ICE Scoring

How to run it

1. 1

   Compile the candidate list: assemble 10–50 experiments, features, or initiatives. ICE is suited to high-volume lists — for fewer than 10 items, a simple discussion may suffice.

2. 2

   Agree on the metric context (5 min): what is the primary metric these items are intended to move? ICE scores are only comparable within the context of a shared goal. If items serve different objectives, score them in separate sessions.

3. 3

   Score Impact (5–10 min): for each item, ask 'If this works exactly as expected, how much will it move our target metric?' Score 1–10 (1 = negligible, 10 = massive). Go item by item quickly — first instincts are often right here.

4. 4

   Score Confidence (5–10 min): ask 'How confident are you that this will work as predicted?' Score 1–10 (1 = pure speculation, 10 = proven by data or prior results). Flag items where confidence is below 4 — these may be worth running as cheap experiments before committing to.

5. 5

   Score Ease (5–10 min): ask 'How easy is this to implement given our current resources and constraints?' Score 1–10 (1 = very hard, many months of work; 10 = can be done today with no dependencies). This is the inverse of effort.

6. 6

   Calculate ICE scores and rank: compute (Impact + Confidence + Ease) ÷ 3 for each item. Sort descending. Items with scores above 7 are strong candidates for the next cycle.

7. 7

   Review the ranked list and calibrate (10 min): ICE often surfaces quick wins (high Ease) over strategic bets (high Impact, low Ease). Discuss whether the ranked list reflects the team's strategic intent or just optimises for speed. Adjust if needed.

8. 8

   Assign ownership: for the top 5–10 items selected for the next sprint or quarter, assign an owner responsible for scoping and executing the experiment.

Tips

• ICE scoring is fast because it is approximate. Don't let groups debate a score from 6 to 7 for five minutes — that level of precision is false anyway.

• Confidence is the most important filter. A 10/10 Impact score on a 1/10 Confidence item is a research project, not a priority. Sequence low-confidence high-impact items as validation experiments, not full launches.

• Ease scoring often gets gamed by technical team members who score items that favour their current sprint high. Calibrate against actual delivery timelines in the last quarter.

• Run ICE in silence first — everyone scores independently before any group discussion — to prevent anchoring and social conformity bias.

• ICE is a shortlisting tool, not a final decision framework. Always sanity-check the top-ranked items against strategic fit before committing to them.

Variations

For growth experiment programmes, add a fourth factor — 'Time to Results' — to create an ICET model that also accounts for how quickly you'll know if an experiment worked. For strategy-heavy teams, weight Impact more heavily (e.g., 2×) rather than using a simple average. Use ICE for weekly experiment triage and RICE for quarterly roadmap prioritisation.