If your Google Ads dashboard reports 400 conversions last month while your CRM shows 120 new opportunities, you probably already suspect the platforms are taking credit for sales that would have happened anyway. Incrementality testing settles that question by measuring how many sales your ads actually caused. It compares a group that saw the ads with a similar group that didn’t. This article covers how lift tests work, which test design fits your budget, how to run your first one, and the mistakes that quietly ruin the results.
The benefit is simple. Once you know which spend is incremental, you can move budget away from campaigns that only look profitable and toward the ones that actually grow your pipeline.
Why Attribution Reports Can’t Prove Causation
Attribution assigns credit. It doesn’t prove cause. Last-click and data-driven models in Google Ads, Meta Ads Manager or HubSpot look at who converted after clicking or viewing an ad and divide credit along that path. They never ask what those buyers would have done if the ad hadn’t run.
Branded search is the classic example. A prospect hears about your product from a colleague, searches your company name, clicks the paid result above your organic listing and books a demo. Google records a conversion. If you removed that ad, most of those people would simply click the organic link right below it. Retargeting has the same problem. It mostly reaches people who already visited your pricing page, and that group converts at a high rate with or without ads.
Platforms also use different lookback windows and count view-through conversions differently, which is why their numbers rarely match. That problem is explained in more detail in why ad reports disagree across attribution windows. Incrementality testing avoids the argument over credit and asks one question: what changed because the ads ran?
How Incrementality Testing Works
Every lift test has the same basic structure. You split a population into a test group that can see ads and a control group that can’t. You run the campaign and compare the conversion rates of the two groups. The difference between them is the incremental lift.
Here is a worked example. A B2B SaaS company spends $20,000 a month on Meta, and the platform reports 400 conversions, so the CPA is $50. In a lift test, the exposed group converts at 2.4% and the holdout group at 1.9%. That means only about 21% of the test group’s conversions were caused by the ads (0.5 divided by 2.4). That works out to roughly 84 incremental conversions, which puts the incremental CPA at about $240. Nobody approves the same budget once they see that number.
There are three common test designs. User-level holdouts, such as Meta Conversion Lift and Google Ads Conversion Lift, randomly hide ads from part of your audience. Randomization is clean, but you depend on the platform’s own measurement, and eligibility usually requires enough spend and conversion volume. Depending on the account, setup often goes through a platform rep.
Geo experiments turn ads on or off in matched regions, such as US DMAs or states, and measure the results in your own CRM. Meta’s open-source GeoLift package for R and Google’s CausalImpact library are the usual tools. Geo tests work for any channel, including LinkedIn, CTV and podcasts, where user-level holdouts are limited or unavailable.
Time-based on/off tests, where you pause a channel for three weeks and watch what happens, are the cheapest and the least reliable. Seasonality, sales team activity and competitor moves all get mixed into the result.
Step-by-Step: Running Your First Geo Lift Test
An experienced growth analyst runs a power analysis before asking for approval, not after the test ends. That is the main difference between a test that produces a decision and one that produces an argument.
1. Ask one narrow question. “Does our non-brand LinkedIn prospecting create pipeline?” works. “Do our ads work?” does not.
2. Measure outcomes in the CRM, not the ad platform. Use SQLs or opportunities from Salesforce or HubSpot. This requires clean region data on every lead.
3. Match regions using 12-24 months of history. GeoLift’s pre-test analysis suggests which regions to use and shows the minimum detectable effect for your budget.
4. Set the duration to match your sales cycle. Most B2B tests need 4-6 weeks of exposure. If your deals close in 60-90 days, measure SQLs as a leading indicator and check opportunities again 90 days later.
5. Freeze everything else. Don’t change budgets, swap creative, adjust pricing or launch a regional sales push in test regions during the test.
Here is a realistic scenario. A marketing team spending $40,000 a month across Meta and LinkedIn is asked by the CFO to justify the budget. They turn off LinkedIn in eight matched states for five weeks. SQLs in those states drop 18% compared with the forecast. That’s enough to keep LinkedIn funded. They run the same test on Meta retargeting, and SQLs barely move. That budget gets cut by half, and the savings go to prospecting.
The ROAS Myth: High Returns Don’t Mean High Impact
Many teams assume their highest-ROAS campaigns are their best investments. In practice, it’s often the reverse. Brand search and retargeting usually show the best ROAS precisely because they reach people who were already about to buy.
Prospecting campaigns often look weak in attribution reports. That’s because they reach people early, long before the click that gets credit for the conversion. Lift tests regularly reveal that these “underperforming” campaigns create more net new pipeline than the channels at the top of the dashboard.
There’s an important exception. If competitors bid on your brand name, branded search ads protect clicks you would otherwise lose, and their incrementality can be substantial. Test it before cutting anything.
Common Mistakes That Invalidate Lift Results
The most expensive mistake is running an underpowered test. A $3,000 test that can only detect a lift above 40% will almost always come back “not significant.” Teams then read that as “ads don’t work,” when it really means the test couldn’t measure anything. Run the power analysis first and accept its budget requirements.
The second mistake is a contaminated control group. National campaigns like Performance Max, YouTube or sponsored newsletters keep running in control regions. Or SDRs keep doing outbound only in test territories. Either way, the gap between the groups reflects something other than the ads. A seasoned revenue operations manager reviews the sales team’s territory plans before the test starts.
The third mistake is treating one result as permanent truth. Audiences change, creative wears out and competitors enter the market. A lift result from Q1 2025 says little about your account in Q4 2026. Retest your largest channels every 6-12 months.
When the Standard Approach Doesn’t Fit
Teams spending under $10,000 a month, or getting fewer than 50 conversions a month, rarely have enough signal for a reliable test. In that case, start by tracking blended CAC and running simple pause tests on obvious suspects like brand search.
ABM programs targeting 200 named accounts can’t use user-level randomization. Split the account list into matched halves instead and compare pipeline creation between them. Small markets with few regions, such as Finland or the Baltics, have too few geographic units for geo tests, so time-based designs with careful seasonality adjustments are often the only option. Regulated industries like healthcare and finance may restrict the data needed for user-level matching altogether.
FAQ
How long should an incrementality test run?
Most B2B tests need 4-6 weeks of exposure, followed by an observation period that matches your sales cycle. Tests shorter than two weeks rarely produce enough conversions to support a decision.
How much budget does incrementality testing require?
That depends on your conversion volume and the size of the effect you want to detect. Mid-market B2B teams typically need $15,000-50,000 in spend during the test. A power analysis in GeoLift gives you the exact number before you commit.
Is incrementality testing the same as marketing mix modeling?
No. MMM tools like Google’s Meridian or Meta’s Robyn estimate channel contribution from historical data, while lift tests measure it directly through controlled experiments. Mature teams use lift test results to calibrate their MMM.
Summary: Test the Spend You Trust Most
Incrementality testing replaces the question of which channel gets credit with the question of what actually caused the sale. Start with your highest-ROAS channel, usually brand search or retargeting, because that’s where the gap between reported and real impact tends to be largest. Measure the result in the CRM, run a power analysis first and rerun tests regularly. A single well-designed test often pays for itself by freeing up budget for campaigns that actually grow the pipeline.