ICE score: how to rank growth experiments

An ICE score ranks a growth experiment by averaging three ratings, each from one to ten: impact, confidence and ease. The highest score is the experiment to run next. It is a quick way to turn a long list of ideas into an ordered backlog without weeks of debate.

What impact, confidence and ease mean

Each experiment gets three ratings. Each one answers a different question, so score them separately and do not let one bleed into the others.

  • Impact: if this works, how much will it move the metric you care about? Think about the size of the change and how many people it reaches.
  • Confidence: how sure are you that it will work? This rests on evidence, such as customer conversations, past tests or a clear pattern in your data.
  • Ease: how cheap and fast is it to run? Count the time, money and people it takes to ship the test and read the result.

The ICE score is the average of the three, shown to one decimal. Seven and up shows in green and marks a strong candidate. Five up to just under seven shows in amber and is worth keeping on the list. Anything below five shows in grey and usually waits.

How to score each rating from one to ten

Scores only work if they mean the same thing every time. Use anchors like these so a seven today is a seven next month.

  • Impact. Low: a small tweak that few people will notice. Middle: a clear lift on one step of your funnel. High: a change that could shift your main metric on its own.
  • Confidence. Low: a hunch with nothing behind it. Middle: some signal, like a few customers asking for it or a similar test that worked. High: strong evidence from your own data or repeated customer feedback.
  • Ease. Low: weeks of engineering or a real budget. Middle: a few days of work from one person. High: something you can ship today with tools you already have.

Score fast. Spend a minute per experiment, not an hour. The goal is a rough order, not a perfect forecast.

A worked example

Here are three experiments for a product that wants more people to finish signing up and start using it.

  1. Shorten the signup form

    7.3
    ICE

    If we remove every optional field from the signup form, then your signup completion rate will rise because each extra field gives people a reason to stop.

    Metric
    your signup completion rate
    Impact
    6
    Confidence
    7
    Ease
    9
  2. Add an onboarding checklist

    6.0
    ICE

    If we show new users a short checklist of first steps, then your activation rate will rise because people will know what to do after they sign up.

    Metric
    your activation rate
    Impact
    7
    Confidence
    6
    Ease
    5
  3. Launch a referral program

    4.7
    ICE

    If we give users a reward for inviting a friend, then your new signups from referrals will rise because happy users will have a reason to share.

    Metric
    your signups from referrals
    Impact
    9
    Confidence
    3
    Ease
    2

The referral program has the biggest possible impact, yet it ranks last. It is hard to build, and nothing yet shows users want to share. The shorter form ranks first because it is cheap, likely to help and quick to read. Run it, learn from it, and score the referral idea again once you have more evidence.

This is the point of ICE. It stops the most exciting idea from jumping the queue when a smaller, surer bet would teach you more this week.

Common mistakes

  • Scoring everything high. If most ideas land above seven, your anchors have drifted. Spread the scores out.
  • Mixing up confidence and impact. A big idea is not a sure idea. Rate them apart.
  • Scoring alone. Ask one other person to score the same list, then talk through the biggest gaps.
  • Never rescoring. When a test ends, update the scores of related ideas with what you learned.
  • Treating the score as the decision. ICE gives you an order. You can still skip an idea for a good reason, as long as you say what it is.

Keep the backlog moving

Run the top experiment, record the result, then rescore what is left. A backlog that is scored once and never touched again goes stale fast. Small, steady loops beat one big planning session.

Keep reading

Rank all 19 channels for your company, free.