# How to measure incrementality on a small budget

> Find out which sales your ads really cause. Three tests you can run without a paid tool: pause test, state holdout and audience holdout, with a rupee example.
- **Author**: Marxx
- **Published**: 2026-11-20
- **Category**: Testing
- **URL**: https://marxx.ai/library-guides/how-to-measure-incrementality-on-a-small-budget

---

To measure incrementality on a small budget, hide your ads from one slice of your market and compare sales with a slice that still sees them. The gap is the sales your ads caused.

You need no paid tool, only a clean split and two to four weeks.

## What you are measuring

[Incrementality](/glossary/incrementality) is the share of your sales that happened because of your ads. Ad platforms credit every sale they can link to an ad. Some of those buyers would have bought anyway. A test shows how many.

## Why not use a lift tool?

Meta runs free Conversion Lift tests in Ads Manager under Experiments, but as a guide your ad account needs a campaign from the past year with $5,000 USD or more of spend and at least 500 conversions. Meta also recommends a test of at least 28 days (checked 2026-10-20). A small account may fall short of that. The three tests below do not depend on those minimums. Paid options are in [the best incrementality testing tools](/best/best-incrementality-testing-tools).

## Three tests you can run yourself

| Test | How it works | Weak spot |
|---|---|---|
| Pause test | Pause all ads for one to two weeks and compare with the same weeks before | Festivals, weather and salary dates also move sales |
| State holdout | Keep ads on in most states and pause them in a few alike states | Needs states that behave alike |
| Audience holdout | Split your customer list at random and exclude one half from ads | Tests only that audience |

The state holdout is the best of the three, because both groups face the same calendar. The steps are in [how to run a geo holdout test](/guides/how-to-run-a-geo-holdout-test). For the audience holdout, Meta lets you build an exclusion-only audience from a customer list (checked 2026-10-20).

## A worked example

Say you sell tea online in 10 states. You normally run ads everywhere. For three weeks you pause ads in 4 alike states and keep them on in the other 6.

- Weekly sales before the test: INR 6,00,000 in the 4 holdout states.
- Weekly sales during the test in those states: INR 5,10,000.
- Drop: INR 6,00,000 minus INR 5,10,000 is INR 90,000 a week.
- Ad spend you switched off in those states: INR 60,000 a week.
- Incremental return: INR 90,000 / INR 60,000 = 1.5.

Say Meta reported a [ROAS](/glossary/roas) (return on ad spend) of 4.0 in those states. The test says each rupee brought back INR 1.50 of sales that would not have happened otherwise. The platform number counted sales that were coming anyway.

Check the 6 states that kept ads. If their sales stayed flat at INR 9,00,000 a week, no wider trend moved your sales.

## Rules for a fair small test

- **Set the pass line first.** For example, "if holdout sales fall less than 5%, ads add little here".
- **Match your groups.** Pick states with similar sales over the last eight weeks.
- **Change nothing else.** No new offer, no price cut, no new channel.
- **Run whole weeks.** Each weekday should count equally.
- **Use big areas.** Meta says location accuracy cannot be guaranteed. A person who lives in a shown area can also see your ad while visiting a held-out one (checked 2026-10-20). States leak less than cities.

## FAQ

### How long should a test run?

Two to four weeks, in whole weeks. A shorter test often shows noise, not a real gap.

### Can an account that spends INR 50,000 a month run this?

Yes, but the result is noisy. Trust only large gaps, and repeat the test before you cut budget.

### How sure do I need to be?

Meta's guide to reading lift results calls a confidence of 90 percent or more high confidence (checked 2026-10-20). Small do-it-yourself tests rarely reach that, so treat the result as a strong hint and run it again.

---
- [More Testing articles](https://marxx.ai/library-guides/category/testing)
- [All articles](https://marxx.ai/library-guides)