Incrementality testing measures how many sales your ads caused, not how many they got credit for. You split people (or regions) into a test group that can see your ads and a holdout group that can't, then compare their results; the difference is the lift. On Meta you can run a Conversion Lift test in Experiments, a geo test with Meta's open-source GeoLift, or, for a small store, a simpler test that turns ads off in a few regions.
Why isn't attribution enough?
Ads Manager tells you which conversions happened after someone saw or clicked an ad. It does not tell you whether those people would have bought anyway. Retargeting is the classic case: it reaches people who were already about to buy, so its reported results can look much better than the value it adds.
| Attribution (Ads Manager) | Incrementality (lift test) | |
|---|---|---|
| Question | Which ads were seen or clicked before a sale? | Which sales would not have happened without the ads? |
| Method | Rules and attribution windows (for example 7-day click) | A randomized holdout group or matched regions |
| Good for | Daily optimization, comparing ads | Budget decisions, channel and campaign value |
| Weak point | Gives credit for sales that would happen anyway | Needs time, volume and a clean setup |
Meta says Conversion Lift results are not meant to be compared with Ads Manager results. Lift tests count all conversions in both groups during the test dates, while Ads Manager counts conversions attributed to specific ads within a window.
How do you calculate lift?
The math is simple. Make sure both groups are the same size first (Meta scales the holdout group to match the test group).
- Incremental conversions = conversions in the test group − conversions in the holdout group
- Lift % = incremental conversions ÷ holdout conversions × 100
- Incremental ROAS = revenue from incremental conversions ÷ ad spend
- Cost per incremental conversion = ad spend ÷ incremental conversions
An example with made-up numbers, only to show the math: the test group makes 300 purchases and the same-size holdout group makes 240. Incremental purchases = 60. Lift = 60 ÷ 240 = 25%. If the ads cost $3,000, each incremental purchase cost $50, even if Ads Manager credited the ads with all 300.
That last number is the one that matters for budget. Compare it with your break-even ROAS and margin, not with the reported cost per purchase.
Which test design should you use?
| Design | How it works | Good for | Watch out for |
|---|---|---|---|
| Meta Conversion Lift | Meta randomly splits your audience into test and holdout; the holdout can't see your ads | Measuring a Meta account, campaign or ad set | Needs enough spend and conversions |
| Geo test with GeoLift | Ads run in some regions and stop in matched regions; a model builds the "no ads" baseline | When you can't hold out people, or want to measure total sales across channels | Needs enough regions and stable sales history |
| Simple region on/off test | You pause ads in a few regions for the whole test and compare your store's sales there with similar regions | Small stores without the budget for the above | Regions are never perfect matches; read it as a rough signal |
| "Go dark" period | You pause all ads for a time and compare with before | Last resort | Season, promotions and stock changes can hide or fake the effect |
| A/B test | Two strategies, both groups see ads | Choosing between creatives, audiences or placements | Not incrementality: there is no group without ads |
Meta Conversion Lift
Meta describes a lift test like this: it splits your campaign's intended audience randomly into test and holdout groups, people stay in their group for the whole test, the holdout is withheld from your ads, and Meta compares the actions of both groups to calculate lift and a confidence percentage. Meta also points out that a lift test is not an A/B test, because an A/B test has no holdout group.
To set one up, go to Ads Manager > All tools > Experiments > Create test > Conversion lift (Meta's steps). The button may be disabled if you don't meet the spend or conversion requirements. Meta recommends a minimum budget of $5,000 and at least 28 days to make a significant result more likely.
Geo tests with GeoLift
GeoLift is a free, open-source R package from Meta's team. It helps you pick test and control regions, checks if your test has enough power, and measures lift with synthetic control methods. Its best practices say:
- Use daily data where you can; it is strongly recommended over weekly.
- Have at least 25 pre-test periods for 20 or more regions.
- Have 4 to 5 times the test length in stable history before the test, and ideally the last 52 weeks.
- Run for at least 15 days with daily data (4 to 6 weeks with weekly data), and cover at least one purchase cycle.
GeoLift works on the data you give it for each region, such as your store's total daily sales. That way it also counts sales that arrive through other channels after someone saw your ads, which a Meta-only test can't see.
How do you plan an incrementality test, step by step?
- Write one question tied to a decision. "Does our Meta prospecting add new sales, or would we get them anyway?" A test that can't change a budget isn't worth running.
- Pick one main metric before launch. Purchases, revenue, new customers or profit. Don't switch after you see the data.
- Choose the design from the table above, based on your budget and how you can split the audience.
- Check size and length. Use Meta's recommendations or GeoLift's power tools. If you don't have the volume, narrow the question (one campaign, not the whole account) or measure a more frequent event.
- Protect the holdout. Don't send the holdout regions special emails, discounts or influencer pushes during the test.
- Freeze other changes. No new offers, price changes or big creative swaps in the middle of the test.
- Set the decision rule in advance. For example: "If the cost per incremental purchase is below our break-even, we keep the budget; if not, we cut it by half."
- Name an owner and a date for the readout.
How do you read the results?
- Look at lift and confidence together. A high lift with low confidence is a hint, not an answer.
- Use the cost per incremental conversion for budget decisions, not the reported cost per result.
- Read a flat result carefully. Meta explains that "similar performance" means it found no clear difference between the groups for that event. That is still a finding: the setup you tested did not change behavior. If the test was too small or contaminated, though, the result is just unclear.
- Follow the buyers after the test. If your store sells products people buy again, compare repeat purchases of new customers over the next weeks. A campaign with modest lift can still bring valuable customers.
- Run it again when your offer, creative or season changes a lot. One test describes one period.
Common mistakes and how to avoid them
| Mistake | What happens | Fix |
|---|---|---|
| Changing the offer during the test | You measure the offer, not the ads | Freeze offers and prices |
| Holdout regions get other promotions | Holdout sales go up, lift looks smaller | Keep other channels the same in all regions |
| Test too small or too short | Result is unclear | Narrow the question or extend the test |
| Comparing lift results with Ads Manager | Numbers never match | Use each tool for its own job |
| Testing by turning ads on and off by hand and calling it an A/B test | Overlapping audiences, unreliable results (Meta advises against it) | Use Meta's A/B or lift tools |
| No decision rule | Endless debate after the test | Write the rule before launch |
How to do this with SearchTheTrend
SearchTheTrend can't measure your lift: only your ad account and your store data can. It helps you choose what goes into a test, so the test answers a question worth asking.
- In Ads, look at how long competitors keep each ad running and how its Ad Rank moves day by day. Ads a brand keeps funding for weeks are good candidates for the creative in your test cell.
- Open a competitor's advertiser profile to see how many new ads it launches and which ones it keeps live, then decide if you test a new angle or your current best ad.
- Media buyers save the shortlist to a board and share it with the team (for media buyers).
For the rest of your measurement setup, read our guides on attribution models and tracking Facebook ad performance.
FAQ
What is incrementality testing in simple words?
It measures the sales your ads caused, not the sales they got credit for. You compare a group that could see your ads with a similar group that could not; the difference between the two is the lift.
How is incrementality different from attribution?
Attribution gives credit to ads that people saw or clicked before buying, even if they would have bought anyway. Incrementality uses a holdout group to show how many of those sales would not have happened without the ads.
How much budget do I need for a Meta Conversion Lift test?
Meta recommends at least $5,000 and at least 28 days to make a clear result more likely, and the test is only open to advertisers who meet its minimum spend. Smaller stores can run a region test instead.
Is an A/B test the same as an incrementality test?
No. An A/B test compares two ad strategies, and both groups still see ads. A lift test keeps a holdout group that sees none of your ads, so it can measure what the ads add.
What does it mean if my test shows no lift?
Either the ads did not change behavior for that event, or the test was too small, too short or contaminated. Check the setup first; if it was clean, treat the result as real and move budget or test a different campaign.
