A friend who runs a logistics company told me about the best decision he ever made: an acquisition that doubled the business inside three years. When I asked what the analysis had looked like, he laughed and said there had not been much. He liked the founder, the price was right, and it worked. Then he told me about the worst one, a warehouse lease that went wrong when a customer left. The analysis on that one had been careful. He still called it the worst decision he ever made.
He was doing what Annie Duke calls resulting: grading the decision by the outcome. It is the most natural thing in the world, and it is how a team learns the wrong lessons from its own history.
Why the outcome is a bad teacher
Duke spent two decades as a professional poker player, and Thinking in Bets (2018) is built on a distinction poker forces on you and business lets you avoid. Every decision is a bet with odds. A good bet can lose; a bad bet can win. If you grade yourself on the result, you will learn to repeat the lucky bad bets and avoid the unlucky good ones, and after a few years your judgment will be shaped by noise.
In poker the feedback is fast and frequent, so players learn to separate the decision from the card. In a company a big bet resolves once, a year later, in a fog of other events. There is no second hand. Which is why the separation has to be deliberate.
The two questions, kept apart
A look-back is a scheduled review of a decision after the fact. It works only if it asks two questions and refuses to let the answer to one contaminate the other.
The first question is about the call: given what we knew and could have known on the day, was this the right decision? Were the conditions we wrote down the right ones? Did we check the ones we could have checked? Did we hear from the people who would have known? Was the alternative we rejected really worse on the evidence, or did we just prefer this one?
The second question is about the outcome: what happened, compared with what we expected? Which conditions turned out true, which false, and which did we never find out about? Did the tripwires fire, and did we act when they did?
Answer the first before you look at the second. A team that starts with the outcome will reverse-engineer the call to match it, and a team that has just had a win will not examine the call at all.
Scoring the confidence
There is a third thing worth writing down at the look-back, and almost nobody does. On the day of the decision, how sure were you? Russo and Schoemaker (1992) asked more than two thousand managers to give ranges they were 90 percent confident about. Fewer than one percent were calibrated. The typical "90 percent sure" captured the truth about half the time.
Philip Tetlock's forecasting tournaments (Tetlock and Gardner, 2015) found that the people who improved were the ones who wrote down a probability, found out what happened, and scored themselves. They were rarely the smartest people in the tournament. They were the ones who kept score. A leadership team that records its confidence at commit and reads it at the look-back gets the same feedback, one decision at a time. After a dozen decisions the pattern is visible: this team is well calibrated on hiring and wildly overconfident on timelines, say. That is a fact about the team you cannot learn any other way.
Making the look-back happen
The reason most companies never do this is not disagreement with the idea. It is that the look-back has no natural owner and no natural date. The decision is made, the meeting ends, everyone moves on. Six months later the outcome is known, the conditions are forgotten, and there is nothing to review against.
So the look-back has to be booked at the moment of commitment, with a date and a name, and the conditions have to be written down where they will still be legible. If the record says "we bet that mid-market customers would accept a 15 percent increase, we were 70 percent sure, we would know by the end of the second quarter", the review writes itself. If the record says "raised prices", it does not.
Where Midfire fits
Midfire books the look-back when a decision is committed and asks the two questions separately when the date comes: first the call, then the outcome. The confidence recorded at commit is scored against what happened, so the calibration of the team becomes a number that moves. None of that is new thinking; Duke and Tetlock wrote the method. What the product adds is the date and the record, which are the two things that were missing every time I meant to do this and did not.
The version with no software
Take the last three big decisions your company made. For each, write the call in two sentences as it looked on the day, without the outcome. Then write what happened. Then ask, honestly, whether the good outcomes came from good calls. My friend with the acquisition did this and concluded, a little uncomfortably, that his best decision had been a lucky one and his worst had been sound. He is now much slower to buy companies on instinct. That is what a look-back is for.