# How to build a creative testing budget

> Size your creative testing budget from your cost per sale. The 50-results rule, how many ads to test at once, and how to split testing and scaling money.
- **Author**: Marxx
- **Published**: 2026-10-24
- **Category**: Testing
- **URL**: https://marxx.ai/library-guides/how-to-build-a-creative-testing-budget

---

Build a creative testing budget by working backwards from your cost per result. Give each test ad set about 50 times your target cost per result each week, because Meta says an ad set needs about 50 results in a week to exit learning.

Put several different ads inside that one ad set, not several ad sets.

## The rule Meta gives you

Meta says ad sets usually exit the [learning phase](/glossary/learning-phase) after about 50 results in the week after the last significant edit. Meta also says to avoid high ad volumes, and to use budgets large enough to get enough results (checked 2026-10-20, [source](https://www.facebook.com/business/help/112167992830700)).

## Steps

1. **Write your target cost per result.** Base it on your margin. See [break-even ROAS](/glossary/break-even-roas).
2. **Multiply by 50.** That is the weekly budget for one test ad set. Divide by 7 for the daily figure.
3. **Put 3 to 5 truly different ads in that one ad set.** Different angle, person and hook, not colour swaps. See [creative diversity](/glossary/creative-diversity).
4. **Use A/B tests for clean answers.** Meta recommends the same budget for both versions. It advises against informal tests where you switch ad sets on and off by hand ([source](https://www.facebook.com/business/help/1738164643098669), checked 2026-10-20).
5. **Write your kill rule first.** For example, turn off an ad that spends three times your target cost per result with no result. See [how to know when to turn off an ad](/guides/how-to-know-when-to-kill-an-ad).
6. **Set the test share.** A starting point is 15% of monthly spend. Adjust it once you see how often tests win.
7. **Name every test the same way,** so you can compare later. Use the [ad naming convention template](/templates/ad-naming-convention-template).

## A worked example

Say you spend INR 4,00,000 a month. Your target cost per purchase is INR 400.

- Test share: 15% of INR 4,00,000 = INR 60,000 a month, about INR 13,850 a week.
- One test ad set on purchases needs 50 x INR 400 = INR 20,000 a week. That does not fit.

Meta's own fix for a thin event count is a more frequent event. Say add to cart costs INR 100.

- One test ad set on add to cart needs 50 x INR 100 = INR 5,000 a week.
- You can run two test ad sets for INR 10,000 and keep INR 3,850 spare.

Add to cart is a weaker signal. Once a test ad shows promise, move it into your main campaign, where it optimises for purchases, and confirm it there.

## What to test with the money

- One new angle per test ad set
- Two hooks on your best angle
- A new format, such as UGC video against a static review card

Keep the budget steady while a test runs. Meta says frequent budget changes can restart learning. If your ad sets still show "Learning limited", read [how to exit learning limited](/guides/how-to-exit-learning-limited).

## Mistakes to avoid

- **Many ad sets with tiny budgets.** None gets enough results.
- **Judging after one day.**
- **Testing tiny tweaks.** A new button colour teaches you little.
- **Changing the budget mid-test.**

We make Marxx. It shows which hooks, angles and formats you have already run, and which you have not tested yet.

## FAQ

### How much should I spend on creative testing?

Enough to give each test ad set about 50 results a week, which is 50 times your target cost per result. Start with about 15% of monthly spend if you have no data.

### How many ads should I test at once?

Three to five different ads in one ad set is a good range. More ads split the budget thinner, and Meta advises against high ad volumes.

### How long should a test run?

At least a week, since Meta measures learning over the week after your last significant edit.

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