"We're getting 500 conversions a month from Google Ads — but how many of those would have happened anyway?" This is the question that regular conversion tracking can't answer. Conversion tracking tells you what happened after an ad click. It doesn't tell you whether the ad caused the conversion.
A customer who searched for your brand name, clicked your ad, and purchased might have purchased anyway — through organic search, a bookmark, or by typing your URL directly. Your conversion tracking counts it as an ad-driven result. But was it really?
Conversion lift testing is the method that answers this question. It measures the incremental impact of your ads: the conversions that would not have happened without the advertising.
1. Why conversion tracking alone isn't enough
Conversion tracking measures correlation, not causation. Consider these scenarios:
| Scenario | What tracking reports | What actually happened |
|---|---|---|
| User searches your brand, clicks your ad, purchases | 1 conversion from brand search | User would have purchased anyway through organic |
| User sees a display ad, later searches and purchases | 1 view-through conversion | Display ad may or may not have influenced the purchase |
| User clicks a retargeting ad for a product in their cart, purchases | 1 conversion from retargeting | User was already going to buy — the ad just intercepted the path |
| User discovers your product through a prospecting ad, purchases | 1 conversion from prospecting | This might be a genuinely incremental conversion |
In each case, conversion tracking reports a conversion. But only the last scenario is clearly incremental — a sale that wouldn't have happened without the ad. The others might be cannibalizing organic conversions: taking credit for results that would have occurred regardless.
Lift testing separates the signal from the noise.
2. How conversion lift tests work
The concept is simple: it's a controlled experiment, like an A/B test but for ad exposure.
-
Split your audience into two groups:
- Test group (exposed): Sees your ads as normal
- Control group (holdout): Does not see your ads (or sees a placeholder/PSA ad)
-
Run the campaign for a set period (typically 2–4 weeks)
-
Measure conversions in both groups:
- Test group: X conversions
- Control group: Y conversions
-
Calculate lift:
- Incremental conversions = X - Y
- Lift percentage = (X - Y) / Y × 100%
If the test group had 500 conversions and the control group had 400, the lift is 100 incremental conversions (25% lift). Those 100 conversions are the ones your ads actually caused.
What makes it different from A/B testing
A/B testing compares two versions of something (ad creative, landing page) to see which performs better. Lift testing compares "ads on" vs. "ads off" to see whether the ads create value at all.
3. Running a lift test on Google Ads
Google's Conversion Lift study
Google offers Conversion Lift as a measurement tool for eligible accounts. It works by:
- Splitting your target audience into test and holdout groups (Google handles the randomization)
- Running your campaign normally for the test group
- Suppressing ads for the holdout group
- Comparing conversion rates between the two groups
Requirements:
- Account must meet minimum spend and conversion volume thresholds (Google doesn't publish exact numbers, but typically $10K+ spend and 100+ conversions during the test period)
- The study is set up through your Google Ads representative or through the Experiments section
- Test duration: typically 2–4 weeks
What you get:
- Incremental conversions (absolute number)
- Lift percentage
- Statistical confidence level
- Cost per incremental conversion (your actual CPA for conversions that wouldn't have happened)
Recent change (2025): Google reduced the minimum spend threshold from roughly $100,000 to approximately $5,000 through Bayesian statistical improvements, making lift testing accessible to a much wider range of advertisers.
DIY approach: geo-based lift test
If you don't qualify for Google's official Conversion Lift study, you can run a geo-based test:
- Select two sets of geographically similar markets (e.g., cities with similar demographics and business performance)
- Run ads in one set (test) and pause ads in the other (control)
- Compare conversion rates (using backend data, not ad platform data) between the two sets
- The difference is your lift estimate
Limitations: Geo-based tests are noisier than user-level randomization. External factors (local events, weather, competition) can skew results. You need enough volume in each geo to detect a statistically significant difference.
4. Running a lift test on Meta
Meta's Conversion Lift study
Meta offers Conversion Lift tests through the Experiments tool in Ads Manager.
How it works:
- Meta splits your target audience into test (sees ads) and holdout (doesn't see ads) groups
- After the test period, Meta compares conversions between the groups
- Results include incremental conversions, lift percentage, and cost per incremental result
Setup:
- Go to Ads Manager → Experiments → Conversion Lift
- Select the campaign(s) to test
- Choose the holdout percentage (typically 10–20% of the audience)
- Set the test duration (7–28 days recommended)
- Start the test
Requirements:
- Sufficient campaign budget (Meta recommends at least the normal budget — don't reduce spending during the test)
- Enough expected conversions to reach statistical significance (Meta will estimate this before you start)
- The Meta Pixel or Conversions API must be active (the measurement uses these signals)
Meta's Incremental Attribution (always-on)
In addition to formal Conversion Lift studies, Meta launched Incremental Attribution in Ads Manager (April 2025). Unlike a lift test, this is an always-on feature that separates ad-driven conversions from conversions that would have occurred naturally — no holdout group required.
How it works: Meta uses its historical lift test data and machine learning to estimate which reported conversions are incremental, applying the estimate at the campaign level in your standard Ads Manager reporting.
Advantage: No holdout cost, no test setup, continuous data.
Limitation: It's a model-based estimate, not a controlled experiment. Less rigorous than a formal lift test, but far better than no incrementality measurement at all. Early adopters report 20%+ improvements in identifying true performance when optimizing based on incremental metrics.
Interpreting Meta's results
Meta reports:
- Conversions (test): Total conversions in the exposed group
- Conversions (control): Estimated conversions in the holdout group (these people weren't shown ads)
- Incremental conversions: The difference
- Cost per incremental conversion: Your true CPA for new conversions
A common surprise: the cost per incremental conversion is almost always higher than your reported CPA. This is because your reported CPA includes conversions that would have happened anyway. The incremental CPA reflects only the conversions the ad actually caused.
5. How to read the results
The key metrics
| Metric | What it tells you | What to watch for |
|---|---|---|
| Lift % | How much your ads increased conversions above the baseline | <10% lift suggests most conversions would happen anyway |
| Incremental conversions | The absolute number of conversions your ads caused | Compare this to your total reported conversions to see the "organic overlap" (if Google Ads and GA4 counts already differ, start with diagnosing the count mismatch) |
| Cost per incremental conversion | Your true CPA | If this exceeds your target CPA, your ads are less efficient than tracking suggests |
| Confidence level | Statistical reliability of the result | Below 90% means the result could be noise — extend the test or increase budget |
What different lift results mean
High lift (30%+): Your ads are genuinely driving new conversions. Prospecting campaigns, new market expansion, and non-brand search typically show high lift.
Moderate lift (10–30%): Your ads are contributing, but a meaningful portion of conversions would happen organically. This is common for brand search and retargeting.
Low or no lift (<10%): Most of your tracked conversions would happen without the ads. This doesn't mean the ads are worthless — they may defend against competitors — but you're paying for conversions you'd mostly get for free.
Negative lift: The control group converted more than the test group. This is rare and usually indicates a test design issue (e.g., the test period was too short, or external factors dominated).
6. Common mistakes that invalidate lift tests
| Mistake | Why it breaks the test | How to avoid it |
|---|---|---|
| Test period too short | Not enough conversions to reach statistical significance | Run for at least 2 weeks, or until the platform reports 90%+ confidence |
| Changing the campaign during the test | Budget, targeting, or creative changes introduce confounding variables | Lock the campaign configuration for the entire test period |
| Too small a holdout group | Not enough control conversions to measure against | Use at least 10% holdout (20% is better for smaller accounts) |
| Seasonal timing | A sale, holiday, or competitor event skews one group | Choose a "normal" business period, or run the test long enough to average out seasonal effects |
| Measuring with platform data instead of backend data (geo tests) | Platform data includes attribution modeling, which biases toward showing lift | Use CRM or order system data as the source of truth |
| Running multiple tests simultaneously | Audiences overlap, confounding both tests | Run one lift test at a time, or ensure audience isolation |
7. Where lift testing fits: the measurement triangle
Lift testing is one of three complementary measurement approaches. Each answers a different question:
| Approach | Question it answers | Strengths | Weaknesses |
|---|---|---|---|
| Attribution (last click, data-driven) | Which touchpoint gets credit for each conversion? | Real-time, granular, actionable for daily optimization | Doesn't prove causation; biased toward lower-funnel |
| Conversion lift testing | Did the ads cause additional conversions? | Proves causation via controlled experiment | Requires holdout (opportunity cost), point-in-time only |
| Media Mix Modeling (MMM) | How does each channel contribute to total business outcomes? | Cross-channel view, accounts for offline factors | Aggregate-level only, requires 2+ years of data, can't guide individual campaign decisions |
The strongest measurement programs use all three: attribution for daily campaign management, lift testing to validate whether specific campaigns are incremental, and MMM for strategic budget allocation across channels. If you can only pick two, start with attribution and lift testing — they cover the operational and causal questions that drive the most day-to-day decisions.
Always-on incrementality tools
Formal lift tests require a holdout group and a fixed test window. A newer category of tools runs continuous synthetic-control tests that estimate incrementality without withholding ads from a control group:
- How they work: These platforms (e.g., INCRMNTAL, Haus, SegmentStream) create a "synthetic twin" of your test market using statistical blending of control markets, then compare actual performance against the synthetic baseline.
- Advantage: No opportunity cost from holdout groups; always-on measurement instead of periodic tests.
- Tradeoff: Less statistically rigorous than a true randomized holdout. Best used for directional guidance, not high-stakes budget decisions.
For advertisers who can't afford the holdout cost of formal lift tests, or who need incrementality signals between periodic formal tests, these tools fill a practical gap.
8. When to run (and not run) a lift test
Good candidates for lift testing
- Brand search campaigns: The classic question — "are we paying for clicks we'd get organically?"
- Retargeting campaigns: Are these converting people who would buy anyway?
- Display/video prospecting: Is this actually driving new customers, or just taking credit?
- When a channel's reported ROAS seems too good to be true
When not to run a lift test
- Low conversion volume: If you get fewer than 50 conversions per week, you won't reach statistical significance in a reasonable timeframe
- New campaigns: You need a stable baseline before testing incrementality
- When you can't afford to pause ads for a subset: The holdout group doesn't see ads — if you're in a competitive market where pausing means losing share permanently, the test has a real business cost
Frequently asked questions
Q. Should we stop brand search campaigns if a lift test shows low incrementality? A. Not necessarily. Low lift on brand search is common — most of those clicks would come through organic. But pausing brand search can let competitors bid on your brand terms and intercept your traffic. The decision depends on your competitive landscape, not just the lift number.
Q. How is this different from the Attribution reports in Google Ads? A. Attribution reports redistribute credit for conversions that already happened (the total doesn't change — the distribution does). Lift testing measures whether the conversions would have happened at all. They answer fundamentally different questions.
Q. Our lift test showed 0% lift. Does that mean our ads are wasted? A. Not automatically. Zero lift means the conversions would have occurred without ads during the test period. But ads may serve other purposes: brand awareness, competitive defense, or priming future purchases that fall outside the test window. Consider the strategic role, not just the immediate conversion impact.
Q. Can we run a lift test on a single campaign, or does it need to be account-wide? A. You can test a single campaign. In fact, single-campaign tests are more actionable — they tell you whether that specific campaign is incremental. Account-wide tests give you the overall picture but don't help you decide which campaigns to scale or cut.
Q. How often should we run lift tests? A. Run one when you have a specific decision to make (should we scale this campaign? is retargeting worth the spend?). Don't run them continuously — each test requires a holdout group that doesn't see ads, which has an opportunity cost. Quarterly or biannual testing is a common cadence for larger advertisers.
Conclusion: measure what tracking can't
Conversion tracking answers "what happened after the ad?" Lift testing answers "what happened because of the ad?" Both are necessary. Tracking gives you the day-to-day operational data for optimization — though consent mode can reduce the volume of tracked conversions, which in turn affects lift test design. Lift testing gives you the strategic data for investment decisions.
The uncomfortable truth that lift testing often reveals: a significant portion of reported conversions would have happened without the ads. This doesn't mean advertising doesn't work — it means it works differently (and often less dramatically) than the tracking dashboard suggests. The advertisers who understand this make better budget decisions.
ConversionOK ensures the foundation is solid — that the conversion signals your site sends actually reach the ad platform. Lift testing builds on top of that foundation: once you know your tracking is accurate, you can ask the deeper question of whether the tracked conversions are incremental. You can't test what you can't measure.