A fast growing company in corporate training space provides its customers an online content marketplace where they can buy courses. When onboarding new customers, the client’s customer success reps manually provide recommended content based on their intuition about what similar companies like. Our client wanted to scale this process so that customer success reps provide better recommendations, faster using a data-driven approach.
Synaptiq developed a series of models in a Python notebook that solve the cold start challenge for new customers and new courses.
Identifying similar customers to the newly onboarded customer
Finding courses used by similar customers
Finding courses that are like the courses used by similar customers
Ranking the resultant list of recommended course
Course similarity was computed using NLP methods based on course abstracts, and we used dimensionality reduction methods to project courses and customers into a latent space in which the sparse metadata we had in the cold start case could be used to compute measures of similarity. Supervised machine learning methods then tuned weights to build a model that accurately predicted the rating a customer would give for any course.
While we set out to solve the cold start problem only, we also provided a solution for recommending courses to existing customers so that the client could also use the recommender system after onboarding. We provided the client two Python notebooks for them to integrate into their applications. They plan to roll out the recommender system to their customer success reps late this summer.