How to check the numbers in their pitch
Four questions that turn any performance claim into either a real number or an admission. You do not need to know anything about advertising to ask them.
The four tests
Every claim on this ledger was checked against these, and they are the whole of the claim hygiene criterion. Run them on whatever is in front of you.
One: over what attribution window?
Meta decides which conversions to credit to an ad using a window: seven days after a click, one day after a view, or some combination. The same campaign reports wildly different results depending on which is used, and the wider the window the better it looks.
A cost per acquisition quoted without its window is not a number, it is a rumour. If nobody in the room knows the answer, that is the answer.
Two: divided by what?
A click-through rate can count every click on the ad, including likes, shares, comments and taps on the page name. Engagement-heavy creative posts spectacular meaningless numbers this way, and a reported thirty per cent click-through is almost always this rather than intent.
The number that describes people who reached your website is the link click. Ask which one they are quoting. Ask the same of any cost figure: cost per what, exactly. Cost per lead and cost per booked job differ by an order of magnitude and only one of them is money.
Three: over what period, across how many accounts?
A seven times return is a different claim if it describes one client for one month than if it describes thirty accounts over a year. Both are quoted the same way in a deck.
One month of one account is an anecdote. It may be a true anecdote and it is still not a basis for predicting your result.
Four: is it platform-attributed, and do they say so?
Meta reports what Meta believes it caused. Presenting that as revenue or as customers is the most common overstatement in this industry, and it is usually made without any intention to mislead, because the dashboard says it.
The strongest answer an agency can give here is a plain one: these are platform-attributed figures, they are not typical, and here is what our clients actually booked. One agency on this ledger publishes a disclaimer to that effect, and it is the reason they score highest on this criterion.
What good answers sound like
Nobody should expect an agency to have all four to hand for every claim in a deck. What separates a good answer from a bad one is whether they know which they can support. "Seven-day click, one client, four months, and it is Meta-attributed so treat it as directional" is a better answer than a bigger number with nothing behind it.