ICE Scoring Model: Quick Prioritization Explained

What is the ICE scoring model?

The ICE scoring model is a lightweight prioritization method that rates each idea from 1 to 10 on Impact, Confidence, and Ease, then multiplies the three numbers into a single score used to sort the list. Sean Ellis, who coined the term “growth hacking” and ran growth at LogMeIn and Dropbox, developed it to compare a continuous flow of experiment ideas without turning prioritization into a second job (Ellis describes the origin himself{rel=”nofollow noopener”}).

The formula is one line:

ICE score = Impact × Confidence × Ease

Each factor means something specific:

• Impact — how much this idea would move the one metric you are focused on if it worked. Ellis frames this as funnel math: if this step doubled, what would that do to the business?
• Confidence — how sure you are that the idea will produce that impact. In Ellis’s telling, this leans toward a binary question: will this test achieve its goal or not?
• Ease — how little effort it takes to test or ship, including the cost of getting a real answer rather than a fake one.

Two clarifications prevent most of the confusion that follows. “Ease” measures cost, not importance — a 9 means “we could start this week.” And the score is relative to one list at one moment; comparing it across quarters is meaningless unless the anchors match.

> Important note: Ellis popularized the method through practice, not through a controlled study. There is no canonical peer-reviewed paper behind ICE. Treat it as a disciplined conversation format with a number attached, which is exactly how it delivers value.

How do you calculate an ICE score step by step?

You calculate an ICE score by fixing the scale definitions first, scoring each idea independently, multiplying the three values, and sorting the result. The arithmetic takes seconds; the discipline is in steps one and two.

Step 1: Name the single objective

ICE only works when every idea is judged against the same target. “Increase trial-to-paid conversion from 12% to 15% this quarter” is a usable objective. “Improve the product” is not, because two people will score impact against different imagined goals and the numbers will be incomparable.

At GrowthHackers, one active objective eliminated roughly 95–98% of a 1,500-idea backlog before scoring began, leaving 30–75 candidates each week (Anuj Adhiya’s account{rel=”nofollow noopener”}). That pre-filter matters more than the scoring. Sorting 1,500 ideas is a week of work; sorting 40 is an hour.

Step 2: Write the anchors before you score

Decide what a 3, a 6, and a 9 mean for each factor, in writing, for this objective. Anchors are what stop a session from becoming a negotiation about vocabulary. A workable set for a conversion objective:

Step 3: Score silently, then compare

Everyone scores alone before discussion. Public scoring produces anchoring: the first confident voice sets the number and the rest adjust toward it. Silent scoring then a short reconciliation on the five biggest disagreements takes about half the time and produces better data.

Step 4: Multiply and sort

Multiply the three values. A 10/10/10 idea scores 1,000; the realistic best ideas in a normal backlog land between 200 and 600. Sort descending and look at the top of the list, not the middle.

Step 5: Record the assumption behind each score

For every idea in the top ten, write one sentence explaining the impact model and one explaining the confidence evidence. Six weeks later, when the result arrives, that sentence is how you learn whether your scoring is calibrated or merely confident.

Worked example: two ideas, one list

A B2B SaaS team is trying to lift trial-to-paid conversion. Two candidates reach the shortlist:

The email sequence wins by 4.7×. Read that correctly: it does not mean the email work is 4.7 times more valuable in the long run. It means that, this week, with this evidence and this capacity, the email sequence is the better bet. The redesign may still be the right project next quarter, once the email test has produced evidence that raises the confidence score for the larger change.

Should you multiply or average the three scores?

Multiply if you want trade-offs to bite, because multiplication punishes any weak factor. Average if you need a stable 1–10 number for reporting or for comparing many teams. Ellis’s version multiplies; several tool vendors and teams average instead, and both produce a usable ranking as long as you never mix the two in one table.

Multiplication is not neutral. It changes how much a single point is worth:

• An idea at 7 × 8 × 9 = 504 falls to 7 × 8 × 5 = 280 when Ease drops four points — a 44% cut.
• Raise Impact from 7 to 8 on the original: 8 × 8 × 9 = 576, a 14% gain.
• A 2 on any factor caps the ceiling: 10 × 10 × 2 = 200, below a modest 6 × 6 × 6 = 216.

That last line is the argument for multiplication. An enormous idea nobody believes in should not outrank a decent idea that will certainly ship. Averaging would put 10/10/2 at 7.3 and 6/6/6 at 6.0, reversing the ranking and quietly rewarding wishful thinking.

How is ICE different from RICE?

RICE adds a Reach factor and inverts Ease into Effort, producing a benefit-per-cost ratio rather than a pure ranking. Intercom developed it after using ICE, specifically because reach differences between features were too large to ignore.

The practical test: if two items on your list could plausibly affect audiences that differ by a factor of fifty, ICE will mislead you and RICE will not. If every item touches the same funnel and similar effort, RICE adds data collection for a ranking you could have had in an hour.

When does ICE work well — and when should you skip it?

ICE works when the list is long, the items are small and comparable, being wrong is cheap, and you will run the loop again next week. It fails when any of those conditions is missing.

It fits well for:

• Growth, marketing, and content backlogs judged against one metric.
• Small product teams sorting a discovery shortlist.
• Operational improvements inside a single workflow.
• Personal project triage when candidates outnumber evenings.

Skip it, or wrap it in something heavier, when:

• The decision is a single irreversible commitment, such as a platform migration or a factory build.
• Items differ enormously in size, so Ease is no longer comparable.
• Dependencies decide the order; nothing in the formula represents “B cannot start until A ships.”
• A stakeholder will challenge the ranking formally and you need a defensible model, not a defensible conversation.

What are the most common ICE mistakes?

The most common ICE mistakes are score clustering, confidence inflation, inverted ease scales, and treating the output as a decision rather than an agenda. Each one is predictable, which means each one is preventable.

Everything lands between 5 and 8. Without anchors, scores collapse toward the middle and the ranking is decided by rounding noise. Fix: anchor the scale against the strongest and weakest item on the list before scoring the rest, then force a spread.

Confidence becomes enthusiasm. Teams routinely assign high confidence on the strength of opinion, which turns ICE into a popularity ranking. Itamar Gilad built a “confidence meter” to counter this by tying numbers to evidence classes: opinion, secondary research, user testing, and A/B result each map to a fixed band. The effect is that an appealing idea nobody has checked gets demoted automatically.

Ease is scored as effort. Someone writes “effort 8” meaning “expensive,” and because Ease is inverted, that row silently reverses. Write “10 = easiest” at the top of the sheet in bold.

Disagreement gets averaged away. If two people score Confidence 9 and 5, averaging to 7 erases the most useful information in the room. Score independently, then discuss only the gaps.

The score becomes the decision. A ranking is executed, the assumptions are never revisited, and the team learns nothing. ICE should produce an agenda for a short conversation, not a verdict.

The list itself goes unquestioned. An efficient ranking of the wrong twenty ideas is still the wrong twenty ideas.

> Common mistake: Scoring twenty ideas that all sound attractive and then feeling disappointed when the top five fail. The failure was upstream: no evidence entered the confidence column, so the ranking measured plausibility, not probability.

Example: a calibration loop that improved the scores

A six-person team ran ICE on 24 ideas against one objective: increase activation within seven days. The first session took 75 minutes and produced five experiments, three of which shipped. Three months later, ideas scored 8–9 on impact had delivered 1–4 percentage points of lift, while two ideas scored 5–6 had each delivered more than 6 points. They had been rewarding ideas that sounded strategic and discounting small friction fixes, so they rewrote the impact anchor to require funnel math for any score above 7.

That loop — predict, ship, compare, adjust anchors — separates a team that gets better at prioritization from one that just gets faster at ranking.

How do you run a 60-minute ICE session?

You run it in five blocks: context, anchors, silent scoring, focused discussion, and commitment. The structure matters more than the tool you use.

Two rules keep the hour honest. Nobody may raise a score without naming the evidence that justifies it, and the session ends with a decision about what to test, not a promise to “revisit the backlog.” A backlog that gets re-ranked but never shipped is a very organized way to avoid work.

Does ICE work outside growth marketing?

Yes, with one adaptation: replace “impact on conversion” with the outcome your team is accountable for. Hiring becomes impact = how much the role unblocks a committed deliverable, confidence = quality of evidence for the need, ease = time to fill. Operations becomes hours saved per month, whether a pilot exists, and implementation days. Content becomes qualified sessions, prior performance of similar pieces, and production hours.

The tolerance for error does not transfer. In a marketing backlog, a mis-ranked experiment costs a week; in a regulated launch or a hiring decision with a six-month ramp, a rough score is not evidence and should not be presented as one.

How do you keep ICE honest over time?

You keep it honest by keeping a prediction log and reviewing it on a fixed cadence. Every scored item that ships gets three fields: predicted impact, actual impact, and the reason for the gap. After two or three cycles, patterns appear that no amount of debate could have produced.

A simple evidence ladder for the Confidence column does most of the work:

The ladder has a side effect that matters: ideas scoring low on confidence but high on impact become obvious candidates for cheap validation rather than a full build.

Where should the scores and the work live?

ICE produces a decision, and decisions decay fast when they are separated from the work they authorize. A spreadsheet handles the ranking fine; the failure happens afterwards, when the top three ideas move into a task board with no link back to the assumption that justified them.

That gap is what Doitify addresses: experiment ideas become tasks with owners and dates, and the predicted-impact note sits next to the task instead of in a separate scoring file. Doitify’s boards, sub-tasks, checklists, and progress reports keep the ranking and the execution in one workspace, which is what makes the prediction log sustainable. Doitify is our product, so we have an obvious interest here — and it is still true that a team already living in a well-maintained spreadsheet does not need to move anything to run ICE properly.

Conclusion

The ICE scoring model earns its place by making prioritization cheap enough to repeat. Three numbers, one multiplication, and a ranked list in under an hour — with the reasoning written down where the team can inspect it.

Use it as triage: filter by one objective, anchor the scales in writing, score alone, and discuss only the disagreements that matter. Keep the prediction log, because that is the only mechanism that turns a scoring ritual into judgment. And step up to RICE or a written business case when reach varies widely, dependencies set the order, or being wrong is expensive.

If this post on ICE scoring model was helpful, you might also enjoy Free Project Meeting Notes Template and What Is a Gantt Chart? Complete Guide.