# The experiment backlog: how to write each idea, how to score it with ICE and where ICE goes wrong

One row per experiment idea: the hypothesis in a fixed sentence, the metric it moves, the ICE scores and their average, the sample it needs, and what happened. The page explains the ICE score as GrowthHackers defines it, the four ways it misleads a team, and the weekly cadence that keeps a backlog moving. Download the sheet as CSV.

## How to use it

Write every idea as a hypothesis before it gets a score. An idea that cannot be written as "we believe this change will move this metric because of this evidence" is not ready to test.

Score in the weekly meeting, together, and write the evidence for the Confidence score in the row. Re-score the ideas that share a page or a metric each time a test on them finishes.

## The columns

One row per idea:

| Column | What goes in it |
| --- | --- |
| ID | A short code you can say out loud in the meeting, such as ACT-014. |
| Hypothesis | We believe [change] for [audience] will [move metric] because [evidence]. |
| Metric | The one number that decides the test, and where it is measured. |
| Impact (1 to 10) | How much the metric moves if the hypothesis is right. |
| Confidence (1 to 10) | How strong the evidence is. Write the evidence in the next column. |
| Evidence | Data, research or a previous test behind the Confidence score. |
| Ease (1 to 10) | How little time and how few people it takes to run. |
| ICE score | The average of the three scores. |
| Sample needed | Visitors or users per variant, from the sample size calculator. |
| Owner | The person who runs it and writes it up. |
| Status | Idea, scored, running, read, shipped or dropped. |
| Result and decision | The measured change with its interval, and what you decided. |

## The ICE score

- **Definition.** Impact, Confidence and Ease are each graded from 1 to 10, and the ICE score is their average. That is the definition GrowthHackers gives for the framework.
- **Worked example.** An idea scored Impact 7, Confidence 4, Ease 8 has an ICE score of (7 + 4 + 8) ÷ 3 = 6.3.
- **Use it to order, not to decide.** The score sorts a long list into a short one. The meeting still decides which two or three tests run this week, and why.

## Where ICE goes wrong, and the fix

- **Scores drift up for ideas the room likes.** Fix: score alone first, then compare. A spread of more than three points on any score means the team does not share the same picture of the idea; talk about that before averaging.
- **Confidence without evidence.** Fix: no Confidence above 5 without a line in the Evidence column. A gut feeling is a 3.
- **Ease pulls the backlog toward small tests.** Fix: keep one slot a week for a high-impact test, whatever its Ease score. Small tests that cannot move the metric are not cheap; they use up traffic.
- **Scores go stale.** Fix: re-score ideas on the same page or metric when a related test finishes. A result is new evidence for every idea near it.

## The weekly cadence

- **Read results.** Each finished test: what moved, by how much, with what confidence, and the decision. Ten minutes at most per test.
- **Pick the next tests.** From the top of the scored list, the tests that fit the traffic you have this week. Check each one's sample against the calculator.
- **Add and score new ideas.** Everyone brings ideas written as hypotheses. Score them in the room.
- **Prune.** Drop ideas that have sat unscored or untested for more than a few weeks. The backlog is a queue, not an archive.

---

From Race to Scale: https://www.racetoscale.com/tools/growth-experiment-backlog-template