This calculator scores a list of features by reach, impact, confidence and effort, then ranks them as the numbers change. The formula is reach multiplied by impact multiplied by confidence, divided by effort.
| Feature | Reach | Impact | Confidence | Effort | Score | Remove |
|---|
The rows above are examples. Replace the names and numbers with your own features.
How do I choose each input
Four numbers, and three of them are a guess dressed up as a measurement.
Reach is how many people the feature touches in a period you fix first and never change mid comparison. The trap is counting one feature over a month and another over a quarter, because the two scores then compare nothing. Pick a month or a quarter for the whole list and write it next to the column.
Impact is how much the feature matters to each person it reaches, on the fixed scale 3, 2, 1, 0.5, 0.25. The trap is impact inflation, where every feature on the list ends up at 3 because everyone proposing a feature believes in it. A list where more than two or three rows sit at the top of the scale has stopped discriminating between features.
Confidence is how sure the team is about the reach and impact guesses, as 100, 80 or 50 percent. The trap is using it as a fudge factor to push a wanted feature back up the ranking after the first two numbers came out low. Confidence answers how much evidence backs the first two guesses, not how much the team wants the result to change.
Effort is person months, one person working for one month, not elapsed calendar time. The trap is the wrong unit twice over. Two people for a month is two person months, not one, and a feature that takes four calendar weeks with two people on it is four times the effort of the same four weeks with half a person on it. Write down who is actually doing the work before writing down how long.
A worked example
The three rows above, worked by hand.
Add dark mode: reach 800 a month, impact 1 (medium), confidence 80 percent, effort 2 person months.
800 × 1 × 0.8 ÷ 2 = 320
Export to CSV: reach 150 a month, impact 2 (high), confidence 100 percent, effort 1 person month.
150 × 2 × 1 ÷ 1 = 300
Rebuild onboarding flow: reach 2,000 a month, impact 3 (massive), confidence 50 percent, effort 6 person months.
2,000 × 3 × 0.5 ÷ 6 = 500
Rebuild onboarding flow ranks first even though half the confidence is a guess, because its reach and impact are large enough to survive being halved. Export to CSV ranks last of the three even though every number going into it is certain, because certainty on a small feature does not beat size on a bigger one.
When RICE is the wrong tool
RICE is a way to rank work nobody has asked for yet, by estimating who it will reach before it exists. Two situations call for something else.
The request already has real votes. A feature already sitting on a feedback board with people attached to it does not need a reach estimate, because the reach already happened and was counted. A board that scores requests from the votes, comments and views they already have replaces three of the four RICE guesses with a count.
The question is what to cut from a release with a fixed date, not what to build next. RICE produces an order with no natural cut line. For a release, sorting work into must, should, could and will not answers the question RICE was never built to answer.
The full definition of RICE, where it came from and where teams get it wrong, is on its own page. A filled prioritization matrix with real requests already placed is the other common format for the same decision. The ICE score calculator drops reach and scores impact, confidence and ease alone, for lists where reach is about the same on every row.
Where the reach number comes from in practice
For anything already requested, the reach number sits on the request itself rather than in a forecast.
A feedback board that keeps a vote count per feature, such as VoteFirst, turns part of the reach guess into a fact. Each vote is one person who said this matters to them, and a vote can optionally carry the weight of the voter's plan, so a request from ten paying customers is not scored the same as a request from ten accounts that never paid. That still leaves impact, confidence and effort as judgement calls. It only removes the one number that used to be a guess about people who had already spoken.