# RICE Framework: Prioritization Scoring and When to Override > The RICE framework scores backlog ideas by Reach, Impact, Confidence and Effort. See the formula, a worked example, and when to override the score. - URL: https://missiongrowth.io/blog/rice-framework - Published: 2026-09-24 · Updated: 2026-09-24 - Author: Ömer Furkan Aktaş, Founder, Mission Growth - Publisher: Mission Growth. Company facts: https://missiongrowth.io/llms.txt More backlog ideas exist than any team can ship, and each one sounds urgent to whoever pitched it. The RICE framework, also known as the rice scoring model, replaces that argument with one number: multiply Reach, Impact and Confidence, then divide by Effort, and rank the list. Treat that ranking as a starting point you can still override. Below: the rice framework formula, and how to calculate a rice score for each input. What follows scores a full rice framework example across five ideas, including a tie, and ends with a rice prioritization comparison against ICE and MoSCoW for teams deciding which one fits. ## What is the RICE framework? RICE is a backlog-scoring formula that ranks competing ideas by multiplying Reach, Impact and Confidence, then dividing by Effort, so every idea gets one comparable number instead of a gut feeling. Sean McBride built it at Intercom and published it on 2018-01-05, a year after Sean Ellis's ICE framework, not the other way around, a chronology worth fixing since it gets reversed often. The four inputs are Reach (how many people or events the idea touches), Impact (how much it moves the outcome), Confidence (how much evidence backs the estimate) and Effort (the person-months it costs). Multiply the first three, divide by the fourth, and you get a score every idea in the backlog can be sorted by. Teams adopt it for three concrete reasons. It's structured: one repeatable calculation replaces a debate about which idea "feels" biggest. It's data-driven: the inputs are estimates a team can defend with real numbers. And it's defensible: a number a PM can point to in a roadmap review does more work than an opinion. RICE isn't limited to product roadmaps, either. Teams have used it to rank a marketing group's own partnership ideas and to prioritize a state labor department's plain-language rewrite program, both outside a traditional product backlog. A prioritization framework, in general, is just a repeatable rule for ranking more ideas than a team can build. RICE is the version that makes you estimate reach and evidence quality explicitly instead of leaving them implicit. ## How to calculate a RICE score A RICE score comes from four inputs, Reach, Impact, Confidence and Effort, each estimated on its own scale before they're combined into one number. The formula is: **(Reach x Impact x Confidence) / Effort** Get each input right first. The scale you pick for Impact especially decides how much one massive idea can outrank a pile of small ones, and that's where most scoring mistakes start. ::figure{src="/blog/figures/rice-framework-1.svg" alt="RICE framework formula diagram showing Reach, Impact and Confidence multiplied together and divided by Effort to produce a single backlog score." caption="The RICE formula multiplies Reach, Impact and Confidence, then divides by Effort, to produce one comparable score." width="720" height="230"} ### Reach Reach is the number of people or events the idea touches in a fixed period, stated as a raw count. A feature that affects 400 signups a quarter has a Reach of 400; a feature that changes every session in a month is counted per session, not per user, if sessions are what moves. Pick one time window, a quarter or a month, and use it for every idea in the same backlog. Mixing windows makes the scores stop being comparable. ### Impact Impact is how much the idea moves the outcome for each person or event it reaches, scored on a fixed scale from minimal to massive. Intercom's original scale runs 3 (massive), 2 (high), 1 (medium), 0.5 (low) and 0.25 (minimal), a 12x range between the top and bottom tier. That range matters more than it looks. Three scales are in circulation, and they compress differently: Intercom's 0.25-3 scale spans 12x, ICE's conventional 1-10 scale spans 10x, and Hustle Badger's suggested 0.1-100 logarithmic scale spans 1000x. A backlog scored on the 1000x scale lets one massive idea outrank dozens of small ones in a way a 12x scale never allows. Which scale your team picks is itself a decision that shapes every ranking after it. ::figure{src="/blog/figures/rice-framework-3.svg" alt="Impact scale ranges compared across Intercom, ICE and Hustle Badger show Hustle Badger’s 0.1 to 100 scale spanning 1000x versus 12x and 10x." caption="Hustle Badger’s 0.1-100 Impact scale spans 1000x, far past Intercom’s 12x and ICE’s 10x ranges." width="720" height="222"} If Impact is the input you're least sure how to size, don't guess. For an SEO-driven idea, run the numbers through the [SEO ROI calculator](https://missiongrowth.io/tools/seo-roi-calculator) first and let a computed lift stand in for a guessed tier; [the SEO ROI formula](https://missiongrowth.io/blog/seo-roi) walks through the same math behind that tool in more depth. Railsware's own case tagged each RICE idea by [which part of the funnel it moves](https://missiongrowth.io/blog/pirate-metrics), from acquisition to referral, one way to add context to Impact without changing the formula itself. ### Confidence Confidence is how much evidence backs the Reach and Impact numbers, expressed as a percentage tier rather than a false-precision guess. Intercom's scale runs 100% (high confidence, backed by real data), 80% (medium), 50% (low), and anything below 50% is labeled a "moonshot." Each percentage converts to its own share of one inside the multiplication, but talk about it as a percentage everywhere else. Confidence is the input teams over-trust most; the AI-features section below names exactly when to discount it. ### Effort Effort is the total person-months the idea costs across every team involved, the one input that shrinks the score instead of growing it. Intercom's convention: anything under a month of work scores a flat 0.5, so a two-week fix doesn't get an artificially tiny denominator that inflates its score past ideas that took real estimation. Sizing Effort accurately matters most for a resource-constrained team. A small team weighing [how long does SEO take](https://missiongrowth.io/blog/how-long-does-seo-take) against a backlog of other work needs an honest Effort number, because a wrong one costs a bigger share of the roadmap. ## A full RICE-scored backlog, worked example A five-idea backlog scored end to end shows what a real RICE ranking looks like, including what happens when two scores land on the exact same number. Every number below is illustrative, built for this post to show the mechanics. ::dataset{key="rice-worked-backlog" name="Worked RICE-scored backlog, five illustrative ideas"} | Idea | Reach | Impact | Confidence | Effort | RICE score | |---|---|---|---|---|---| | Weekly usage digest email | 8,000/month | 0.5 | 80% | 1 person-month | 3,200 | | Bulk CSV export | 1,200/month | 1 | 100% | 0.5 person-month | 2,400 | | Redesigned onboarding checklist | 3,000/month | 2 | 80% | 2 person-months | 2,400 | | In-app referral prompt | 5,000/month | 1 | 50% | 1.5 person-months | 1,667 | | SSO login for the enterprise plan | 400/quarter | 3 | 100% | 4 person-months | 300 | ::figure{src="/blog/figures/rice-framework-2.svg" alt="Five worked RICE scores rank from a weekly usage digest email at 3,200 down to an enterprise SSO login at 300, with two ideas tied at 2,400." caption="A weekly usage digest email outranks the other four ideas in this worked backlog, while two ideas tie for second." width="720" height="315"} Work the arithmetic once so the pattern is clear: - **Weekly usage digest email:** 8,000 × 0.5 × 80% ÷ 1 = 3,200 - **Bulk CSV export:** 1,200 × 1 × 100% ÷ 0.5 = 2,400 - **Redesigned onboarding checklist:** 3,000 × 2 × 80% ÷ 2 = 2,400, the same score as the CSV export, by coincidence, not by any shared input - **In-app referral prompt:** 5,000 × 1 × 50% ÷ 1.5 = 1,666.7, rounded to 1,667 - **SSO login for the enterprise plan:** 400 × 3 × 100% ÷ 4 = 300, last despite the top Impact tier, because a small Reach and a large person-month cost pull the other way Two ideas can tie on the same RICE score for very different reasons, and the tie only breaks once you know which input is the shaky one. Here, the CSV export's Confidence is a real 100%, backed by direct usage data. The onboarding checklist's 80% Confidence is closer to a guess. Recompute the checklist at a more conservative Confidence of 60%: 3,000 × 2 × 60% ÷ 2 = 1,800, which drops it below the CSV export's 2,400. The ranking flips once the shaky input moves, so the tie was never real. That test, not a coin flip, is what breaks it. ## Common RICE mistakes, and when to override the score Most RICE mistakes come from trusting the number past the point where the inputs are still solid, and the fix is a short list of conditions that justify working out of score order. Three mistakes show up most often: - **Treating Confidence as a formality.** A team fills in 80% by habit rather than by asking what evidence actually backs the number, which erases the one input meant to catch a weak estimate before it inflates the whole score. - **Picking an Impact scale and never revisiting it.** A 1000x scale and a 12x scale produce different backlogs from the same raw judgment, so switching scales mid-project without re-scoring the whole backlog silently reorders it. - **Reading a low score as "skip it."** A low RICE score does not always mean an idea is wrong; it can be blocking a dependency or covering a contractual table-stakes requirement that the formula was never built to weigh. Override the top-ranked score for exactly three conditions, never for "it feels more exciting": - **A hard dependency.** A lower-scored idea unblocks something else on the roadmap, so it ships first regardless of its score. - **A contractual table-stakes item.** Intercom's own guidance calls this out: a required feature can outrank a higher score. - **A live near-tie.** The sensitivity recompute above still shows two scores close enough to flip. Outside those three conditions, the ranking stands. Fold that judgment into [the backlog review rhythm](https://missiongrowth.io/blog/growth-experiment-cadence) your team already runs, rather than re-deciding it from scratch every time a tie shows up. If your team is resource-constrained, the override decision carries more weight. A startup running [SEO on a small budget](https://missiongrowth.io/blog/startup-seo) can't afford to chase a technically-higher score that ignores a dependency; the override rule exists precisely for that situation. ## Scoring AI features and prompts with RICE A brand-new AI feature is exactly the low-information judgment a team has no real data behind yet, which is where a self-rated Confidence score is least trustworthy. The fix is discounting the number rather than inventing a fifth letter. The reasoning comes from judgment research outside product management. Moore and Healy's 2008 study in Psychological Review tested how confidence tracks accuracy as people get more information. Confidence exceeded accuracy at every stage they tested, but the gap was largest when participants were poorly informed: a 55.1% hit rate against a stated confidence interval of roughly 90% before real information arrived. Once real information came in, the gap narrowed, not closed: hit rates of 85.6% and 84.4% at later stages, still measurably below the stated confidence. Overprecision, the researchers' own term for this, shrinks as information accumulates. It never fully closes. The same variable drives both results: how much real information backs the estimate. Applied to RICE: cap a first-of-its-kind AI feature's Confidence at "medium" (80%) by default, regardless of how sure the team feels, until a working prototype's real usage data replaces the guess. Don't fully trust even a well-informed Confidence score afterward, since Moore and Healy's own informed-phase participants were still overconfident. The transfer has a real limit: the underlying study ran a general-knowledge quiz, a reasoned application to RICE rather than a RICE-specific replication. It would be disproved if teams scoring low-information features turned out well-calibrated in practice, with no real gap against better-informed estimates. A first AI or ML feature with no comparable shipped work is the textbook low-information case: no prior data, no comparable feature to anchor the estimate against. One practitioner has proposed a fifth multiplier for this exact problem: RICE-A adds an "AI Complexity" factor for the data, training and deployment overhead specific to AI features. It's a single essay, one person's idea rather than an established convention, and the discounting rule above solves the same problem without adding a new axis to a formula every stakeholder already understands. ## How often to update RICE scores Update a RICE score on a trigger, not a calendar: when a new fact replaces a guess, when any single input moves by a full tier, or when two scores are close enough that the ranking could flip. A fixed "review quarterly" rule misses the week a score actually goes stale; a trigger rule catches it the day it happens. Three triggers cover it: - **The fixed backlog-review cadence.** Re-score every idea at [the same backlog-review rhythm](https://missiongrowth.io/blog/growth-experiment-cadence) your team already runs for weekly and monthly planning, rather than a separate process just for RICE. - **Any single input moves a full tier because new data replaced a guess.** A Confidence estimate scored at 50% moving to 100% once real usage data arrives is enough on its own to change the ranking. - **Two scores are close enough to flip before you finalize the ranking.** Run the sensitivity recompute from the worked example above; if the order changes, treat it as unresolved rather than trusting whichever number happened to land first. ## RICE vs ICE vs MoSCoW: how to choose RICE, ICE and MoSCoW answer the same question with different amounts of math, and the right one depends on whether you have a Reach number worth estimating at all. Framed as rice vs ice, the difference comes down to one input: ICE skips Reach for speed, RICE adds it back for a defensible score. ::figure{src="/blog/figures/rice-framework-4.svg" alt="RICE, ICE and MoSCoW compared side by side shows only RICE requires a real Reach number before scoring, while MoSCoW skips scoring entirely." caption="RICE is the only one of the three that requires a real Reach number before it scores anything." width="720" height="317"} One common claim calls ICE "just a shortened version of RICE," using the formula "(Impact x Confidence) / Ease." Both are backwards. ICE came about a year before RICE, and multiplies all three of its inputs, Impact, Confidence and Ease, rather than dividing by one of three. Sean Ellis formalized it with Morgan Brown in *Hacking Growth*, first published 2017-04-25. It skips Reach entirely, which makes it faster for early teams with more ideas than they have traffic to measure. For the full mechanics and a worked example, see [ICE prioritization](https://missiongrowth.io/blog/growth-experiment-cadence), which covers ICE and PIE in depth; this post doesn't re-teach them here. MoSCoW skips scoring altogether. Dai Clegg created it in 1994, sorting requirements into four buckets: Must-have, Should-have, Could-have and Won't-have. It's the right call when a team can't estimate Reach with a straight face, such as early scoping conversations where the backlog is still requirements rather than shippable ideas. - **Best for RICE:** teams that can put a real number on Reach and want a score that's defensible in a roadmap review. - **Best for ICE:** early teams moving fast with more ideas than traffic to size Reach against. - **Best for MoSCoW:** requirement-gathering conversations before there's anything to estimate Reach or Effort for. A RICE score is a starting rank you're meant to question. Watch for the two places the number is most likely to mislead you: an Impact scale you never questioned, and a Confidence rating scored with too little real information. Use the sensitivity recompute above before you trust a near-tie. Score your own backlog with the worked table as a template. Run the tie-break test on whatever two ideas land closest, and only override the ranking for a dependency, a table-stakes requirement, or a confirmed near-tie, never a hunch. ## FAQ ### What does RICE stand for? RICE is an acronym for Reach, Impact, Confidence and Effort, the four backlog inputs multiplied and divided into one comparable score. Sean McBride published the framework at Intercom on 2018-01-05. ### What does the RICE score formula look like? The formula is (Reach x Impact x Confidence) / Effort. Reach is a count of people or events touched, and Impact and Confidence are scored on fixed scales. Effort, measured in person-months, is the input that shrinks the score rather than growing it. ### How often should you update a RICE score? Update on a trigger, not a calendar: when a guess gets replaced by real data, when any single input moves a full tier, or when two scores are close enough that the ranking could flip. A quarterly-only review misses the week a score actually goes stale. ### Is RICE better than ICE? Neither is strictly better. ICE skips Reach and suits early teams with more ideas than they have traffic to measure against; RICE suits teams that can estimate Reach with real numbers and want a more defensible score. See [the full ICE and PIE mechanics](https://missiongrowth.io/blog/growth-experiment-cadence) for more on when ICE fits better. ### Can RICE be used outside product management? Yes. Teams have used it to rank a marketing group's own partnership ideas and to prioritize a state labor department's plain-language rewrite program, both outside a traditional product roadmap. The same four inputs apply anywhere ideas compete for limited effort. ### What is RICE-A? RICE-A is one practitioner's proposal to add a fifth "AI Complexity" multiplier for the data, training and deployment overhead of AI features. It's a single essay, not an established convention. Capping Confidence at "medium" for first-of-its-kind AI features solves the same problem without a new axis. If you do use RICE-A, score Reach, Impact, Confidence and Effort as usual and apply the AI Complexity multiplier to AI features only, so the rest of the backlog stays comparable. ### Is there a RICE framework template for Excel? A RICE template is four input columns and one formula column. In Excel or Google Sheets, put Reach, Impact, Confidence and Effort in columns B to E, enter `=B2*C2*D2/E2` in column F, copy it down and sort by F. Enter Confidence as a percentage so the formula reads it as a fraction.