How a conversions lift study works
Ordinary reporting answers a narrow question: of the people who converted, how many had seen or clicked an ad? It cannot tell you whether they would have converted anyway. A conversions lift study answers that harder question by taking the audience you would have advertised to and splitting it in two. One group sees the campaign. The other, the holdout, is deliberately kept ad-free for the duration.
Both groups are then measured for the same outcome — purchases, leads, whatever you configured. The difference between them is the lift: conversions that happened because of the advertising rather than alongside it. Meta runs the split and the reporting itself, so the advertiser does not choose who lands in which group.
The study runs for a fixed period and reports once. It is not a dashboard you watch daily, and it cannot be judged part-way through without undermining the result.
Why lift studies matter
Attributed conversions and caused conversions are different quantities, and the gap between them is largest exactly where advertisers spend most confidently. Retargeting is the clearest case: an audience already close to buying will convert at a flattering rate whether or not you advertise to them, so platform reporting makes that budget look excellent while lift may show far less.
Brand campaigns, broad prospecting and any spend a client is sceptical about are the other candidates. A lift result is also the only measurement argument that survives a finance director asking what would have happened if you had spent nothing.
Where lift studies go wrong
The first problem is eligibility. Meta sets minimum requirements around audience size, budget and conversion volume, and small accounts frequently cannot run a valid test. A study that runs anyway on thin volume produces a result too uncertain to act on, which is worse than no study, because people act on it regardless.
The second is contamination. Running other campaigns to the same audience during the study, changing budgets part-way, or launching a promotion mid-test all leak into the holdout group and blur the comparison.
The third is over-reading a single test. A lift result describes that audience, that creative, that period and that market. It is evidence, not a permanent constant, and treating it as a fixed property of the channel leads to confident mistakes later.
How to use one properly
Decide before you start what result would change your behaviour, and write it down. A test whose outcome would not alter a single decision is not worth the sales you deliberately give up in the holdout — because that is the real cost of a lift study, and it should be acknowledged rather than glossed over.
Freeze the account for the test period: no budget changes, no new campaigns to the same audience, no promotions you did not plan for. Make sure the conversion being measured is reliably recorded first, since a lift study inherits every weakness in your pixel and Conversions API setup. Then run one test on the question that matters most, act on it, and consider a geographic holdout test if you need an answer the platform is not able to give.