I want to distinguish between two kinds of regret, because only one of them produces anything useful.
The first kind is regret as punishment. Something went wrong — a company failed, a relationship ended, a decision produced consequences you did not want — and you carry the weight of it indefinitely. Replaying it. Feeling bad about it. Using it as evidence that you are fundamentally flawed. This kind of regret is not useful. It does not change the past. It does not produce better future behaviour. It produces a diminished sense of self, which makes future behaviour worse.
The second kind is regret as data. Something went wrong, you feel bad about it, and you use that feeling as the entry point for a specific analytical process: what was the decision, what was wrong about it, what would the better decision have been, and how would I recognise the same situation if it arose again? This kind of regret is genuinely valuable. It updates the formula.
When my companies failed, I had a period of the first kind of regret. The media coverage. The legal consequences. The public characterisation of events that were far more complex than any single label could capture. The loss of the businesses I had built — businesses that had employed people, that had paid taxes, that had tried to solve real problems even when the structure was ultimately fragile.
What converted that into the second kind was a simple discipline: I stopped asking what happened to me and started asking what I would build differently. Not as a deflection — as a genuine engineering question. The answer produced structural changes I have implemented in everything I have built since. Separate legal entities. Geographically diversified operations. Regulatory navigation built in from the start rather than added as an afterthought. Each of those changes came from a specific regret about a specific structural decision.
The profound insight is that guilt is a signal, not a sentence. It is telling you something about a decision that diverged from a standard you hold. Listen to it long enough to extract the specific lesson. Then move. A signal that runs past its useful period is just noise — and painful noise at that. Extract the data. Update the formula. The regret has done its job.