- A RICE score looks like a measurement. What each factor really measures, and where the number misleads.
- The ten-minute sensitivity test: a ranking that flips signals a decision resting on a guess.
- When the score still helps, and what it cannot see: dependencies, energy, the cost of saying no.
Monday, three serious ideas are on the table: redo the homepage of your site, write a free guide, add a module to your course. You open a spreadsheet, fill in four columns, and the homepage comes out on top with 450 points, against 320 for the guide. The relief is immediate. A precise number feels like a decision made with method.
This is where the RICE score turns risky. The formula is not the problem. It compares serious options well. The problem is the authority a number takes over a decision, while nobody checks what feeds it. In that spreadsheet, three of the four values are estimates. The result looks like a measurement. It is not one.
What each factor really measures
RICE multiplies reach, impact, confidence, and divides by effort: (Reach × Impact × Confidence) / Effort. Sean McBride described the model at Intercom in 2018, in a post that admits impact is hard to measure precisely and has to be picked from a multiple-choice scale (Intercom, 2018). The admission is honest. It already names the problem: the four factors do not carry the same solidity.
Reach is the only factor you can count. Intercom recommends taking it from real product metrics rather than from a number pulled out of the air: how many people reach this step, over which period. If you have that figure, the score gains weight. If you do not, you are writing an estimate, and you need to remember it when you read the result.
Impact is an ordinal scale: 3 for massive, 2 for high, 1 for medium, 0.5 for low, 0.25 for minimal. It ranks the options, but nothing says a massive impact is exactly three times a medium one. Multiplying ranks as if they were measurements gives a number that looks exact for a meaning that is not. This is the part of the model that bothers me most, and the part people forget when they read “450”.
Confidence is declared by the person defending the project, so by the person who most wants it to go ahead. Moore and Healy (2008) describe overconfidence in three forms, one of them overprecision: excessive certainty about the accuracy of your own beliefs (Moore and Healy, 2008). A confidence of 80 % is exactly where overprecision settles in, because it looks like prudence.
Effort, the divisor, is usually underestimated. Buehler, Griffin and Ross (1994) showed that people predict completion times that are shorter than the times they actually take, and they do it even when they recall past overruns (Buehler et al., 1994). An effort that is too small inflates the score. And the projects you think are simple are the ones that benefit most.
The ten-minute sensitivity test
This test does not measure how good your estimates are. It measures how much your ranking depends on them. Run it on two or three options, no more:
- For each option, write its four values (reach, impact, confidence, effort) on one line, in a spreadsheet or a plain text file.
- Calculate each score and note the ranking.
- Set the confidence of the option in first place to 50 %, the “low” value on the scale.
- Recalculate. If the ranking holds, the decision is solid: choose and act.
- If it flips, your decision rests on an estimate you have not checked. Do not choose yet. Identify the missing data point, a traffic figure, a cost, a customer comment, and go and get it.
An example, invented for illustration. The homepage: reach 150 visitors a quarter, impact 3, confidence 100 %, effort 1 person-month, so a score of 450. The guide: reach 400, impact 2, confidence 80 %, effort 2 person-months, so 320. The homepage wins. But if you never checked the visitor count and you drop its confidence to 50 %, its score falls to 225 and the guide takes the lead. The formula changed nothing. What decides is a figure nobody looked at.
In that case, the right answer is not a third score. Go and get the homepage visitor count, then recalculate once. If the ranking stays the same with a real figure, you can decide. If not, you have found the real subject of the decision, and it had nothing to do with the formula.
Mistakes that come back
Four mistakes come back often once people move to scoring.
The first is scoring everyday tasks. The model was built for projects competing for the same slot. Applied to “answer emails” or “tidy the desk”, it produces numbers that help nobody and cost half an hour.
The second is scoring before eliminating. If your list holds ten options and three of them are leftovers from old projects, the spreadsheet will only rank them. A first pass with an Eisenhower matrix removes the noise: only the important options deserve a score. The detail of that filter is in the prioritization lesson.
The third is comparing scores produced by two teams or two people. Impact and confidence scales depend on who fills them in. A 2 for one person is not a 2 for another. A score compares options evaluated by the same person, with the same criteria, at the same time.
The fourth is using the same confidence everywhere. Putting 80 % on everything amounts to evaluating nothing. Honest confidence varies from one option to the next: an option backed by a metric deserves 100 %, one backed by a hunch deserves 50 %, and that gap should show in the table.
Making estimates less wrong
You will never remove the estimate. You can make its source visible. Add a “source” column next to each value and write one of three notes: measured, remembered, hoped. A reach marked “measured” carries more weight than one marked “remembered”. The column does not change the calculation, but it changes the discussion: when someone challenges a figure, you know whether it rests on data or on an impression.
This answers the overprecision problem directly. A number with its source can be defended. A number without one gets repeated until it passes for a fact.
Then apply one simple rule: a value without a source gets 50 % confidence before you compare, whatever it shows. You will quickly see which options rest on something concrete and which rest on the wish to do them. This is not a penalty for your ideas. It is the difference between an option you have observed and one you are hoping for.
When the score still helps
The score remains useful when three conditions hold together. The options are serious and close, otherwise the gap alone settles it. You can replace at least one estimate with a measurement. And you have to explain your choice to someone else.
In that last case, the score serves less to decide than to discuss. When two people disagree, the spreadsheet lets them argue about one specific figure instead of a general feeling. The disagreement becomes useful: you go looking for the missing data instead of defending a position.
What the score cannot see
The score does not see dependencies, your energy, or what saying no costs. The option at the top may need another task finished first, and the spreadsheet cannot know that. An option that looks average may unlock three other projects. And the score says nothing about what you give up by choosing it, which is often the question that matters most.
A list that is too long does not need a score. It needs reducing first. If your options number in the dozens, reread why your to-do list wears you out before you even start before you open the spreadsheet. Sort, then score: that order works.
What this changes for you
Tonight, take the two options you hesitate to launch. Write their four values, with a source for each, then run the sensitivity test. It takes ten minutes and tells you whether your decision rests on data or on an impression. If the ranking holds, go. If it flips, find the data before you choose.
To make that choice stick over your week, the prioritization lesson sets out the full grid: importance, what it unlocks, the cost of waiting, what you give up. It also sets up the weekly review that keeps one dominant priority. The score comes after that sorting, never before.
Sources
Sources checked on 5 November 2026.
See the 3 sources of this article
- Sean McBride, Intercom, 5 January 2018, RICE: Simple prioritization for product managers
- Roger Buehler, Dale Griffin and Michael Ross, Journal of Personality and Social Psychology, 1994, Exploring the “planning fallacy”: Why people underestimate their task completion times
- Don A. Moore and Paul J. Healy, Psychological Review, 2008, The trouble with overconfidence