The research points one way, the pipeline points the other, and both decks were produced by competent people. This pattern explains why conflicting data recurs and how to decide when the numbers refuse to agree.
Executives are told to be data-driven, then handed two datasets driving in opposite directions. The instinct is to commission a third study. The better move is usually to understand why the first two disagree, because that disagreement is often the most informative number in the room.
Last reviewed 3 July 2026 · Free and ungated
Brief the Global Board on both sets of evidence and receive confidential input from senior operators who have carried decisions like this through a full cycle.
Pressure-test this decisionThe customer research says demand is softening. The pipeline data says the strongest quarter in three years is forming. Both were assembled carefully, both survive a first round of questioning, and each function stands behind its own. The meeting adjourns with an action to "reconcile the numbers", which over the following fortnight turns into two teams building better cases for the figures they already had.
The datasets are answering different questions and were built under different incentives. A pipeline reflects what salespeople are rewarded for entering; research reflects what respondents say, which is not the same as what they will do; finance's model reflects the assumptions someone chose when it was built years ago. Each function trusts the instrument it controls, partly because it understands that instrument's flaws and has learned to live with them. Data does not interpret itself. Every number arrives carrying the choices of whoever collected it, and those choices, not arithmetic errors, are where most conflicts live.
Interrogate provenance, not conclusions: what exactly was measured, from whom, over what period, and what incentives touched the collection. Then decompose both cases into their assumptions. Conflicts that look total usually localise to one or two variables, a conversion rate or a market definition, and a targeted test of that variable is cheaper than another full study. Ask each side in advance what evidence would change its mind. If the answer is nothing, you have learned that this is a values conflict wearing data's clothes, which is useful to know before the next meeting. Where the stakes justify it, run a small real-world trial both sides accept as decisive before it starts.
Operators who have run comparable businesses have watched these instruments perform over full cycles. They know which kind of signal predicted reality and which kind flattered the forecast, because they carried the consequences either way. Independent perspectives do not add a third dataset to argue about. They add pattern recognition about which class of evidence deserves the benefit of the doubt in this specific kind of call.
Usually behaviour beats intention, but behavioural data carries its own distortions: a pipeline can be groomed, and past behaviour misleads when conditions shift. The practical question is not which dataset is pure. It is which dataset's biases matter less for the decision in front of you.
When it targets the specific assumption where the conflict localises, and both sides agree in advance what result settles it. An open-ended third study without that agreement mostly buys delay and a third position to argue about.
Make the pick explicit. Record which dataset the decision relies on and what result would falsify it. That converts a preference into a testable position, and it protects the CEO as much as anyone: retrofitted evidence is what turns an honest misjudgement into a credibility problem later.