ProductBeginner5 min read

ICE Scoring

ICE is a lightweight prioritization framework popularized by Sean Ellis for ranking growth experiments: Score = Impact × Confidence × Ease. Each input is rated 1-10 (or 1-5). Unlike RICE, ICE has no Reach term — it assumes most growth experiments touch a similar audience. Unlike RICE's person-months Effort, ICE flips it to Ease (high score = easy). The output is a single number used to triage 20-50 experiment ideas in a backlog. ICE is built for speed: a team can score 30 ideas in 20 minutes. The framework's strength is also its weakness — it's fast because it's vague, which makes it easy to game.

Also known asICE ScoreImpact Confidence EaseSean Ellis ICEGrowth Hacking ICE

Challenge a friend Browse library

The Trap

The trap with ICE is that all three inputs are wishful thinking dressed up as numbers. A team scores their idea Impact 9, Confidence 8, Ease 7 = 504. Their colleague's idea gets 7×6×6 = 252. The first idea 'wins' but no one has cited evidence for any score. ICE rewards the most enthusiastic advocate, not the highest-leverage experiment. Second trap: ICE compresses everything into a single comparable number, which hides disagreement. Two scorers averaging Impact 8 and 2 = 5 looks identical to two scorers both saying 5 — but the disagreement signals a missing experiment, not a settled answer.

What to Do

Use ICE for fast triage, then graduate the top candidates to a stricter framework. Process: (1) Score 1-10 on each axis, individually, no discussion. (2) Reveal scores — flag any input with >3 points of disagreement for discussion. (3) Re-score after discussion. (4) Top 5 by score get one-page experiment briefs with hypothesis, success metric, and decision criteria. Skip steps 2-4 and ICE becomes a popularity contest. Time-box the scoring to 20 minutes so people don't optimize for politics.

Formula

ICE Score = Impact (1-10) × Confidence (1-10) × Ease (1-10)

In Practice

Sean Ellis — the marketer who coined the term 'growth hacking' while at Dropbox — popularized ICE through GrowthHackers.com and his book Hacking Growth. Ellis's argument was that growth teams need a triage framework that takes 30 seconds per idea, not 30 minutes. His original ICE used a 1-10 scale with deliberately vague anchors because the framework's job was to surface the obvious top 5 from a pile of 50 — not to produce a precision ranking. Ellis explicitly warned against treating ICE as a decision-maker: 'It's a discussion starter, not a decision tool.' Most teams ignore that warning. (Source: Sean Ellis & Morgan Brown, Hacking Growth, 2017)

Pro Tips

01
Anchor each axis with concrete examples before scoring. Impact 10 = '10x improvement to a north star metric.' Impact 5 = 'Measurably moves a secondary metric.' Impact 1 = 'Vanity win at best.' Without anchors, scorers drift toward 7-8 on everything because no one wants to look pessimistic.
02
Track post-experiment outcomes against ICE scores in a spreadsheet for 3 months. The correlation will be embarrassingly low at first. The exercise of measuring it is what calibrates the team — scorers learn their personal biases.
03
Use ICE for growth experiments (small, reversible, fast to test). Use RICE for product features (larger, harder to undo, slower to validate). Mixing them muddies both frameworks.

Myth vs Reality

Myth

“ICE and RICE are basically the same — pick whichever you like”

Reality

They're built for different decisions. ICE is a 30-second triage for experiments where Reach is similar across ideas. RICE is for committing engineering capacity to features where Reach varies wildly (a feature for enterprise admins reaches 5% of users; a feature for free-tier signup reaches 100%). Using ICE for product roadmap decisions hides the Reach difference and overweights novelty.

Myth

“Higher ICE scores produce better outcomes”

Reality

Multiple studies of growth team backlogs show ICE-rank correlation with measured experiment lift is roughly 0.2-0.4 — barely above random. The framework's value isn't predictive accuracy; it's that it forces the team to compare ideas on the same axes. A score of 400 vs 300 is noise. A score of 400 vs 80 is signal.

Try it

Run the numbers.

Pressure-test the concept against your own knowledge — answer the challenge or try the live scenario.

🧪

Knowledge Check

Your team scored 5 growth experiments with ICE. The winner scored Impact 9, Confidence 9, Ease 9 = 729. The runner-up scored Impact 6, Confidence 8, Ease 7 = 336. What's the most likely truth here?

Industry benchmarks

Is your number good?

Calibrate against real-world tiers. Use these ranges as targets — not absolutes.

ICE Score (1-10 scale on each axis)

Growth experiment backlog — typical SaaS

Top tier — run this week

> 400

Solid candidate

200-400

Marginal — re-examine inputs

80-200

Skip

< 80

Source: Sean Ellis / GrowthHackers community norms

Real-world cases

Companies that lived this.

Verified narratives with the numbers that prove (or break) the concept.

📈

Sean Ellis / GrowthHackers

2010s

success

Sean Ellis introduced ICE as part of the growth hacking methodology. His teams at Dropbox, LogMeIn, and Eventbrite used ICE to triage hundreds of experiment ideas per quarter. The framework was deliberately simpler than RICE because growth teams needed to ship 5-10 experiments per week — too fast for deep prioritization. Ellis's key innovation was pairing ICE with a strict experimentation cadence: every Monday, score 20-30 ideas; every week, ship the top 3-5; every Friday, review measured outcomes against scores. The framework was ritual, not arithmetic.

Scoring time per idea

~30 seconds

Typical weekly scoring volume

20-30 ideas

Experiments shipped per week

3-5

Score-to-outcome correlation

Low (~0.3)

ICE works when paired with disciplined post-experiment measurement. Without the feedback loop, scores drift toward optimism and the framework becomes ceremony. With the feedback loop, scorers calibrate and the top-5 selection becomes meaningful within 2-3 quarters.

Source ↗

Related concepts