RICE prioritization compares product ideas using four estimates: how many people they will reach, how much they will help, how confident you are, and how much work they will take. Customer feedback can improve those estimates, provided you keep the original problem and the evidence behind each number.
A board with hundreds of requests still leaves a difficult decision: which change will remove the biggest obstacle for the people you want to serve? The worked example below compares three ideas and shows how one weak assumption can change their ranking.
What the RICE score means
The formula is:
RICE = (Reach × Impact × Confidence) ÷ Effort
Intercom introduced RICE as a way to compare projects more consistently. In its original framework, reach counts people or events over a defined period, impact uses a shared scale, confidence discounts uncertain estimates, and effort measures total person-months.
You can use a different effort unit if the team needs it, but use that unit for every idea in the comparison. A score expressed per person-week cannot be compared directly with a score expressed per person-month. The score also has no universal threshold for a “good” feature: it is useful within a comparable set of options.
If you need to organize the requests before scoring them, start with our broader guide to prioritizing feature requests.
Start with one decision and one audience
Write down the decision the session should produce. For example: “Choose the next improvement to help administrators finish account setup.” This makes it easier to compare two solutions to the same problem, and harder to squeeze an unrelated dashboard redesign into the conversation.
Then define who counts toward reach and the time period you will use. In this example, the audience is administrators starting setup during the next quarter. People who finished setup last year do not belong in that estimate unless the proposed change helps them too.
Keep separate lists for work with different decision rules. A critical security fix or a contractual obligation may need to happen regardless of its score. Record that constraint explicitly instead of adjusting its impact estimate until it appears at the top.
Use feedback to improve each input
Reach: count affected people, not messages
Suppose six support tickets and four comments describe difficulty importing a team list. Before assigning reach, check whether some came from the same customer. Repeated messages can show the seriousness of a problem, but counting them as separate people inflates the estimate.
Now look beyond the people who wrote in. How many administrators encounter the import step? How many use the relevant file type? If those counts are unavailable, write down the missing information and lower confidence. A precise-looking number assembled from guesses is still a guess.
Votes help identify an interested group. They do not represent every person affected, and a person who voted for several requests should not be counted repeatedly within one request. Use the discussion to understand the task behind the vote.
Impact: describe the change in the user's task
Agree on a small scale before reviewing individual ideas. For the worked example below, impact is 0.5 for a small improvement, 1 for a meaningful improvement, and 2 for a large improvement to the chosen setup outcome. These are illustrative scoring choices, not measured effects.
Attach a sentence to each estimate. “Administrators can correct a bad CSV row without restarting the import” is more useful than “improves onboarding.” It tells engineering what needs to change and tells the team what it can check after release.
Avoid turning the scale into a negotiation. If two people disagree, identify the assumption that separates them. One might believe the import error blocks every new account; the other might know that most teams use a different setup path. Find evidence for that difference.
Confidence: show what you know and what you inferred
Use confidence to reflect the quality of the estimates. Direct observations of the problem, relevant usage data, and a reviewed implementation approach make a different case from one sales request and an untested technical guess.
Record the source beside the score. A short note is enough: “Support examples from three distinct accounts; no measurement of affected setup attempts yet.” That note remains useful when someone asks why the score changed next month.
Do not average away a crucial unknown. If the feature depends on a third-party capability that engineering has not checked, investigate that dependency before presenting the delivery estimate as reliable.
Effort: include the complete usable change
Ask the people who will do the work to estimate it. Include design, implementation, testing, rollout, and necessary support or documentation changes. A two-day coding estimate may leave out a week of data migration and verification.
For small teams, person-weeks are often easier to discuss than person-months. The important point is consistency. Also distinguish effort from elapsed time: one person waiting for an external review can delay a release without doing a week of additional work.
A worked example with three feature requests
The following is an invented planning example for an account-setup flow. It is not Productroad customer data or a claim about conversion improvements. All options use the next quarter as the reach period and person-weeks as the effort unit.
| Idea | Reach | Impact | Confidence | Effort | Score |
|---|---|---|---|---|---|
| Explain CSV import errors inline | 400 | 2 | 80% | 2 | 320 |
| Add reusable invite templates | 250 | 1 | 80% | 1 | 200 |
| Build a new setup dashboard | 600 | 1 | 40% | 4 | 60 |
For the import improvement, the arithmetic is (400 × 2 × 0.8) ÷ 2 = 320. Use 0.8 for 80%, rather than 80, or the result will be inflated by a factor of 100.
The import improvement leads in this example, but the team should still inspect the evidence. Is 400 the number of people who attempt an import, or the number who actually encounter an error? If only 100 are affected, the score becomes 80. The invite template would then rank above it.
Verify the affected population before committing to implementation. A short investigation may resolve the disagreement more cheaply than another scoring meeting.
Keep dependencies and exceptions visible
A ranked list is a starting point for a plan. It does not capture every sequencing constraint. If the import improvement requires a parser update, include that work in the estimate or document it as a dependency. Do not promise the smaller visible change while omitting the prerequisite.
There are also sensible reasons to take a lower-scoring option first. It may unblock a committed release or let you test a risky assumption sooner. Keep a brief decision note with the reason, the owner, and what would cause you to reconsider. This makes the exception inspectable without forcing every consideration into the formula.
Limit the size of the scoring session. Bring a manageable set of prepared candidates, with duplicate requests grouped and the original customer language available. If most of the session is spent figuring out what a request means, it is too early to compare scores.
Tell customers what the decision means
The person who requested a feature usually needs an answer about their problem. A score of 320 will not help them decide whether to wait, use a workaround, or choose another approach.
For the import example, a useful update might say:
We are working on clearer import errors so you can fix an invalid row without starting again. We are checking the approach against the examples you sent. We will update this request when the change is ready to try.
Only use that wording once the team has actually started the work. For an idea still under consideration, say what you are investigating and avoid implying a delivery date.
In Productroad, feedback boards hold requests and discussion, roadmaps make planned work visible, and the changelog communicates releases. Keep the RICE calculation in your team's existing planning worksheet and link it to the relevant requests. A teammate can then trace a score back to the customer evidence and see what happened after the decision.
Revisit the estimate after release
Return to the problem statement once people can use the change. Did administrators stop restarting the import? Are the same error messages still appearing in support conversations? Did a different obstacle become visible?
Compare the result with the expectation you recorded. If you overestimated reach, keep that observation for the next planning session. If the implementation took longer because migration work was missing, adjust how the team estimates similar work.
You do not need to reconstruct a RICE score for every historical item. Start with the next real decision, document the uncertain inputs, and keep the evidence close enough that another teammate can understand the choice.
Productroad helps you collect user feedback, prioritize it, make a roadmap of a project and inform customers via changelog about new cool features.



