To me this sounds - a litte - overengineered. To do more than 99.9% of the companies out there it would be enough to just send generic follow up autoresponders

To me this sounds - a litte - overengineered. To do more than 99.9% of the companies out there it would be enough to just send generic follow up autoresponders based on sign up date. Email them on day 0, 1, 3, 7, 11, 14 ... etc. You can do this with AWeber, Mailchimp, Getresponse and a dozen other services. Just tell the customer what he can do with your service and ask a lot of questions: "Have you tried this?", "Did you know you can...", "How can we help?". Use the feedback you already have to come up with ideas for content. Point them to a lot of your help and learning resources. In each email tell them you are really interested in their feedback and give an easy option to contact you. Then read the feedback as it comes in - like manually reading, with your eyes ;-) - and respond promptly and indvidually. Setting his up takes a day or two - and mostly consists of coming up with ideas for the content. After having this running for a few weeks you might come to the conclusion that you don't need a complicated machine learning, metrics based system :-) I've written about this in another context earlier this year, basically about how we handle the trial period: http://news.ycombinator.com/item?id=2979048 http://news.ycombinator.com/item?id=2979048 TL;DR: "Eighty percent of success is showing up." (Woody Allen)