Why Did It Fail? Probably Not for Any of the Reasons You Listed

A project misses badly. In the review, three explanations get offered, roughly in this order:

• The strategy was wrong.
• The execution was poor.
• We didn't have the resources.

All three are plausible. All three are easy to argue for. And all three can be confidently, unanimously wrong, because the thing that actually moved the outcome was never written on the whiteboard.

The shape of the problem

Here is a scenario that happens constantly. A company runs 64 stores. Someone plots marketing spend against revenue, store by store.

The relationship is about as clean as real business data ever gets. The obvious read is that marketing works and there should be more of it.

Now press the button.

Split the same 64 stores by footfall — how many people walk past — and each group collapses into formless scatter. Footfall was driving both numbers all along. Busy locations get larger marketing budgets and they sell more anyway, for reasons that have nothing to do with the budget. The correlation was real. The causal story built on it was fiction.

What this is called, in each field that names it

The idea shows up everywhere with a different label:

• Statistics calls it omitted variable bias — a model that leaves out a relevant variable attributes that variable's effect to whatever it left in.
• Epidemiology calls it confounding, and has a precise definition: a confounder is something that influences both the supposed cause and the effect, and isn't on the path between them.
• Everyday argument calls it "correlation is not causation", which is true but too vague to act on. The useful version names the mechanism: there is usually a third thing, and your job is to find it.

The medical example is the cleanest. A study finds people taking a supplement are healthier. Supplement-takers also tend to exercise more, eat better, and see doctors more often. Unless the study accounts for that, it isn't measuring the supplement — it's measuring the kind of person who takes supplements.

Why smart teams miss it

It isn't in the data you have. Footfall wasn't on the chart because nobody pulled it. The variables that end up in a review are the ones that were convenient to collect, and convenience is not correlated with importance.

The available story is satisfying. "Marketing drives revenue" is coherent, actionable, and flattering to the marketing team. A story that fits is the enemy of looking for a better one.

Confirmation bias does the rest. Once a hypothesis is on the table, evidence gets sorted into supporting and ignorable. Nobody decides to do this.

Post-mortems are social events. The explanations on offer in a room are constrained by who is in it. "Bad execution" is a claim about people who may be sitting there. "There was a structural factor none of us controlled" is nobody's fault, and therefore nobody's job to raise.

Four questions that find it

"What else changed at the same time?" The most common hidden variable is timing. A competitor launched. A platform changed its algorithm. A rate rose. Your campaign did not run in a vacuum, and the calendar of everything else is a cheap thing to check.

"What determined who got the treatment?" In the store example: what decides which stores get big budgets? If the answer is "the ones we expect to do well", your data is not going to tell you whether the budget worked, because the assignment was made on the outcome.

"Does the relationship survive slicing?" This is the button in the chart above. Split by any plausible third variable — region, size, tenure, segment, month — and see whether the effect holds inside each slice. An effect that vanishes when you cut the data was probably never there.

"What would have happened anyway?" The comparison that matters is never before-and-after. It's treated-versus-what-they-would-have-done-untreated. Any argument that skips this is a story about a trend.

Back to the failed project

Take the three explanations from the top and apply the questions.

The strategy might have been fine and a competitor might have launched into the same window with four times the budget. The execution might have been fine and a key integration partner might have changed their terms in month three. The resources might have been adequate for the plan as understood, and the plan might have rested on a market assumption that quietly stopped being true.

None of those are on the whiteboard, because none of them are anybody's fault, and the list of candidate explanations was assembled from the things people could see and feel responsible for.

The question worth asking at the start of every post-mortem is the one nobody asks: what was going on that we haven't put on the list?