I come back to this essay every time I get stuck writing. Thanks, Daniel!
Very interesting piece Daniel,
I myself was reluctant at first to introduce Session context or any other type of affinity driven models.
It has taking us (the EmpathyBroker team) over 2 years of trial and error to reach some fair judgement on the matter. Here a summary if it helps:
I think equally important is HOW MANY options to present to the user. Type-ahead can be about predicting what a user is thinking but it can also be about suggestions about what is possible. The tradeoff of popularity, probability, and other metadata likely looks different when predicting than when suggesting. But I’ll leave it to you to prove to me otherwise since you are much at better at that than I am :)