Base rates: how often does this actually work?

The inside view builds a forecast from the details of your situation: your team, your plan, your reasons for confidence. The outside view asks what happened to the reference class of things like this. The inside view is more engaging, more specific, and reliably more optimistic.

What it looks like in practice

"Our first sales leader will work out because we have a strong process and a clear brief." The outside view asks how often first sales leaders work out, and the honest answer across most sources is somewhere around a third, with failures clustered early. That does not mean do not hire. It means budget for the median outcome rather than the one you are picturing, and design for the failure mode with the highest frequency.

Your situation is genuinely different. So was everyone else's.

Building a reference class when there is no dataset

• Define the class narrowly enough to be relevant, widely enough to have members. "B2B software companies our size changing pricing" beats both "companies" and "companies exactly like us".
• Count what you can reach. Five founders who tried it is a small sample and vastly better than none.
• Look for the failures deliberately. They are systematically harder to find, because nobody writes up the pricing change that went nowhere.
• State a number and a range. "Probably fine" cannot be checked in six months. "Seventy per cent, give or take fifteen" can.

Why this is trainable

In the largest forecasting tournament run on the question, roughly an hour of training in probabilistic reasoning improved accuracy by about 10% on Brier scores, and the effect persisted across four years. Part of that training was explicitly about reference classes and stating coherent probabilities. It is a method, not a talent.

Five Peers editorial note. Sources named in the text.