Prioritising experiments with the ICE score model

Your growth team always has more ideas than capacity. Forty experiments on the backlog, three you can genuinely run this month. The question is not which ideas are good, but which one you do first. The ICE score model gives you a simple, repeatable answer. You rate every experiment on three axes: Impact, Confidence and Ease. The number that comes out determines the order. In this article you will read what the three components mean, how to score them and how the maths works, using a concrete example.

ICE is not an oracle. It does not make choices for you, it makes your choices visible and debatable. That is exactly why it works: it takes the discussion away from whoever shouts loudest and hands it to the numbers you fill in together.

What is the ICE score model?

ICE stands for Impact, Confidence and Ease. For every experiment on your backlog you give each of those three a score, usually on a scale of 1 to 10. You then combine the three scores into a single ICE score and sort your entire list from high to low. The experiments at the top are the ones you run first.

The model became popular within growth marketing because it is fast. You need neither a complicated spreadsheet nor a data specialist to prioritise a backlog. Ten minutes with your team and a shared document are enough for a first round. That speed is deliberate: prioritising must not become a project in itself, because then you run fewer experiments, and that is precisely where growth comes from.

Important to understand: ICE is a comparison instrument, not a prediction. A score of 240 does not mean a test will deliver 240 percent more revenue. It only means that this experiment, based on what you know today, looks like a better use of your time than an experiment scoring 90. You are comparing ideas with each other, not with reality.

The three components dissected

Impact: how far does this move the needle?

Impact is about the size of the effect if the experiment succeeds. How many extra leads, conversions or how much extra revenue do you expect? A score of 10 means a successful experiment delivers a big jump on the metric that matters to you. A 2 means a marginal improvement you will barely see in your numbers.

When scoring impact it helps to look at two things: how large the effect is per user, and how many users it touches. Moving a button on a page that thousands of visitors see every month reaches a lot of people. That same change on a page with ten visitors a month scores low on impact, however clever the idea is.

Confidence: how sure are you?

Confidence is the honesty test of the model. How strong is your evidence that this experiment works? You put down a 10 when you have hard data: earlier tests, clear user feedback, a pattern in your analytics. A 2 or 3 belongs to an idea that rests purely on gut feeling.

This axis is the reason ICE is more than a popularity contest. Anyone can come up with an idea that has enormous impact. The question is whether you believe it will actually work, and why. Confidence forces you to name that evidence explicitly. Nobody has any backing? Then the score drops, and rightly so. That is how you keep expensive gambles at the bottom of the list until you have gathered cheaper evidence.

Ease: how much effort does it take?

Ease is about the effort needed to run the experiment. A 10 is something you have live this week without help from development. A 2 demands weeks of work, several teams or a technical intervention. Mind the direction: high ease means little effort. The easier it is, the higher the score.

Ease is what makes ICE practical for small teams. A brilliant idea that takes three months to build blocks ten quick tests in that time which together could deliver just as much. By weighing effort, you choose learning speed: more experiments in the same period means knowing sooner what works.

The worked example

The classic ICE formula is a simple multiplication:

ICE score = Impact x Confidence x Ease

Suppose you have three experiments on your backlog:

Experiment A: rewriting the email subject lines for your lead nurturing. You expect a noticeably higher open rate, but it only reaches people already in your flow. Impact: 5. You have seen earlier tests where this worked, so your evidence is reasonable. Confidence: 7. You can change it yourself, no development needed. Ease: 9. Score: 5 x 7 x 9 = 315.

Experiment B: redesigning the entire homepage. If it succeeds, it reaches everyone who arrives, so the impact could be large. Impact: 9. But you do not really know which elements make the difference, it is a bet on the whole. Confidence: 3. And it costs weeks of design and development. Ease: 2. Score: 9 x 3 x 2 = 54.

Experiment C: removing a field from your contact form. Fewer fields can raise conversion, but the effect is usually modest. Impact: 4. There is plenty of research and your own data supporting this direction. Confidence: 8. It is a five-minute change. Ease: 9. Score: 4 x 8 x 9 = 288.

Sorted by score: A (315), C (288), B (54). The homepage redesign may have felt like the big, ambitious idea, but it sinks to the bottom because you have little confidence in it and it costs a lot. The two quick, well-founded tests are the ones you run first. By the time you get to B, A and C may already have taught you which elements on your homepage really count, which raises B’s confidence by itself.

Some teams use the average of the three scores instead of the product. That gives a scale of 1 to 10 instead of 1 to 1000. Which variant you choose does not matter, as long as you apply it consistently across your whole backlog. The order is what counts, not the absolute number. If you also want to weigh the reach of an experiment explicitly, look at the RICE framework, which prioritises on reach, impact, confidence and effort.

The pitfalls that make the score worthless

ICE is only as reliable as the people filling it in. The biggest pitfall is one person assigning all the scores on their own. You then get no objectivity, but a single opinion cast in numbers. So calibrate with your team: have a few people score independently and discuss the outliers. It is precisely the disagreements that produce the sharpest insights, because they expose hidden assumptions.

Second pitfall: inflating scores to get your favourite idea to the top. If everyone knows that a higher score means their idea runs sooner, an incentive to score optimistically appears. Keep Confidence strict. No evidence, no high confidence, no exceptions.

Third pitfall: treating your scores as concrete. An ICE score is a snapshot of what you know today. As soon as an experiment produces data, the confidence of related ideas changes. Revise your backlog regularly instead of scoring it once and forgetting it.

Where ICE fits into your growth process

ICE is a prioritisation instrument, not a strategy. It determines which experiments you run first, but not which experiments belong on your list or which goal they serve. That question is solved one level higher. Growth marketing is not a loose collection of tests, but the system that orchestrates SEO, CRO, content and lead generation into one predictable growth engine. ICE is the dial with which you choose the right order inside that system, so your limited capacity goes to the most promising ideas.

Prioritising well only works if the ideas themselves are good. A sharp backlog starts with a clearly formulated growth hypothesis per experiment: what do you expect, for whom and why. And if you doubt whether a variant really performs better, a reliable A/B test is the way to raise your confidence on solid grounds for the next round.

Would you rather not set up a prioritisation method yourself, but have a team that runs the entire growth engine for you? As a growth marketing agency, we build the backlog, score the experiments and steer on leads, revenue and pipeline instead of isolated numbers that look nice but move nothing.

Ready to put your experiments in order and grow structurally? Get in touch and together we will look at which tests will push your growth engine forward hardest.