Every interesting project contains at least one thing nobody on the team has done before. That is usually why it is interesting, and it is always why the estimate is wrong.
Separate the known from the unknown, out loud
We split every estimate into two lists. The first is work we have done shapes of before; those numbers are reliable within about twenty percent. The second is work we have not; those are not estimates at all, they are guesses wearing a number.
The second list gets a timeboxed spike before anyone quotes it. Two days to find out is cheaper than six weeks of finding out slowly, and a partner would far rather buy two days of certainty than inherit our optimism.
- Name the riskiest assumption in the brief. Estimate that first.
- Quote ranges, not points, and say what would collapse the range.
- Re-estimate after the spike, in front of the client, using the same format.
None of this makes the guess accurate. It makes the guess legible, which is the part that actually protects the relationship.