Is Meta Retargeting Creating Sales or Claiming Them? An Incrementality Test Plan
A strong retargeting ROAS does not tell me how many sales would disappear without those ads. The audience already showed interest. Some people may buy through email, direct traffic or another route anyway. I would separate the platform's attribution from the additional contribution the campaign creates before using that ROAS to justify more budget.
The useful question is specific: what changes when eligible people can receive this retargeting campaign compared with a comparable group held out of it? Define the campaign, audience and outcome first. Then choose a test that can actually answer that question.
Check which experiment you can run
Meta's current Conversion Lift documentation describes randomized test and control groups and says the self-serve option is available if you meet its criteria. Check availability and feasibility for the actual account. Spending $15,000 a month does not by itself establish eligibility or enough conversion volume.
A supported lift study can hold people out of the ads being tested. A custom customer-level holdout requires reliable random assignment, identity matching and enforceable exclusions. A geographic design requires enough comparable regions, appropriate analysis and protection against spillover. Picking one city to switch off is not automatically a valid experiment.
If you cannot maintain the holdout or detect a commercially useful effect at your volume, record that limit. A simple before-and-after pause can inform an operating decision, but seasonality, promotions, stock and other channels prevent it from isolating causation on its own.
Write the experiment contract before launch
Define eligibility at a clear point, such as the first qualifying visit during enrollment. Keep people in their assigned group. Analyze everyone assigned, rather than comparing only people who happened to see an ad with people who did not. Actual exposure can reflect auction delivery and behavior, which makes those groups different before the outcome.
Specify whether the test measures one retargeting campaign or a broader account policy. Keep other advertising and email treatment comparable across cells if they are outside the intervention. Audit duplicate audiences, account overlap, identity changes and suppression delays that could expose the control group.
Set the conversion and return windows, primary outcome, spending exposure and analysis plan in advance. Choose the sample requirement from baseline conversion, outcome variation and the smallest effect worth acting on. There is no universal number of days or conversions that makes every retargeting test decisive.
Calculate lift with the right denominator
This example is hypothetical. It assumes randomized assignment, a working holdout and complete measurement over the same window. It shows arithmetic only; no confidence interval or statistical significance is implied.
| Measure | Test group | Control group |
|---|---|---|
| Assigned eligible people | 20,000 | 10,000 |
| Purchasers | 600 | 250 |
| Purchaser rate | 3.0% | 2.5% |
| Retargeting spend | $2,000 | $0 |
The absolute rate difference is 0.5 percentage points. Relative lift is 20%: 3.0% divided by 2.5%, minus one. Applying the control rate to the 20,000-person test group gives 500 expected purchasers without the intervention. The estimated difference is 100 purchasers, not the raw 350-person difference between unequal groups.
Assume one order per purchaser, $100 net revenue and $35 contribution before ads per order in both cells. Incremental revenue is $10,000, giving a point estimate of 5 times incremental revenue divided by the $2,000 spend. Incremental contribution before ads is $3,500; after the tested spend it is $1,500, before other experiment costs and fixed overhead.
Real order values and margins vary. Calculate outcomes from the observed data rather than forcing the same average on both cells. Report uncertainty around the estimate and check whether plausible values change the budget decision.
Read the result without overstating it
A flat, uncertain result does not prove the campaign has no effect. It may mean the test cannot distinguish the effects that matter. A positive result also does not establish that doubling spend will preserve the same marginal return.
Check assignment balance, missing outcomes, contamination and mid-test changes before explaining the result. Compare incremental contribution with the cost and operational risk of the next budget move. Use the new-customer economics guide and discount experiment when customer mix or offers complicate the readout.
Bring the question to the account review
I managed $544,397.42 in BFCM Meta spend for one US jewelry brand. At that level, each budget decision needs a clear reason. If your brand already spends at least $15,000 a month on Meta, see my ecommerce work and schedule a call. Bring the retargeting setup, order economics and the decision you need to make.