
Ads Manager says your prospecting campaign drove 1,000 purchases last month. Your founder asks the only question that matters: how many of those people would have bought anyway? Attribution cannot answer that. It hands out credit; it does not prove cause.
A Meta conversion lift test can. It keeps a random slice of your audience away from your ads, then compares what both groups bought. Here is how to get access, set one up, read it, and what to do if Meta says no.
A Meta conversion lift test is a randomised experiment that measures how many conversions your ads caused. Meta splits your audience into a test group that can see your ads and a control group that cannot, then compares conversions between the two.
Meta's own lift study documentation describes it this way: Meta randomises the audience, assigns each account to test or control, and calls the control share the holdout percentage. The difference in conversions between the groups is your incremental lift.
The output is a set of numbers you will not find in normal reporting:
Attribution assigns credit to ads a buyer saw or clicked; a lift test measures what changed because the ads ran. An A/B test compares two versions of an ad or setup against each other, while a lift test compares ads against no ads.
| Method | Question it answers | Has a no-ads group? |
|---|---|---|
| Attribution (7-day click, 1-day view) | Which ads touched the buyer before purchase? | No |
| Incremental attribution setting | Which conversions does Meta's model predict the ad caused? | No, it is a model |
| A/B test | Which of two versions performs better? | No |
| Conversion lift test | How many sales did the ads add? | Yes |
The gap between attribution and lift can run either way. Meta's 2025 "Suite of Truth" white paper, as reported by PPC Land, found a median 31% undervaluation of Meta under rules-based attribution across 307 studies. That is Meta grading its own channel, so treat it as one data point, not the verdict for your brand.
If your Meta and Shopify numbers already disagree, start with why Meta purchases don't match Shopify orders. A lift test does not fix tracking gaps; it inherits them.
Maybe. Meta's developer docs open with a warning that "Conversion Lift Measurement is currently limited" and tell advertisers to contact their Meta representative for access. Measurement partners can also set studies up for advertisers through the API.
In practice you have three routes:
Volume decides most of this. Haus, which sells incrementality testing, says Meta's platform studies require Meta's approval and that Meta may reject tests without enough volume. Meta's docs we could read do not publish a minimum spend. Triple Whale lists its own prerequisites for its Meta lift integration: $5,000 of campaign spend in the past year and at least 500 optimised conversions. Those are Triple Whale's numbers, not Meta's, but they show the scale vendors expect.
Set up a lift test by choosing one question, sizing the holdout, checking tracking, and freezing the plan before the start date. Meta will not let you change the core design once the study starts.
Write the question in one line before you open Experiments. Good questions for an Indian D2C brand:
Meta's white paper, per PPC Land, recommends single-cell tests to calibrate how much budget Meta deserves and multi-cell tests to compare strategies.
The holdout is the share of people who will not see your ads. Meta's API calls it control_percentage. A bigger holdout gives a clearer answer faster but costs you more sales during the test.
Triple Whale calls a 10% to 20% holdout ideal. Haus says Meta requires at least 10% of the audience per cell. If you use Meta's guided setup, Triple Whale describes four modes: optimise for speed, balanced (suggested for first tests), minimise revenue impact, and manual.
Worked example (illustrative numbers): say a campaign spends ₹10L a month and you hold out 15%. You are not cutting spend by 15%; Meta keeps spending on the test group. What you give up is the sales those held-out people would have made because of the ads. If the ads matter, that is the real cost of the test.
Lift results are only as good as the conversion data Meta receives. Meta's docs say lift objectives use Conversions API-based pixel or app events, and Madgicx recommends sending purchases through both the browser pixel and the Conversions API.
Before launch, confirm that purchases fire once per order, that checkout apps such as GoKwik or Shopflo pass the event, and that nothing changed in your theme last week. If you are unsure, run through our Meta pixel not firing on Shopify checks first.
Meta's docs are strict here. The study start time must be in the future. After the start you cannot change the start time, the treatment percentage or the objective, and you cannot remove campaigns or ad accounts from the study. You can still extend the end time and the observation end time, which keeps counting conversions after the test window closes.
Plan for 3 to 4 weeks for a Meta conversion campaign, including a short post-test window, and never less than 2 weeks. The right length depends on your buying cycle and on how many conversions you get per week.
Haus's duration guide (March 2026) gives 3 to 4 weeks as the baseline for Meta conversion tests and calls 14 days the minimum norm. In its review of 640 Meta tests (July 2025), the average ran 18.6 active days plus an 8.8-day post-treatment window. Haus's advertisers average $14M a year on Meta, so a ₹15L-a-month brand should expect to need the longer end of that range.
Three things push the test longer:
Stopping early is the classic mistake. Triple Whale warns that ending early weakens statistical confidence, and Haus says it risks false negatives.
Read three numbers together: incremental conversions with their range, cost per incremental conversion, and confidence. A big lift with a wide range and low confidence is a hint, not a decision.
Meta's results include conversions in the test group, conversions in the control group scaled to the same size, incremental conversions with lower and upper bounds, a confidence value and cost per incremental conversion. Madgicx defines the confidence figure as the probability that the ads drove more than zero incremental conversions. PPC Land's explainer treats a result as highly conclusive when more than 90% of the probability sits above zero.
Say a prospecting campaign spends ₹10L during the test.
| Metric | Value |
|---|---|
| Purchases in test group | 1,000 |
| Purchases in control, scaled to test size | 700 |
| Incremental purchases | 300 |
| Lift (300 ÷ 700) | ~43% |
| Cost per purchase, Ads Manager view (₹10L ÷ 1,000) | ₹1,000 |
| Cost per incremental purchase (₹10L ÷ 300) | ~₹3,333 |
The ₹1,000 CPA looked fine. The ₹3,333 cost per incremental purchase is what the campaign really costs you for each extra sale. Compare that with your contribution margin per order, not your ROAS target. Our guide to POAS vs ROAS shows how to set that bar.
For COD-heavy brands, one more check: a lift test counts purchases, not delivered orders. If 25% of a campaign's COD orders come back as RTO, the incremental revenue you keep is smaller than the incremental orders suggest. See RTO-adjusted ROAS for the maths.
Most failed lift tests die from changes made mid-test, too little volume, or broken tracking. The design is sound; the account around it moves.
If you cannot get a conversion lift study, use Meta's incremental attribution setting as a daily signal and run a geo or pause test for a harder answer. Neither replaces a randomised lift test, but both beat trusting 7-day click alone.
Incremental attribution is an attribution setting, not an experiment. Meta describes it as using models that predict whether a conversion was caused by an ad, and PPC Land dates the global rollout to June 2025. Jon Loomer notes that choosing it locks your attribution settings and that he has seen little difference from standard results.
Seer Interactive tested it on about $1.05M of spend in April 2025: Meta counted 87% of conversions as incremental, against 67% in GA4. Seer calls it "a directional signal, not gospel". We covered the wider attribution changes in Meta's attribution change.
A geo test turns ads off in some regions and compares their sales with similar regions where ads keep running. Meta's Marketing Science team built GeoLift, a free R package based on synthetic control methods, for cases where people-based Conversion Lift "is not feasible". Meta still recommends people-based tests where possible because they have higher statistical power.
Geo tests suit Indian brands with sales spread across many cities and pincodes, and they capture marketplace and offline sales that a Meta lift test misses. At the e4m D2C Summit in September 2026, Truelift's founder described a six-week test that paused Meta and quick commerce ads to see what each drove. The cost is the same as any holdout: sales you may give up in the paused regions.
Clevrr AI does not run lift tests. It helps you keep a test clean and read it against the rest of your business.
Klev, Clevrr's AI agent, watches your Meta and Google accounts every morning, so it flags the edits, stock-outs and tracking breaks that quietly ruin a test while it runs. Clevrr also joins Meta data with Shopify, Amazon and Flipkart sales, so you can see whether a holdout moved marketplace orders as well as website orders. Quick commerce and Myntra are not connected today.
Yes, if your account has the volume. A Meta conversion lift test is the cleanest way to learn how many sales your ads add, and what each extra sale costs in rupees. Check access first, size the holdout, fix tracking, freeze the plan and give it 3 to 4 weeks. If Meta will not grant a study, use incremental attribution for daily direction and a geo test for the big budget calls.
Want the leaks found before the test starts? Start free with 500 credits.
No. Madgicx, citing Meta's Experiments page, says there is no additional cost to create a test. The real cost is opportunity: the people in your holdout do not see your ads, so you give up any sales those ads would have caused. A larger holdout gives a faster, clearer answer but raises that hidden cost.
A conversion lift test compares ads against no ads; an A/B test compares two versions against each other. Use an A/B test to pick a better creative, audience or bid strategy. Use a lift test to learn whether a campaign adds sales at all, and what each extra sale costs you. Multi-cell lift tests can compare strategies too.
The confidence number is the probability that your ads drove more than zero incremental conversions. It does not tell you how big the lift is. Read it with the range of incremental conversions: a 95% confidence with a narrow range is a decision, while a 70% confidence with a wide range means you need more volume or more time.
Not always. In Haus's July 2025 review of 640 Meta tests, 58% of brands got higher incremental ROAS from manual campaigns than from Advantage+ in head-to-head tests, while Advantage+ still won for many others. Haus sells testing and its advertisers are large, so treat this as a reason to test your own account, not a rule.
Join hundreds of D2C brands using Clevrr AI to automate their growth and efficiency.
Join 500+ D2C founders using Clevrr AI to automate their operations and scale faster.
Get Started Free