Every Google Ads account manager has been in a meeting where someone asks a version of the same question: if we paused this campaign tomorrow, would anything actually change? Attribution reports say yes. Incrementality testing is the only way to find out if that is actually true.
Attribution tells you which touchpoints appeared in the path to conversion. Incrementality tells you which touchpoints caused the conversion. For campaigns running at meaningful spend, those two answers can be very different and the gap between them is where budget decisions go wrong.
Why attribution is not the same as incrementality
Data-driven attribution and last-click attribution both assign credit to touchpoints that appeared before a conversion. Neither tells you whether removing those touchpoints would have changed the outcome. A brand search ad that shows up after a user has already made their decision gets attribution credit even if the conversion was inevitable. A prospecting campaign that introduced the brand three weeks ago might get little credit even though it was the reason the user is converting at all.
Incrementality tests create a counterfactual: what would have happened to this group of users or geography if the ads had not run? The difference between the control group and the exposed group is the true incremental lift.
What incrementality means in practical marketing terms
If your Google Ads campaigns report 200 conversions in a month and an incrementality test shows 60 percent true lift, the campaigns actually drove approximately 120 incremental conversions. The other 80 would have happened anyway through organic search, direct, word of mouth, or other channels. The effective cost-per-acquisition for the incremental conversions is therefore higher than the reported number by a significant margin.
This does not mean the campaigns are not worth running. It means the budget should be calibrated to the incremental contribution, not the reported attribution total.
When paid search should be tested for true lift
- –Brand search campaigns with strong organic presence — these frequently show low true incrementality
- –Accounts where ROAS or CPL looks strong but downstream lead quality or revenue does not match
- –Accounts scaling budget based primarily on last-click or platform attribution reports
- –Campaigns targeting existing customers or CRM lists where organic re-engagement may be high
- –Any campaign where a budget cut question is being seriously considered
When paid social should be tested differently
Meta and other paid social platforms have their own lift testing tools built in. These use audience holdouts within the platform and are generally the most accurate method for social-specific incrementality. However, they only measure lift within the platform's reported outcomes. An independent geo holdout test captures cross-channel effects that platform-native tests miss.
Four common test designs
Geo holdouts
Select a set of comparable geographic markets and suppress ads in the holdout markets while running normally in the exposed markets. After the test period, compare conversion rates between groups, controlling for known differences in baseline performance. This method works well for campaigns with clear geographic targeting and sufficient volume in each region.
Audience holdouts
Within Google Ads campaign settings, configure a percentage of the target audience to be held out from seeing ads. This method is less precise than geo holdouts because the holdout group may still encounter the ads through brand search or non-holdout channels, but it avoids the geographic confounds that can affect geo tests.
Budget suppression tests
Reduce campaign budget significantly for a defined period in a specific channel or geography and measure the conversion impact. Less methodologically clean than a full holdout, but useful for accounts where a complete suppression is operationally impossible or where you need a quick directional read.
Brand term reduction tests
Specifically for brand search campaigns, pause or significantly reduce brand keyword bids for a period and measure whether organic brand search picks up the volume. If organic brand click volume increases proportionally to the paid brand reduction, true incrementality is low. If organic does not absorb the volume, the paid brand campaign has higher true lift.
How to choose the right KPI
Qualified lead volume
For lead generation accounts, raw form fills are a poor incrementality KPI because they include unqualified submissions. Use SQL volume, opportunity creation, or closed business as the outcome metric where the data pipeline allows for it. This requires CRM integration and a longer test window.
Revenue contribution
For ecommerce accounts, revenue is generally the right KPI, but margin contribution is better if product mix varies significantly across geographies. A test that shows 70 percent lift on revenue but the holdout region primarily buys high-margin products may overstate the incremental profit contribution.
CAC / MER / contribution margin
For accounts that track blended efficiency metrics, measure the test through the lens of incremental customer acquisition cost and its impact on marketing efficiency ratio. These metrics contextualize the lift number within the broader business model.
Common ways incrementality tests fail
| Failure mode | What it looks like | How to prevent it |
|---|---|---|
| Test period too short | Results appear significant but reverse after the test ends | Run for minimum 2 weeks; 4+ weeks for low-volume accounts |
| Non-comparable control regions | Holdout markets have different seasonality or competitive dynamics | Match markets on historical conversion rate and volume |
| Budget reallocation during test | Budget moves into holdout regions mid-test | Lock campaign geo targeting for the full test duration |
| Conversion window mismatch | Leads from during the test convert after the test ends | Extend measurement window to include the full conversion lag |
| Insufficient holdout size | Too small to detect real differences statistically | Holdout should represent at least 20–30% of comparable volume |
Lead generation example
A B2B software company running non-brand search campaigns tests incrementality using a geo holdout across six comparable US metro markets. Three markets suppress ads for four weeks. Result: the suppressed markets show a 25 percent reduction in demo requests compared to the exposed markets, after controlling for baseline differences. True incrementality is approximately 65 percent. The CPL for incremental demos is 54 percent higher than the reported CPL. Budget allocation is adjusted accordingly.
Ecommerce example
A direct-to-consumer brand runs a brand search campaign and an incrementality test using the brand term reduction method. Over three weeks, brand keyword bids are reduced by 80 percent. Organic branded search volume increases by 60 percent, absorbing most of the suppressed paid volume. Incremental revenue from the brand campaign is found to be approximately 20 percent of reported revenue. Brand search budget is reduced and reallocated to non-brand prospecting.
What to do with the results operationally
- –Recalculate effective CPA and ROAS using the incremental conversion count, not the attributed total
- –If true lift is high, that is a case for maintaining or increasing spend in that channel
- –If true lift is low, investigate whether the budget would perform better in channels with higher causality
- –Do not pause campaigns immediately based on a single test — run a follow-up test before making permanent cuts
- –Document the test design, results, and decisions made so future teams can build on the methodology