Scores and ranks product initiatives using the RICE

RICE Prioritisation Skill

Apply consistent, criteria-based RICE scoring to a list of features or initiatives to produce an objective prioritisation ranking.

Reads from / Writes to the Brain

If a professional-brain (brain/) exists, ground in it instead of re-asking for what you already know:

• Read first: knowledge/strategy.md (so the ranking serves the direction), the items as entities/, and impact hypotheses/. Run python3 ../professional-brain/scripts/brain_query.py ./brain "<initiative theme>" and carry each fact's provenance tag through — an impact estimate is usually a [hunch], not [data].
• 📥 Propose to the Brain: after producing, propose recording the ranking decision to decisions/ and the reach/impact estimates as hypotheses/ tagged by evidence strength. Show them, get a yes, then write with ../professional-brain/scripts/brain_write.py … --commit (append-only, dry-run by default).

Required Inputs

Ask the user for these if not provided:

• List of initiatives or features to score (names and brief descriptions)
• Reach estimates (users affected per quarter — from analytics if available)
• Impact estimates (use the standard scale below)
• Effort estimates (person-months — from engineering if available)
• Quarter or planning period

RICE Definitions (adapt to your context)

• Reach: Number of users affected per quarter (use actual DAU/MAU data where available)
• Impact: Effect on your primary metric — use scale: 3=massive, 2=high, 1=medium, 0.5=low, 0.25=minimal
• Confidence: How certain are we about R and I estimates? 100%=high, 80%=medium, 50%=low
• Effort: Person-months required across all functions

RICE Formula

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

Programmatic Helper

This skill ships with a stdlib-only Python script that calculates and ranks RICE scores so the maths is consistent and the quick-win / moonshot flags are applied by rule, not by feel. Feed it the initiatives once R, I, C, and E are gathered.

# From a JSON file (confidence accepts 0.8 or 80)
python3 scripts/rice_calculator.py initiatives.json

# Or from a CSV with header: name,reach,impact,confidence,effort
python3 scripts/rice_calculator.py initiatives.csv --format csv

# Or piped in
echo '[{"name":"Onboarding","reach":5000,"impact":2,"confidence":0.8,"effort":3}]' \
  | python3 scripts/rice_calculator.py -

It outputs a ranked table with computed RICE scores and auto-flags quick-win (strong score, low relative effort), moonshot (high impact, high effort), and low-confidence (≤50%) items. Use the computed ranking as the starting point, then apply the validation step below — never accept a surprising top rank without checking the estimates behind it.

Deeper Materials

• references/estimate-calibration.md — how to anchor each of the four estimates (reach sources, the impact scale with reserve-it-for examples, evidence-based confidence, cross-functional effort) and the cross-checks to run on the finished ranking. Apply it when challenging the user's inputs.
• templates/scoring-worksheet.md — a fill-in worksheet whose evidence columns force each score to name its source. Offer it when a team wants to score together rather than have the ranking generated.

Where this sits — scoring on the spine

Third in the product-decision spine: /assumption-mapper → /prd-template → rice-prioritisation → /roadmap-narrative. It receives the success metric from each initiative's PRD — RICE's Impact is the estimated move on that baselined number, not a fresh guess — and hands /roadmap-narrative the ranked initiatives with their scores to group into themes. The four RICE terms are defined once in docs/craft/product-decisions.md; Confidence there is the honesty valve, and this skill lives or dies on using it.

The loop

RICE fails when estimates are invented to produce a desired ranking. The loop's job is to keep every score honest; Phase 2 is where that happens.

1. Gather the four estimates per initiative. Reach (real count per period), Impact (magnitude on the PRD's success metric), Confidence (0–1), Effort (person-months). Pull Impact from the upstream PRD's metric where it exists. Done when: every initiative has all four, and each carries a provenance tag on its source.
2. Interrogate confidence — the anti-gaming phase. For each estimate, confidence must reflect evidence, not enthusiasm: a bold impact with no data gets a low confidence, and the score self-corrects. Challenge weak inputs and name what data would raise them (the disclosed estimate-calibration reference is the how). Done when: no [hunch] estimate wears a high confidence, and the person who owns the estimate would defend each number out loud.
3. Score, rank, and stress the top. Compute RICE, rank, flag quick wins (high score, low effort) and moonshots (high impact, high effort), note dependencies. Then the cross-check: if the top item surprises the team, an estimate is probably inflated — RICE is a tool, not a verdict. Done when: the ranking is computed and the top result has survived one honest "does this feel right, and if not, which estimate is lying?"
4. Hand off. Pass the ranked table (with scores and dependencies) to /roadmap-narrative so it groups by theme rather than re-deriving priorities. Done when: /roadmap-narrative could theme these without re-scoring.

Output Structure

RICE Prioritisation: [Backlog/Quarter]

Initiative	Reach	Impact	Confidence	Effort	RICE Score	Notes
[name]	[n]	[score]	[%]	[months]	[score]	[flags]

Recommended Sequence

[Top 5 initiatives with rationale]

Quick Wins (high score, low effort)

[Items to pick up alongside bigger bets]

Data Gaps to Address

[What information would most improve scoring accuracy]

Scoring Rubric (0–40)

Score any output of this skill before handing it over; 32+ is ship-quality.

Dimension	0	5	10
Estimate credibility	Round-number guesses at 100% confidence; effort estimated by PM alone	Reach grounded in analytics but confidence uniform across items regardless of evidence	Each estimate names its source; anything without data sits at 50% confidence; effort comes from engineering, and the doc says so
Impact discrimination	Everything scored 2–3 — the scale produces no signal	Some spread across the scale but anchors undefined, so scores aren't comparable	Full scale used with a stated anchor for each level; "massive" reserved for genuinely rare items
Ranking interrogation	Raw sorted output accepted as the verdict	Quick wins and moonshots flagged, but surprising ranks and dependencies unexamined	Surprising top ranks investigated with the inflated estimate found or defended; dependencies noted where they change sequencing
Actionable sequencing	A scored table with no recommendation	Table plus a top-5 list, but no rationale or data-gap follow-ups	Recommended sequence with per-item rationale, quick wins slotted alongside bigger bets, and named data gaps that would sharpen the next pass

Quality Checks

• Every initiative has all four RICE components estimated (even roughly)
• Confidence is 50% for anything without data backing (not 100% as a default)
• Quick wins and moonshots are explicitly called out
• Dependencies that affect sequencing are noted
• Any surprising ranking is investigated before accepting it

Anti-Patterns

• Do not default to 100% confidence on estimates that lack supporting data — this inflates scores and misleads planning
• Do not treat RICE scores as a final decision — a ranking that surprises the team must be investigated before it is accepted
• Do not omit effort estimates from engineering — PM-only effort estimates are frequently optimistic and skew results
• Do not forget to note dependencies that would change the sequencing even if RICE scores suggest otherwise
• Do not score every initiative at the same impact level — if everything is "high impact," the framework produces no useful signal