Main idea

Writing down first the questions a reference model will have to answer is the standard device for delimiting its ground. The exercise is usually done in a meeting room, cold, and always produces the same thing: questions that look like what we think matters.

The other route is not to write them at all. Product questions already arrive, archived with their answers in support threads and team exchanges. You reread them all, condense them into around twenty generic questions, and each one then covers a set of questions that were actually asked. None comes from anyone's imagination.

These questions are not elegant — "what can still be cancelled, and from what moment on can it no longer be?", "who owns an automatically generated order if that person leaves the company?". They are the ones that cost someone half an hour, several times a month.

Why it matters

They immediately give the order of work, and that order stops being a preference: each question touches one or more areas of the product, you only have to count. When more than half the questions land in the same area, the starting point is decided by a tally nobody argues with.

And the cost of collection is nil. No workshop, no priority negotiation between participants: the material exists, it is waiting to be reread.

Nuances and limits

The support corpus carries what is blocking today, never what nobody dares raise. A question never asked because everyone gave up on getting an answer does not appear in it — and that is precisely the kind of area a reference model ought to cover.

Two customer corpora do not, moreover, say the same thing: interviews carry what customers want to build, tickets carry what breaks. Framing cannot rest on only one of the two.

Open questions

• How do you spot the questions that are no longer asked because the organization has learned not to ask them?