Estimating in the Dark

For our first three years we did what almost every small studio does: we estimated projects at the moment we understood them least, wrote a number down, and then absorbed the difference between that number and reality as unpaid overtime. We knew this was happening. We had a spreadsheet showing exactly how much it was happening. We kept doing it, because the alternative — telling a prospective client that we could not yet say what the work would cost — felt like an admission of incompetence.

What changed our minds was tracking hours honestly for a year. The pattern was not that we underestimated evenly; it was that our estimates were roughly correct on the projects we had done something similar to before, and wildly wrong on everything else. The variance was not noise, it was information: it was telling us precisely which projects we did not yet understand.

So we split the estimate in two. Any project we have not done a close cousin of gets a fixed, modest fee for a scoping phase — reading, interviews, a written problem statement — and no number at all for the work beyond it. The number for the work is produced at the end of that phase, when we can actually produce one. Clients accepted this far more readily than we expected. In hindsight this is obvious: they have also been burned by a confident number that turned out to be fiction.

The uncomfortable part is what it revealed internally. Two of our recurring project types were quietly unprofitable and had been for years, subsidised by the ones that ran clean. We had assumed we were bad at estimating. We were actually bad at one specific kind of work, and the averaging had hidden it. Fixing that meant declining a category of project we had thought of as core.

None of this is a pricing model — it is a refusal to produce a number under conditions where a number cannot be produced. It has cost us two projects we wanted and saved us from about six we would have regretted. On balance the arithmetic is not close.