Decision-making & prioritisation
RICE Prioritisation
A prioritisation score developed on Intercom's product team that ranks competing ideas by multiplying Reach, Impact and Confidence and dividing by Effort, so that quiet, high-value work can beat the loudest request in the room.
Also known as RICE score, RICE scoring model. First set out by Sean McBride (Intercom) in 2016; the primary source is cited in full below.
- Format
- Scoring model
- Level
- Product · Team
- Best for
- Prioritise · Evaluate options · Allocate resources
- Decision stage
- Decide · Plan
- Difficulty
- Introductory
- Time to apply
- An hour or two to score a twenty-item backlog once reach data and effort estimates exist; gathering those inputs is the real work.
Plate · The model
The components
Reach
How many people or events the initiative touches in a defined period. Reach is estimated in real units, customers per quarter or transactions per month, and is the factor most anchored in measurable data.
Signals of strength
Expressed in a countable unit per named period · Drawn from product analytics rather than intuition · The same period used for every item being compared · Counts the people affected, and no one else
Impact
How much the initiative moves the goal for each person reached, scored on a coarse tiered scale from 0.25 (minimal) to 3 (massive). Coarseness is a feature: impact is judgement, and the scale should admit it.
Signals of strength
Tied to a stated goal, such as conversion or retention · Scored on the agreed tier scale, never invented mid-meeting · Backed by user evidence where any exists · Challenged hardest on the items people love most
Confidence
A percentage discount for how much evidence sits behind the Reach and Impact estimates: 100, 80 or 50 per cent. Confidence is what stops an exciting guess outranking a well-evidenced improvement.
Signals of strength
Uses the fixed 100/80/50 tiers, resisting invented values · High confidence traceable to actual data or research · Anything below 50 per cent flagged as a moonshot · Confidence downgraded when estimates rest on one person's opinion
Effort
Total work required across product, design and engineering, in person-months. Effort is the divisor, so underestimating it flatters a score more than any other error in the model.
Signals of strength
Estimated by the people who would do the work · Whole person-months, minimum a half · Includes design, QA and rollout, and never just the build · Revisited when scope changes, with the score recalculated
When it earns its keep
- A product backlog has grown past the point where the team can hold the trade-offs in its head, and roadmap arguments keep restarting from scratch.
- Ideas of very different shapes are competing for the same quarter: a growth experiment against a feature against an infrastructure fix.
- Prioritisation is currently being settled by seniority or volume, and you need a shared structure that forces estimates into the open.
- You have usage data good enough to estimate how many customers each idea would actually touch in a given period.
And when it doesn't
- The decision is strategic rather than comparative. RICE ranks a list; it cannot tell you whether the list serves the strategy. Set direction first with something like Three Horizons or OKRs.
- The work is non-negotiable: regulatory deadlines, security fixes and contractual commitments should be scheduled, and scoring them wastes everyone's time.
- Dependencies dominate. RICE scores items independently, so where item B is worthless without item A the ranking will mislead unless you score the bundle.
- You have no reach data at all. Without a defensible estimate of how many people each idea touches, the R in RICE is decoration and a simpler impact-effort judgement is more honest.
How to run it
Before starting, gather the inputs the analysis depends on:
- A list of candidate initiatives written at roughly the same granularity, so the comparison is fair.
- Product analytics or customer counts from which Reach can be estimated per period rather than guessed.
- An agreed impact scale and confidence scale, fixed before any scoring starts.
- Effort estimates from the people who would do the work, in a consistent unit such as person-months.
- Someone with the authority to act on the ranking, or at least to explain publicly why it was overridden.
- 1
Assemble comparable candidates
Write every candidate at similar size and specificity. Scoring 'redesign onboarding' against 'change the button copy on step three' produces a ranking of wording, not of value. Split epics or bundle tweaks until the items are honestly comparable.
- 2
Estimate Reach
Estimate how many customers, users or events each item touches per period, using real numbers from your product data: sign-ups per quarter, active teams per month, support tickets per week. Reach is the factor that keeps the model honest, because it is the one you can actually look up.
- 3
Estimate Impact
Judge how much the item moves the goal for each person reached, on Intercom's tiered scale: 3 for massive, 2 for high, 1 for medium, 0.5 for low, 0.25 for minimal. The scale is deliberately coarse. Impact is a judgement, and pretending to a second decimal place only hides that.
- 4
Estimate Confidence
Discount enthusiasm by evidence. Use 100 per cent where you have data for reach and impact, 80 per cent where you have data for one and judgement for the other, 50 per cent where it is mostly judgement. McBride's advice is blunt: anything below 50 per cent is a moonshot, and should be treated as one.
- 5
Estimate Effort
Estimate total work across product, design and engineering in person-months, using whole numbers and a minimum of half a person-month. Get the estimate from the people who would build it. Effort estimated by the person who wants the feature is a wish, not an input.
- 6
Calculate, rank and interrogate
The arithmetic is Reach times Impact times Confidence, divided by Effort, giving value per unit of work. The model extends Sean Ellis's ICE score (Impact, Confidence, Ease), built for ranking growth experiments, by adding an explicit Reach factor and treating Effort as a divisor. Sort by score, then interrogate the surprises: a ranking that merely confirms what everyone already wanted has probably been gamed, and one that produces a shock usually contains either an insight or a bad estimate. Find out which before you commit the quarter.
Reading the result
A ranked list of initiatives, each with a RICE score representing estimated value per unit of effort, plus the four underlying estimates that make every ranking position arguable in the open.
- Large gaps between scores are meaningful; small gaps are noise. Treat items within 10 or 20 per cent of each other as tied and settle them on strategy, sequencing or appetite.
- Read the confidence column before the score column. A high score at 50 per cent confidence is a bet dressed as an answer, and might best be handled by a cheap test to raise confidence first.
- The score ranks efficiency, and efficiency is silent about coherence. Check the top of the list against strategy before committing, and record the reason whenever you override.
A worked example
A recruitment software product team ranks its quarterly roadmap
The product team behind a UK applicant-tracking system for recruitment agencies has eight engineers and four credible candidates for the coming quarter: improving CV parsing accuracy, calendar-integrated interview scheduling, a redesign of the client-facing portal, and automated GDPR data-retention rules. The portal redesign has the loudest internal champions. The team scores all four with RICE, using recruiters reached per quarter as the reach unit.
- Reach
- From product analytics: CV parsing touches every active recruiter, about 5,000 per quarter. Scheduling would be used by roughly 3,000. The retention rules run silently for all agencies, so credit 4,000 recruiters whose compliance exposure changes. The client portal is used by agency-side administrators only, about 800 per quarter, a number the redesign's champions had never actually looked up.
- Impact
- Parsing accuracy scores 0.5: a real but incremental time-saving on every CV. Scheduling scores 1: it removes a daily coordination chore that shows up constantly in churn interviews. The retention automation scores 0.5, valuable but invisible until an audit. The portal redesign scores 2 on the strength of client feedback that the current portal embarrasses agencies in front of their customers.
- Confidence
- Parsing gets 100 per cent: usage data and a completed accuracy benchmark exist. Scheduling gets 80 per cent: reach is measured, impact is inferred from interviews. Retention gets 80 per cent on the strength of the legal team's analysis. The portal redesign gets 50 per cent, because the impact claim rests on anecdote and no one has tested whether a redesign, rather than missing features, is what clients actually want.
- Effort
- Engineering estimates: parsing improvements 2 person-months building on the existing pipeline, scheduling 4 including two calendar integrations, retention automation 3 including legal review cycles, and the portal redesign 6 across design and engineering. The portal number drew protest from its champions, which the team treated as information.
The read. The scores: CV parsing (5,000 x 0.5 x 1.0 / 2) = 1,250; scheduling (3,000 x 1 x 0.8 / 4) = 600; retention automation (4,000 x 0.5 x 0.8 / 3) = 533; portal redesign (800 x 2 x 0.5 / 6) = 133. The loudest request finished last, on measured reach a fraction of what its champions assumed. The team committed parsing and scheduling, scheduled retention automation next, and commissioned a two-week discovery on the portal to raise confidence before spending six person-months on a guess. The score did not make the decision; it made the argument inspectable.
Pitfalls
- Scoring items of wildly different granularity against each other. An epic will usually lose to a tweak on effort alone, which rewards thinking small rather than thinking well.
- Letting advocates estimate their own effort. Effort is the divisor, and shaving it is the easiest way to game the model without visibly lying.
- Running compliance, security or contractual work through the score. Obligations are scheduled, and putting them in the ranking teaches people the ranking is negotiable.
- Reading a 610 as genuinely better than a 590. The inputs carry nothing like that precision; treat near scores as ties and decide them on other grounds.
- Scoring once and filing the sheet. Estimates decay as scope and evidence change, and a stale RICE table confers false authority on decisions it no longer describes.
What the critics say
The confidence factor invites false precision. Itamar Gilad, whose work builds on the related ICE score, argues that self-assessed confidence is close to worthless without an explicit evidence test, since teams routinely assign 80 per cent confidence to ideas backed by nothing beyond opinion. His confidence meter, which ties the percentage to named classes of evidence, is a direct response to how the factor fails in practice.
Gilad, I. 'ICE Scores: All You Need to Know', itamargilad.com. https://itamargilad.com/ice-scores/
The formula multiplies three estimates and divides by a fourth, so errors compound rather than cancel, and modest misjudgements in individual factors can reorder the whole list. Practitioner critiques note that the output inherits all the bias of its inputs while adding an aura of arithmetic objectivity, garbage in, gospel out.
'RICE Ain't So Nice', The Honest Product Manager. https://honestpm.substack.com/p/rice-aint-so-nice
RICE structurally favours safe, incremental work. Novel bets score low on confidence by definition and often low on measurable reach, so a team that follows the ranking faithfully will optimise the existing product while systematically deferring the ideas that could change its trajectory.
Work it through
Score each candidate on the four factors. Reach, Impact and Confidence multiply; Effort divides. Confidence is a percentage, so 80 counts as 0.8. The score is value per unit of work; sort by it, then interrogate the surprises before committing the quarter.
| Item | Reach | Impact | Confidence | Effort | RICE score |
|---|---|---|---|---|---|
| – | |||||
| – |