Part IV
What counts as evidence in strategy?
Strategy rarely suffers from a complete absence of information. It suffers from different kinds of information being treated as though they carry the same weight.
A sales figure, a client's belief, a focus-group comment, an analyst's interpretation and a synthetic audience response can all appear in the same deck. By the time they reach the recommendation, their origins have often disappeared. The argument looks solid because every sentence has been smoothed into the same confident voice.
Good strategy needs a more explicit account of what we know, what we infer and what we are currently assuming.
Facts still need interpretation
A fact is something supported directly by a source: revenue fell in a particular period, aided awareness measured at a certain level, or a competitor launched at a recorded price. Facts can still be wrong because sources and measurements can be wrong, but at least the claim can be traced and checked.
Facts do not explain themselves. A decline in sales may reflect distribution, price, competition, seasonality, creative wear-out or something else. The explanation is an interpretation. It may be excellent and strongly supported, but it should not quietly inherit the status of the number it explains.
This distinction protects the work from false certainty. It also makes disagreement more useful. People can agree on the fact and challenge the interpretation without pretending one of them has rejected reality.
Hypotheses are working explanations
A hypothesis is an explanation or prediction that deserves testing. Strategy depends on them because the evidence is rarely complete enough to remove uncertainty.
The problem starts when a plausible hypothesis survives several presentations and becomes client truth through repetition. “Younger buyers find us old-fashioned” may begin as a sensible explanation. Six months later, nobody remembers that it was never directly established.
Hypotheses need a visible status and a route to validation. What evidence would support or weaken this one? Does the decision depend on resolving it, or can the team act while uncertainty remains?
Client belief is evidence about the client
Clients know their business, history and organisation in ways an outside strategist cannot. Their beliefs deserve serious attention. They are still beliefs.
“Our customers value service above price” might be supported by data, frontline experience or years of customer conversations. It might also be the story the organisation prefers to tell about itself. Either way, the statement is evidence about how the client sees the situation. It becomes evidence about customers when something beyond internal conviction supports it.
This does not require correcting every client opinion in the room. Some beliefs are useful constraints even when their empirical status is uncertain. The important thing is to know which kind of constraint you are working with.
Different evidence answers different questions
Quantitative evidence can show scale, frequency, association and change. Qualitative evidence can reveal language, meaning, tension and possible explanations. Behavioural data records what happened under particular conditions. None deserves automatic supremacy.
A large sample does not rescue a badly framed question. A vivid interview quote does not establish prevalence. Behaviour does not always reveal motive, and stated motive does not reliably predict behaviour. The strategic job is to understand what each source can support.
Strong evidence often comes from convergence. Different methods, with different weaknesses, point toward the same conclusion. Disagreement between sources can be even more valuable because it exposes where the diagnosis is too simple.
Model inference needs its own status
An LLM can interpret evidence, identify patterns and generate hypotheses. A synthetic audience can produce useful reactions. These outputs are model inferences. Their fluency does not turn them into observed human evidence.
They may be grounded in strong source material and validated against known outcomes. They may also reflect patterns in training data that fit the prompt while missing the reality of the audience in question. The status should travel with the output so later work does not cite a simulation as though people were interviewed.
Provenance makes strategy easier to challenge
I want a recommendation to show enough of its foundations for somebody to ask a precise question. Which fact supports this? Where did the interpretation enter? What assumption would change the answer? Which conclusion did a person approve?
This does not mean turning every strategy into an academic paper. It means preserving the joins that matter. Evidence gives strategy a basis. Interpretation gives it meaning. Hypotheses let it move before certainty arrives. Keeping them distinct makes the final choice more honest and usually more robust.