When to push back on a metric ask: a skill nobody explicitly teaches
Sometimes the most useful response to 'can you pull this number' is a quick clarifying question, so nobody acts on a misleading number two days later.
Sometimes a stakeholder asks for a specific number, but what they actually need is a different number, they just don't know that distinction exists yet. Pushing back isn't refusing the request, it's asking one clarifying question before doing the work.
🎯 Explain Like I'm Hired Someone asking "can you pull total signups this month" to judge a marketing campaign might be asking the wrong question without realizing it. Example: if signups include a spike of bots or low-quality users from a promo, "total signups" will make a bad campaign look good. The real question is "how many of those signups actually stuck around," which is an entirely different number to pull.
Three situations worth a clarifying question first
- The requested number doesn't match the stated goal. If the goal is "did the feature launch work" but the ask is a raw count with no comparison baseline, that number alone can't answer the real question.
- The metric is known to mislead in a specific way (inflated by bots, a tracking bug, a seasonal spike) and the person asking probably doesn't know that context yet.
- The ask implies a decision the metric can't actually support, like asking for one week of A/B test data to make a permanent pricing call, before the test has run long enough to mean anything.
How to push back without being difficult
The strong move is: deliver the number they asked for and flag the caveat, rather than refusing until the "right" question gets asked. "Here's total signups, heads up, about 15% came from a promo that historically converts much worse, want me to break that out separately?" respects their time while still preventing a bad decision.
Why this builds trust rather than friction
A stakeholder who's been burned once by acting on a misleading number quickly learns to trust whoever flags issues proactively, and quietly stops trusting whoever just delivers whatever was asked, no context, leaving them to discover the problem themselves later.
Next: The metric tree →