"many companies I have worked with have “sticky features”, although not all of them know it. (Some day when I don’t have another few thousand words to write I’ll tell you about what they are for Bingo Card Creator and how I know.)"
Thanks, I'll try to remember to write that up sometime.
Capsule summary: if you were to personally log into an arbitrary trial customer's account, you could predict whether they were likely to buy the software based on your intimate knowledge of customer behavior, and your guess would be far better than random. You can code heuristics to make that guess. You can make your heuristics adapt to data you collect, such that they no longer depend on your (flawed and limited) intimate knowledge.
Pretend that you assign customers something like a credit score, and simply chunk them into four buckets, from "people I'm most confident will buy the software" to "people I'm least confident will buy the software". My conversion rate for the free trial of my software is ~2%. Bucket A is closer to 20%. Bucket D, the largest bucket, is below .1%.
I think there is a way to take the knowledge that the customer who is using the app right now is in Bucket B in a fashion which makes the business a staggering amount of money. (Identify the customers at the margin, then push them over it.) What I lack, which has prevented me from blogging about this, is an empirical example of it working that I can talk about.
I think it is potentially as simple as "Offer them a coupon." or "Give them explicit directions on the interface to try the features that you know distinguish bucket A from bucket B." These interventions are, of course, split testable.
"many companies I have worked with have “sticky features”, although not all of them know it. (Some day when I don’t have another few thousand words to write I’ll tell you about what they are for Bingo Card Creator and how I know.)"