The halo and the post-mortem

You can tell a great deal about a firm from the stories it tells after the result is in. Win the pitch, and it was the strategy, the chemistry in the room, the nerve to hold the price. Lose it, and it was the client's politics, the incumbent's relationships, the brief that was wired from the start. The same firm, often the same people, narrating two events that may have been a coin toss apart, with complete and opposite confidence about cause.

This is not dishonesty. It is the most natural movement of the human mind, and it has a name. Phil Rosenzweig calls it the halo effect: we observe the outcome first, and then read every contributing factor in its light. A company that posted good numbers is described, after the fact, as focused, bold and customer-obsessed. Let the numbers turn, and the very same company is rigid, reckless and complacent — though nothing about it may have changed except the result. The qualities did not cause the outcome. The outcome coloured how we now see the qualities.

The reason this matters for a post-mortem is that the post-mortem is the halo's natural habitat. By definition it begins from the result and reasons backwards — which is exactly the direction in which causation cannot be reliably read.

Why the tidy story is almost always wrong

Rosenzweig's deeper target is what he calls the delusion of single explanations. The stories businesses tell themselves crave one clean cause — we won because of the work — when the real picture is a tangle of factors, most of them correlated with each other, several of them outside anyone's control. A single cause is satisfying precisely because it is actionable: do more of the thing, avoid the other thing. But satisfaction is not accuracy, and the monocausal story is usually the first sign that the analysis has stopped too early.

Francis Gavin lists this among what he calls the marks of historical malpractice — the tells that someone is misusing the past. There are four: the facile analogy, the claim of inevitability, the single cause, and the division of the story into clean good and evil. A post-mortem that produces a hero and a villain, a single decisive moment, and the conclusion that it could only ever have gone this way, has committed all four. It feels like analysis. It is closer to myth-making.

The historian's first reflex against this is suspicion of the tidy account — particularly the flattering one. Margaret MacMillan describes the historian's proper stance as that of an examining magistrate: weighing the witnesses, taking the evidence seriously especially when it cuts against the verdict you have already reached. That last clause is the whole discipline. The post-mortem that only gathers evidence for the explanation the room already prefers is not an inquiry; it is a defence.

Separating the decision from the outcome

The single most useful move thinking historically offers a retrospective is to prise apart two things that the halo welds together: the quality of a decision and the quality of its outcome.

A good decision is one that was sound given what could reasonably have been known at the time. A good outcome is one that turned out well. Under genuine uncertainty these come apart constantly: good decisions produce bad outcomes, and bad decisions are rescued by luck. Gavin calls the failure to see this outcome bias, and he has a vivid illustration in what he names the puzzle of the Third Balkan War — the way 1914, which looked eminently avoidable to the people living through it, came to seem inevitable the moment we knew how it ended. Hindsight does not merely record the past; it smooths it, squeezing out the genuine uncertainty that everyone actually faced.

Gerd Gigerenzer sharpens the same point by insisting on the difference between risk and uncertainty. Risk is a world of calculable odds, where a poor result really can indict the method. Uncertainty is a world where not everything can be known, and where even an excellent decision carries a real chance of turning out badly. Most consequential agency decisions — the pitch, the hire, the new service line — live in the second world. A post-mortem that judges them as if they lived in the first, treating every bad outcome as proof of a bad call, is asking a question the situation cannot answer.

Marc Bloch put the corrective most plainly, writing as a historian rather than a manager: the task is to understand before you judge. The retrospective that races to assign praise and blame has skipped the only step that could make the praise or blame worth anything.

A post-mortem the historian would recognise

So what does this change on Monday morning, in the room where the team gathers to work out what happened? Four habits, none of them elaborate.

Reconstruct the decision before you know the result. Neustadt and May taught their students to sort a situation into three columns — what is Known, what is Unclear, what is merely Presumed. Run a post-mortem the same way, but as the decision-makers would have filled it in at the time. The point is to see the choice through what one of Gavin's borrowed phrases calls the foggy windshield, not the rear-view mirror. Most of what now looks obvious sat squarely in the "presumed" column when it mattered.

Name several causes, on purpose. Against the delusion of the single explanation, require the room to hold at least three contributing factors and to resist collapsing them into one. Complexity here is not woolliness; it is accuracy. The honest answer to "why did we win?" is usually "for several reasons, and partly luck."

Ask what would have changed your mind. Neustadt and May borrowed a device from a public-health official they call Alexander's Question: what new evidence would change my presumption? — asked before the event, then watched for. Built into a decision at the outset, it gives the later post-mortem something real to test against, instead of a story assembled to fit the result.

Compare against the firms that did the same and lost. Rosenzweig's warning against connecting the winning dots is that studying only your successes, and reading off their shared features, proves nothing without the failures that shared those very features. The team that "held its price and won" should ask how many teams held their price and lost. Performance, as he insists, is relative: you can do everything better than last year and still come second to a rival who improved faster.

The caveat

There is a way to get this wrong, and complexity is the safeguard against it. Refusing the single cause does not mean replacing "it was the team" with "it was all luck" — that is merely another single explanation wearing humbler clothes. Some decisions really were better than others; some failures really do have a locatable cause worth fixing. The discipline is not to abandon judgment but to earn it: to hold several causes at once, to weigh the role of chance honestly, and to separate what the team controlled from what it did not.

Done well, the result is not a softer post-mortem but a more useful one. It stops the firm learning false lessons from its wins — which is the more dangerous error, because a flattering story is so much easier to keep believing.

So the question to put to the room is not "whose fault was it?" or even "what did we do right?" It is the historian's question, and it is harder: if we ran this decision again, knowing only what we knew then, what would we actually change — and how much of the result was ever ours to determine at all?

This is a companion piece to the argument at the argument. If it resonates — or if you want to push back on it — I'd be glad to hear from you: chris@christhurling.com.