What is the ICE score
The ICE score is a method for prioritizing test ideas that rates each hypothesis from 1 to 10 on three axes, Impact, Confidence and Ease, and combines them into a single score to decide what to test first.
By Fernando Blanco Dosil · Updated November 2, 2026
The three axes
The point of ICE is speed. It doesn't try to be exact: it tries to stop you arguing every Monday about which test goes first.
- Impact: if it works, how much can it move the metric? A change on the product page, which almost every purchase goes through, is usually worth more than one on the About page.
- Confidence: what evidence do you have that it will work? A figure from your analytics or a pattern seen across other stores gives more confidence than a hunch.
- Ease: how much does it cost to build? Changing some copy is a 9; building a new component, a 4.
How it is calculated
There are two common versions. The most usual is the average of the three axes, which gives an easy-to-read score from 1 to 10. Other teams multiply them, which penalizes ideas with one very low axis more. Either works as long as you always use the same one.
Worked example
Three hypotheses from a clothing store, scored with the average of the three axes.
| Hypothesis | Impact | Confidence | Ease | ICE |
|---|---|---|---|---|
| Size guide next to the size picker on mobile | 8 | 6 | 9 | 7.7 |
| Redesign the whole checkout | 9 | 4 | 3 | 5.3 |
| Free shipping highlighted on the product page | 5 | 7 | 8 | 6.7 |
The first one goes first: it isn't the highest impact, but it is the best bet between what it could deliver and what it costs.
The limits of ICE
It is subjective: two people can score the same idea differently. Writing one line on why each score was given helps, so it can be reviewed.
It also ignores how much traffic each page gets, unless you fold that into impact. A brilliant idea on a page almost nobody visits can take months to produce a result.
How Bixlow does it
When Bixlow analyses your website and proposes hypotheses, each one comes with its ICE: a score from 1 to 10 per axis, with one line justifying each score, and the average of the three to rank the list. That ICE is computed by code with fixed rules, not by the AI model: the AI proposes the change and the rules decide how it ranks. Bixlow shows no uplift percentages or money figures next to the score.