no fixing that one. Are you in the vendor’s target market, or not?
Algorithms and data sources. This is the guts of the solution. Most organizations I’ve talked to need a range of methods for making recommendations: collaborative filtering, predictive modeling, rules, lists. And the methods must draw on a range of data: history, current behavior, segmentation, demographic data, geographic data, explicitly stated preferences. Having a range of methods using a variety of data means that recommendations can be extensively tuned for maximum performance, and you have maximum control over how recommendations are selected. Does the solution support all of these methods and data sources, or not?
Interface you’ll use. You must expect to specify your recommendation tactics (what methods to use, and where to use them). And then you must be prepared to monitor the results, and test various approaches – in short, you are now in the continual optimization game. The interface must make it easy to see where your attention is needed, easy to make changes, easy to see the results. Does the solution provide a single interface to manage recommendations across touchpoints, segments, and marketing programs, or not?
Automation and self-optimization. Part of easy relates to how many actions you need to take to be successful. You may be able to devote significant attention and activity to your top pages, customers, or items, in order to get results. But you really need results from all your pages, customers and items. How much automation and self-optimization is built in to the system? Does it allow you to set a global strategy, and then refine it for individual cases? Will the solution try out algorithm variants to find the optimum, without your help? Does it create segments for you that you can use in refining the recommendations? So, is the vendor constantly looking for more ways to automate and self-optimize, or not?
APIs and data access. Recommendations are a great first step in personalizing your visitor experience. You will likely be able to deploy them quickly using out-of-the-box tactics. But the day will come when you want to refine how recommendations use data, and how they interact with other systems such as content management, ecommerce, CRM, and digital marketing. You’ll want program access to the recommendation services and recommendation-related data. So, are there APIs for all the services, or not?
After you’ve filled your dance card with the partners who meet these criteria, you waltz them through the detailed evaluation (here).