A big budget does not rescue a poor fit. The quickest way to lose leverage is to accept work you should have declined at the first call, especially when the budget is large enough to make everyone polite about the warning signs.
The proposed engagement looked attractive on paper: a clear commercial sponsor, a visible operational problem, and enough runway to build something meaningful. The problem was ownership. Nobody in the client team could say who would run the system after hand-over.
That matters because a data platform is not a package delivery. It is a decision system. Someone has to own definitions, exceptions, permissions, support routes, and the dull rituals that keep the numbers trustworthy after launch.
We tested the fit with three questions. Who signs off metric definitions? Who can stop a release when data quality fails? Who explains the system to a new analyst six months later? The answers moved between departments with each conversation.
The right commercial answer was to sell a discovery phase and hope the organisation caught up. The right delivery answer was to stop. We wrote a short refusal note, named the missing operating decisions, and suggested a smaller internal exercise before bringing in builders.
That answer bought us trust rather than losing it. Refusal is not posture when it is specific. It should leave the other side with a clearer map of what must be true before the work becomes worth buying.
The useful lesson is not that agencies should say no more often. It is that scoping should produce evidence for both sides. If the evidence says the project is not ready, pretending otherwise is just a slower and more expensive refusal.