How to use Conversion Lift Test
A practical guide to platform-run conversion lift tests — how Meta and Google randomise users into exposed and control groups, what minimum spend you need, and when geo holdouts are the better call.
Conversion Lift Test
A platform-run experiment that randomises users into exposed and control groups to measure the causal lift of paid ads.
A conversion lift test is an incrementality experiment run inside an ad platform — most commonly Meta Conversion Lift or Google Conversion Lift — where the platform splits your audience into a treatment group that sees your ads and a holdout group that doesn't. By comparing conversion rates between the two groups, you measure the true causal impact of the campaign rather than the inflated last-click numbers your ads manager reports.
Unlike attribution-based reporting, a conversion lift test answers one question: how many of those purchases would have happened anyway? The output is a lift percentage and an incremental cost per acquisition, which is the number worth optimising against.
Most paid-acquisition teams discover the same uncomfortable thing within a quarter of running their first lift test: somewhere between 20% and 50% of the conversions credited to Meta or Google would have happened without the ad. That gap is the difference between reported ROAS and incremental ROAS, and it's the number that decides whether a channel actually deserves the next €10k.
Conversion lift tests sit inside the broader practice of incrementality measurement. They're the easiest format to spin up because the platform handles randomisation, exposure tracking, and the statistical readout — you just commit budget and wait. The tradeoff is that you're trusting the platform to grade its own homework.
How a conversion lift test actually works
When you launch a lift study, the platform takes your target audience and splits it — usually 50/50, sometimes 90/10 — into a treatment cell and a control cell. Treatment users are eligible to see your campaign. Control users are explicitly excluded; the platform suppresses your ads from them for the full test window, even if they would have qualified.
The platform then tracks conversions in both groups using its own pixel or conversions API data. After the test window closes — typically 2 to 4 weeks — you get a readout: conversion rate in treatment, conversion rate in control, the absolute lift, the percentage lift, and a confidence interval. If the confidence interval crosses zero, the test was underpowered and you can't conclude anything.
On Meta, this runs through the Experiments tool and requires a Conversions API connection plus a sufficiently large audience. On Google, Conversion Lift is gated to managed accounts and tied to specific campaign types (YouTube, Display, and some Search variants). Both platforms require you to pre-declare the conversion event you're measuring against before the test starts.
The control group isn't free
A 10% control holdout on a campaign spending €50k/month means roughly €5k of revenue you would have generated is intentionally suppressed for the test window. That's the price of clean data. Budget for it before launching, and don't run lift tests during peak season unless you've decided the learning is worth missing the wave.
Minimum spend and statistical power
The single biggest reason lift tests fail is underpowering. To detect a 10% lift with 80% power and 90% confidence, you typically need 4,000 to 8,000 conversions across both cells combined. If your campaign generates 200 purchases a month, a 4-week lift test produces 200 conversions — nowhere near enough to detect anything but a huge effect.
This is why Meta won't even let you launch a Conversion Lift study below a rough monthly spend floor of around €30k on the campaign under test, and why Google asks for similar volume. The chart below shows how the minimum detectable lift drops as monthly spend climbs — and why most online stores under €2M annual revenue should think hard before committing.
Minimum detectable lift by monthly campaign spend (4-week test, 50/50 split)
Read the chart this way: if you spend €25k/month on the campaign and the true incremental lift is 15%, your test will probably come back inconclusive because it can only reliably detect lifts of 24% or more. You'll burn four weeks and €25k of control-group holdout to learn nothing. Below €50k/month, lift tests are generally a poor use of budget unless you suspect the campaign is doing very little.
Platform minimums and what to expect
Each platform has its own thresholds, timing, and quirks. The table below captures the practical numbers an online retail team needs to plan around — not the official marketing copy, but what tests actually require to return a usable result.
Note that the spend thresholds are per-test, not per-account. If you want to lift-test three campaigns simultaneously, you need to clear the floor for each one independently — and each one needs its own control holdout.
Conversion lift test requirements by platform
| Platform | Min monthly spend | Test duration | Typical holdout | Conversion events needed |
|---|---|---|---|---|
| Meta Conversion Lift | €25k–€40k | 2–4 weeks | 10–50% | 4,000+ combined |
| Google Conversion Lift (YouTube) | €30k–€50k | 3–6 weeks | 10–50% | 5,000+ combined |
| Google Conversion Lift (Display) | €20k–€35k | 3–4 weeks | 10–50% | 4,000+ combined |
| TikTok Conversion Lift | €20k–€30k | 2–4 weeks | 10–30% | 3,000+ combined |
Across platforms, two patterns hold. First, the bigger the holdout, the faster you reach significance — but the higher the revenue cost during the test. A 50/50 split is the fastest read; a 90/10 split takes roughly four times longer but costs less in foregone revenue. Second, lift confidence is much higher for hard conversion events (purchase, signup) than soft events (add-to-cart, view content), because the underlying volume is smaller and noisier.
When to choose lift tests vs geo holdouts
Conversion lift tests and geo holdout testing both measure incrementality, but they answer slightly different questions. A platform lift test measures the incremental effect of a specific campaign within one ad platform. A geo holdout test measures the incremental effect of an entire channel — including organic spillover, brand search, and cross-device behaviour the platform can't see.
Use a conversion lift test when you want a fast, cheap read on whether a specific Meta or Google campaign is pulling its weight, and you have the spend to power it. Use a geo holdout when you want to measure the full channel — including all the second-order effects — or when you've stopped trusting the platform to mark its own homework. Most mature acquisition teams run both: lift tests for in-platform optimisation, geo holdouts once or twice a year as a calibration.
There's also a sequencing argument. Start with a geo holdout to establish whether the channel is incremental at all. If it is, switch to conversion lift tests to optimise within the channel. If the geo holdout comes back flat, no amount of in-platform lift testing will save the campaign — you have a strategy problem, not a measurement problem.
Rule of thumb for picking a method
Under €30k/month per campaign: skip platform lift tests, run a geo holdout on the whole channel instead. €30k–€100k/month: lift tests work, but expect inconclusive results unless the campaign's real lift is north of 15%. Over €100k/month: lift tests are the right default, and you can run them quarterly without breaking the bank.
Conversion lift test FAQ
An A/B test compares two creatives or audiences within the same campaign — both groups see ads. A conversion lift test compares an exposed group against a true control group that sees no ads at all, measuring whether the campaign drives incremental conversions versus doing nothing.
Most Meta lift studies run 2 to 4 weeks. Shorter windows rarely accumulate enough conversions to reach significance; longer windows risk seasonality and creative fatigue contaminating the result. Meta will recommend a duration based on your spend and historical conversion volume.
Generally no. Below roughly €25k/month per campaign, you won't generate enough conversions in either cell to detect a realistic lift. If your spend is smaller, a geo holdout test on the whole channel is usually a better use of measurement budget.
Almost always underpowering. Either your campaign spend is too low for the test window, your conversion event is too rare, or the true lift is smaller than the test can detect. Increase spend, lengthen the window, or pick a more frequent conversion event like add-to-cart instead of purchase.
Mechanically they're similar — both randomise users into treatment and control. Google's version is more restricted: it's only available on managed accounts and tied to YouTube and Display, while Meta's is open to most advertisers above the spend threshold. Google tests tend to need slightly more volume to reach significance.
Use conversion lift for measuring a specific campaign inside one platform; use geo holdout for measuring an entire channel including spillover effects. Many teams run a geo holdout annually as a calibration and conversion lift tests quarterly for in-platform optimisation.
It means the treatment group converted 20% more than the control group. If control converted at 2% and treatment converted at 2.4%, that's a 20% relative lift. The incremental conversions are the difference — 0.4 percentage points — multiplied by the treatment audience size.
Slightly, yes. The control holdout suppresses ads to a portion of your audience for the test window, so reported revenue dips for those weeks. The tradeoff is learning whether the campaign is actually incremental — which can save much more than the test cost if the answer is no.
Quarterly is a reasonable cadence for major campaigns above €100k/month, especially after big creative refreshes or audience changes. For smaller campaigns, once or twice a year is enough — the underlying incrementality doesn't shift that quickly.
You can, but the results don't combine cleanly. Each platform measures lift only against its own control group, so you can't sum Meta lift and Google lift to get total incremental conversions. For a cross-platform read, run a geo holdout on the combined channel mix instead.
Track CAC, channels, and funnel conversion in one place
Metricuno connects ad spend, funnel events, and revenue so you can see CAC by channel, cohort, and campaign — without stitching together five tools.