Evidence or it didn't happen
Agencies, platforms, consultants and AI tools all give marketing recommendations. Here's the five-question test any of them should survive before you act.
Marketing advice is abundant. Your ad platform recommends raising budgets. Your agency recommends new creative. A tool recommends restructuring campaigns. Some of this advice is excellent. Some of it is confident noise. The skill is telling them apart, and it doesn't require being a marketer. It requires asking for evidence.
Five questions, in order:
1. What exactly changed, with numbers? A real recommendation starts from an observed fact: this campaign's cost per lead rose from X to Y over these dates; this search term spent this much with nothing to show. If the opening claim has no figures ("performance is softening," "there's headroom here"), stop. Vague premises produce unfalsifiable advice.
2. Where do those numbers come from? Which system, which date range, which definition of a conversion? Advice built on a platform's own generous counting deserves different scrutiny than advice built on your revenue data. You don't need to audit every figure: you need the source named, so that someone could.
3. Why does the recommender benefit? Not an accusation: a calibration. A platform recommending higher budgets earns from the spend. An agency recommending more scope earns from the scope. An honest recommender survives this question comfortably; a salesman changes the subject.
4. What is the expected result, stated before you act? "Do this" without "and here's what should happen" is unaccountable by design. Insist on the expectation in writing (even a rough one), because it converts advice into a testable claim. The recommenders worth keeping are the ones willing to be checked.
5. What happens if you do nothing? The cost of inaction is the honest price tag of any recommendation. Sometimes it's real and urgent: a broken tracking tag corrupts every decision downstream. Sometimes it's nothing, and the recommendation is motion for its own sake.
Advice that survives all five questions is worth taking seriously, whoever gave it. Advice that fails question one isn't advice yet.
We built OmniCommand to be held to this exact standard: every recommendation it makes states what changed with the figures, names the source system, shows the evidence behind the finding, states the expected impact, and tells you the risk of acting, because a recommendation that can't answer those questions shouldn't be acted on, including ours.
Request early access at omnicommand.ai.