---
title: "What is incrementality testing and how do you run it?"
author: "Performetic Ekibi"
url: "https://www.performetic.com/en/blog/incrementality-testing-guide"
published: "2026-02-16T08:00:00.000Z"
updated: "2026-10-05T02:41:56.023Z"
---

# What is incrementality testing and how do you run it?

> An incrementality test compares a test group that can see your ads with a control group that cannot, to measure whether the ads really create additional sales. Attribution divides credit; incrementality measures causality. In ecommerce you can run it with Meta Conversion Lift, Google Ads Conversion Lift or city-based geo tests.

## What is incrementality testing and how do you run it?

An incrementality test compares a test group that is eligible to see your ads with a control group that does not see them, to measure whether the ads really create additional sales. To run one, you set a hypothesis, split people or regions randomly into two groups, run the test long enough without changes and calculate the difference in conversions between the groups.

Here is why this matters. Your retargeting campaign reports a 12 ROAS. How many of those people would have bought anyway without the ad? Someone who filled a cart and knows your brand often comes back without being prompted. Attribution gives the ad credit for those sales; an incrementality test answers "what would have happened without the ad?".

## What is the difference between incrementality and attribution?

Attribution splits the credit for a completed sale among the touchpoints that came before it. However advanced the model, it only uses observed paths and cannot prove that the ad caused the sale.

| Attribute | Attribution | Incrementality test |
|---|---|---|
| Question it answers | Who should get credit? | Did the ad create extra sales? |
| Method | Touchpoint data and a model | Randomized test and control groups |
| Frequency | Continuous, daily | Periodic, time-boxed |
| Cost | No extra cost | Control group sees no ads, budget requirements may apply |
| Best use | Daily optimization | Validating large budget decisions |

Meta's Conversion Lift page defines the difference clearly: the test group is made up of people who have the opportunity to see your ads, the control group of people who do not, and the difference in conversions between them is reported as lift.

## What incrementality testing methods are there?

### Platform lift studies

**Meta Conversion Lift:** You create it under Experiments. Meta says there is no additional cost for the test but certain budget requirements may apply. As a guide on its help page, your account needs a campaign that started in the past year with at least USD 5,000 in spend and at least 500 conversions under a 1-day click, 7-day click or 1-day view attribution setting. Self-serve tests also have signal quality requirements, such as using the Conversions API with a minimum event match quality score.

**Google Ads Conversion Lift:** Comes in two types, based on users and based on geography. User-based studies randomly split users into groups; geography-based studies split regions into test and control. Eligibility and the required budget are determined during setup based on campaign type and historical conversion data, and some campaign types require an account representative.

### Geo holdout tests

If there is no ready-made platform tool, or you want to test several channels at once, split cities or regions with similar sales history into two groups. Ads run in the test regions and pause in the control regions, then you compare sales trends between the two. Sales data comes from your store admin, so the result does not depend on platform attribution.

### On/off (time-based) tests

Pausing a channel for a few weeks and watching sales is the simplest method but also the weakest. Seasonality, the promotional calendar and competitor moves easily distort the result. If nothing else is possible, comparing against the same period last year can still give a directional read.

## How do you run an incrementality test step by step?

1. **Write the hypothesis.** For example: "Meta retargeting adds at least 10% extra sales on top of existing sales."
2. **Pick the success metric.** Purchases, revenue or new customers. Choose one primary metric.
3. **Choose the design.** Platform lift study or geo test? The number of channels and your budget decide.
4. **Plan duration and size.** Cover your purchase cycle; usually several weeks rather than a few days.
5. **Protect the test.** Do not change budget, targeting or creative during the test, and avoid major sale periods.
6. **Calculate the result.** Lift = (test group conversion rate - control group conversion rate) / control group conversion rate.
7. **Document the decision.** Turn the result into a budget plan and a correction factor for reading attribution reports.

### Example: calculating incremental ROAS

As an example, a cosmetics brand runs a four-week lift test on its retargeting campaign:

| Item | Test group | Control group |
|---|---|---|
| People | 200,000 | 200,000 |
| Purchases | 2,400 | 2,000 |
| Conversion rate | 1.20% | 1.00% |
| Incremental purchases | 400 | - |
| Ad spend | 160,000 TRY | 0 |
| Average order value | 900 TRY | 900 TRY |

Lift is 20%. Incremental revenue is 400 x 900 = 360,000 TRY and incremental ROAS is 360,000 / 160,000 = 2.25. The same campaign may claim most of the 2,400 purchases in its dashboard and report a much higher ROAS. The decision should rest on whether incremental ROAS clears your break-even level, not on the dashboard ROAS.

### Which channels should you test first?

You do not need to test every channel at once. Prioritize channels that show the highest ROAS in the dashboard while carrying the biggest risk of reaching people who would buy anyway:

- **Retargeting:** Cart abandoners and product page viewers are already interested; incremental contribution may be lower than reported.
- **Brand search:** People searching your brand could arrive through the organic result too; the ad's extra contribution is worth testing.
- **Your largest budget line:** A small misreading in the channel that takes most of the spend affects the most money.

Upper-funnel campaigns carry the opposite risk: they may look weak in the dashboard while contributing more in reality. Testing them before cutting budget is the smarter move.

## What are common incrementality testing mistakes?

- **Tests that are too short.** Tests shorter than the purchase cycle miss delayed sales.
- **Optimizing mid-test.** Touching the campaign distorts the difference between test and control.
- **Small samples.** Low-conversion tests produce statistically uncertain results. Meta reports an interval of the most probable outcomes alongside its estimate; do not make firm decisions on wide intervals.
- **Generalizing from one test.** One period's result does not fit every season; retest major channels a few times a year.

General A/B testing principles apply when designing a test; our [ecommerce A/B testing guide](/en/blog/ecommerce-ab-testing-guide) covers sample size and duration in detail.

## How do incrementality tests fit with marketing mix modeling?

Marketing mix modeling (MMM) statistically estimates each channel's contribution from long-run aggregate data. Meridian, Google's open-source MMM, supports feeding experiment results into the model as priors so estimates are anchored in causal evidence. In other words, incrementality tests are not just the answer for one campaign; they are calibration points for your whole budget model. This is especially valuable when balancing awareness and performance budgets, which we discuss in our [brand awareness vs performance ads guide](/en/blog/brand-awareness-vs-performance-marketing-budget).

## Key takeaways

- Incrementality tests measure true ad contribution from the difference between exposed-eligible and control groups.
- Attribution divides credit; incrementality measures causality.
- Meta and Google Ads offer built-in lift studies with spend and conversion eligibility requirements.
- Geo tests are a strong alternative for measuring several channels independently of platforms.
- Base decisions on incremental ROAS, not dashboard ROAS, and retest major channels regularly.

If you want a test plan that shows which channels truly drive extra sales, the Performetic team can assess your account in a [free growth analysis](/en/contact).

## FAQ

### How long should an incrementality test run?

Long enough to cover your product's purchase cycle and collect sufficient conversions. Tests of a few days miss delayed sales, so ecommerce tests usually run for several weeks. Platform tools give duration and power estimates during setup; avoid shortening what they recommend.

### Can a small brand run an incrementality test?

Self-serve tools such as Meta Conversion Lift have spend and conversion requirements, so small accounts may not qualify. In that case, a simple geo test splitting similar cities into test and control, or a careful on/off test compared with last year, can still give a directional answer.

### How do you calculate incremental ROAS?

First find the number of incremental purchases from the conversion difference between test and control. Multiply it by average order value to get incremental revenue. Divide incremental revenue by ad spend during the test to get incremental ROAS, then compare it with your break-even ROAS.

### Does Meta's incremental attribution setting replace a lift test?

No. Meta's incremental attribution optimizes and reports delivery using models that predict whether a conversion was caused by an ad. That is a model. Conversion Lift is an actual experiment with a randomized control group and gives stronger evidence of causality.

## Sources

- [About Conversion Lift (Meta Business Help Center)](https://www.facebook.com/business/help/221353413010930)
- [About Conversion Lift (Google Ads Help)](https://support.google.com/google-ads/answer/12003020)
- [About attribution models and attribution settings (Meta Business Help Center)](https://www.facebook.com/business/help/460276478298895)
- [Meridian (Google for Developers)](https://developers.google.com/meridian)