RICE and Scoring Models
RICE is the most widely used prioritization scoring model in product. It's simple, it forces four useful questions, and it produces a comparable number. It's also the framework I most often see misused, because the number looks objective while being built entirely on estimates. This note is RICE done honestly: the mechanics, and the failure modes that bite teams who trust the output too much.
#The formula
RICE was popularized by Intercom. It scores each initiative on four factors and combines them:
$$\text{RICE} = \frac{\text{Reach} \times \text{Impact} \times \text{Confidence}}{\text{Effort}}$$
| Factor | What it measures | Typical unit |
|---|---|---|
| Reach | How many people/events this affects in a period | # of users per quarter |
| Impact | How much it moves the goal per person | Scale: 3 = massive, 2 = high, 1 = medium, 0.5 = low, 0.25 = minimal |
| Confidence | How sure you are of your estimates | %: 100% high, 80% medium, 50% low |
| Effort | Total work to deliver | Person-months |
The three numerator factors are value-ish; effort is the denominator because you want value per unit of work. The output is a relative score with no inherent meaning. It's only useful for ranking against other items scored the same way. A RICE of 32 means nothing alone; it means something next to a 12 and a 90.
#Why it's good
I keep RICE in the toolkit because it does three things well. It forces the Reach question, since PMs systematically overrate features that delight a tiny segment. The Confidence multiplier is a humility tax that discounts the seductive high-impact idea you have no evidence for, which is the single best thing about the model. And it makes effort a first-class citizen, which kills a lot of "great idea, impossible to build" entries. As a prioritization tool for a backlog of comparable features, it earns its place.
#The failure modes: where RICE lies to you
Here's where I get blunt, because the dangerous thing about RICE is that the number looks like math.
RICE produces a decimal. A decimal implies precision. But Reach is a guess, Impact is a five-point opinion, Confidence is a vibe converted to a percentage, and Effort is engineering's estimate (itself a range). You are multiplying four fuzzy numbers and reporting the result to two decimal places. The output is exactly as reliable as your worst input, and usually less, because errors compound under multiplication.
The specific traps I watch for:
- Gameable inputs. Anyone who knows the formula can reverse-engineer a high score by nudging Impact from 2 to 3 or shaving an effort estimate. If RICE controls the roadmap, people optimize the RICE, not the product. (Goodhart's law, in a spreadsheet.)
- Effort estimates that don't exist yet. You're scoring things you haven't designed, so Effort is frequently wrong by 2-3x. Since it's the denominator, an effort error swings the whole ranking.
- Impact is the soft spot. A 3 vs a 2 is a judgment dressed as data, and it has the largest leverage on score. Two PMs will score the same feature differently and both feel objective.
- It's blind to strategy and dependencies. RICE has no idea about your vision, your defensibility, or that feature B is worthless without feature A. A pure RICE sort will happily rank a strategically irrelevant crowd-pleaser at the top.
- It crushes long-term bets. High-effort, hard-to-estimate, low-confidence strategic investments score terribly, which is exactly the work that builds moats. RICE has a structural bias toward small, safe, near-term wins.
#How I actually use it
I treat RICE as a conversation-starter and a sanity check, never an autopilot. Concretely: score the backlog, sort it, then read the list against the strategy and override with documented reasons. The overrides are the most interesting output. Every time I move a low-RICE item up, I have to articulate why it matters beyond the number, and that articulation is usually the real strategic insight. And I never present RICE to executives as justification, because they should hear the bet and the reasoning, not a spreadsheet's arithmetic.
The most underused factor is Confidence. Instead of just discounting low-confidence items, route them: high impact + low confidence means run a test or discovery sprint before it competes on the roadmap, because scoring an idea you don't understand is theater. RICE then becomes a triage tool for what to learn, not just what to build.
#How AI is changing it
This is squarely in Reforge's "auto-prioritization" bucket. A model will populate a RICE table from your backlog and ticket data in seconds, and that's a real chore eliminated. The risk is that AI makes the false precision worse: a confident-sounding, well-formatted score from a model feels even more authoritative than one you built yourself, and it's no less of a guess. My rule holds and tightens: let AI draft the inputs and do the arithmetic, then spend the saved time on the part that was always the actual work, which is interrogating the Impact and Effort estimates and deciding which high-score items to override because they fight the strategy. The model can't make that call, and it has no stake in defending it.
#Continue Reading
- Prioritization Frameworks is the full map of methods and when RICE is the right one.
- The Impact-Likelihood Matrix is a coarser, faster alternative for portfolio triage.
- OKRs and Goal-Setting for how the goals that define "Impact" should be set.
- Experimentation and A-B Testing for how to raise Confidence before betting effort on an idea.
- Outcomes Over Outputs for why the "Impact" you score must be an outcome, not a shipped feature.