RICE and ICE scoring: what the numbers mean and where they stop

What it is

The Intercom article that introduced RICE defines four factors. Reach: how many people or events the change affects in a defined period, taken from product metrics where possible. Impact: the effect per person on the goal, on a fixed scale (3 massive, 2 high, 1 medium, 0.5 low, 0.25 minimal). Confidence: how well the reach, impact and effort figures are supported (100% high, 80% medium, 50% low; below that is a "total moonshot"). Effort: person-months across product, design and engineering, in whole numbers with half a month as the minimum. The score is reach × impact × confidence ÷ effort, which the article describes as "total impact per time worked". ICE (impact, confidence, ease) is the lighter variant, commonly described as three scores on a 1–10 scale multiplied together, with no reach term and ease in place of effort.

Why it matters

Without a shared scheme, prioritisation is decided by whoever argues longest. A score forces each proposal to state the same four assumptions, so a disagreement becomes "you think reach is 2,000, I think 400" rather than "I just feel this is important". The article itself says scores should not be a hard rule: dependencies and table-stakes features can legitimately jump the queue, and the scheme makes those exceptions visible.

How to apply

• Fill in reach and effort from data (funnel counts, past similar work) and label the rest as judgement; the confidence factor exists to discount judgement.
• Score in a group, one factor at a time across all items, so the scale is applied consistently; then sort and challenge the outliers.
• Keep the spreadsheet with the date and the assumptions; re-score when a number changes, not when someone wants a different order.
• Use ICE for quick triage of many small ideas and RICE when reach differs by orders of magnitude between candidates.
• Treat the ranking as an input to a decision, alongside strategy, risk and dependencies, and write down when you override it.

Pitfalls

False precision: a score of 1,344 versus 1,290 means nothing when confidence is 50%. Effort as the divisor punishes large but essential work (platform, security, debt), which then never ranks; reserve capacity for it outside the ranking. Impact scales drift over time unless anchored with examples. Reach counted in different units across items (customers versus events) makes scores incomparable. A framework cannot supply the goal; it only orders candidates against one you have already chosen.