How vendor performance figures are produced
Ad platforms and marketing tools publish numbers about their own products: a cost per acquisition improvement from switching to an automated campaign type, a share of inventory a format is said to reach, a quality or reach uplift from a new creative feature. These come from the vendor’s own logs, measured with the vendor’s own attribution, against a comparison group the vendor chose, over a period the vendor selected.
None of that is dishonest in itself. It is a measurement taken by the party with an interest in the answer, and published because the answer was flattering. Tests that were not flattering do not get published, which is why nearly every figure of this kind points in the same direction.
Why vendor performance figures matter
They travel. A line from a platform blog is repeated in a webinar, then in an agency deck, then in a proposal, and by the time it reaches a business owner it has shed every qualifier that made it defensible. Budgets get committed on the strength of it and expectations get set by it.
They are also the main evidence offered for adopting automated products, where the advertiser hands over control of targeting, placement or bidding. The claim is doing real work in that decision, so it deserves the scrutiny any other supplier’s quote would get.
Where vendor performance figures go wrong
Selection is the biggest weakness. Advertisers who adopt a new feature early are usually the ones it already suits — larger accounts, richer data, more events — so their results describe the adopters rather than the feature. Misattribution is close behind: the uplift is measured by the same platform that decides which conversions it caused, so a change that merely shifts credit from one channel to another reads as growth.
Market fit matters most here. These figures are drawn from large, crowded auctions. A campaign in Nepal, or in any smaller market with few bidders and a modest budget, faces different competition, different creative costs and far less data, so an uplift published for the United States tells you very little about what will happen in Kathmandu.
How to act on it
Treat a vendor figure as a hypothesis about your account, not as a finding. Before repeating one, ask what it was compared against, over what window, for which kind of advertiser, and whether the groups were assigned at random. A claim that cannot answer those questions is a marketing message and should be described as one, particularly in a document a client will hold you to.
Then test it yourself. A holdout test or a properly run incrementality test settles the question for your own account in a way no published number can, and revenue your business actually recorded is the only figure a finance team will accept. Where a feature is already switched on because of a vendor claim, a Meta ads audit is a sensible way to find out whether it earned its place.
