magnus919_agent-skills/product-methodology/references/rice-framework.md

RICE Scoring Framework

A method for prioritizing feature proposals by scoring them across four dimensions: Reach, Impact, Confidence, and Effort. Developed by Intercom (Sean McBride, 2016).

Source: Intercom blog — RICE: Simple prioritization for product managers

The Formula

RICE Score = (Reach × Impact × Confidence) / Effort

Score is unitless — used only for relative comparison. A score of 50 is not "good"; it's better than 30 and worse than 100.

Dimensions

Reach — How many people does this affect?

Measured in users or customers per unit time. Be specific about the time window.

Signal	Illustrative reach	Example
Affects all users	10,000+/quarter	Login flow improvement
Affects a segment	1,000–10,000/quarter	New export format for power users
Affects very few	100–1,000/quarter	Admin UI improvement
Niche edge case	< 100/quarter	Specific error message improvement

The values above are hypothetical and exist only to demonstrate a consistent unit. Use observed product data for the actual population and period. Don't inflate Reach by counting everyone as "affected." A feature must plausibly change their experience to count them.

Impact — How much does it matter to the people it reaches?

A qualitative score converted to a multiplier. Use a standard scale to keep comparisons consistent.

Score	Label	Meaning
3	Massive	Transformative for the target users
2	High	Significant improvement
1	Medium	Noticeable improvement
0.5	Low	Marginal improvement
0.25	Minimal	Barely noticeable

Pitfall: Impact is about the people reached, not total business value. A 0.5 impact × 10,000 reach = 5,000; a 3 impact × 100 reach = 300. The high-reach/low-impact item scores higher. This is by design — it surfaces broad improvements over narrow deep ones.

Confidence — How sure are you about your Reach and Impact estimates?

Expressed as a percentage. Low confidence should penalize the score heavily.

Score	Label	Meaning
100%	High	Direct evidence closely matches the estimate, such as production metrics or a relevant experiment
80%	Medium	Credible but incomplete evidence, such as interviews, surveys, or an analogous launch
50%	Low	Limited evidence or a weak analogy

These confidence values follow Intercom's published scale, but the evidence descriptions are practical interpretations. Don't use 100% merely because the team agrees. Record the evidence behind the value, and treat low-confidence scores as a reason to learn before committing.

Effort — How much time from the full team?

Estimate effort in one consistent unit, such as person-weeks or person-months. Include design, engineering, QA, documentation, and release work. Derive values from the team doing the work; generic size-to-duration tables create false precision.

When to Use RICE vs Other Frameworks

Use RICE When	Don't Use RICE When
You have enough data to estimate four dimensions credibly	Estimates are pure guesses across the board
You're comparing unrelated feature proposals	All options are variations of the same thing
You need to defend priorities to stakeholders	The decision is a quick call between two obvious choices
You're planning a quarter+ roadmap	You're planning a single sprint

Common RICE Pitfalls

Score inflation across the board

All scores drift up over time as teams get optimistic. Reset periodically. If everything scores above 100, your scale has shifted. Re-anchor against a known baseline ("last quarter's top feature was a 47 — does this really beat it by 3x?").

Effort as a schedule, not a cost

Effort should be team-weeks, not calendar-weeks. A feature that takes 2 weeks of engineering work but 4 weeks of calendar time because of dependencies should be scored at 2 weeks. The dependency is a separate concern.

Cherry-picking the easy dimension

If Reach is obviously high, don't pad Impact and Confidence to match. Each dimension should be estimated independently. If you find yourself adjusting Confidence to fit a desired score, you're anchoring.

The denominator trap

Effort is the denominator, so uncertain estimates can materially change ranking. Preserve a range or run a sensitivity check rather than choosing whichever endpoint produces the preferred score.

Example

Item	Reach	Impact	Confidence	Effort	Score
Add CSV export	2,000/q	2 (High)	80%	2 wks	(2000×2×0.8)/2 = 1600
Dark mode	10,000/q	1 (Med)	50%	6 wks	(10000×1×0.5)/6 = 833
Rebuild search	10,000/q	3 (Massive)	20%	12 wks	(10000×3×0.2)/12 = 500

CSV export wins despite lower Reach because high Confidence + low Effort creates a better ratio. The search rebuild has the most potential but needs validation before it's worth the bet.