ICE (Impact, Confidence, Ease) Score

ICE is a prioritization method that scores each candidate idea — an experiment, a feature, a campaign — on three dimensions, usually 1 to 10: Impact (how much will this move the metric if it works?), Confidence (how sure are we it will work?), and Ease (how cheap and fast is it to do?). The scores combine (commonly multiplied or averaged) into a single number that ranks the backlog by expected payoff per effort. Popularized in the growth-hacking era (Sean Ellis and the GrowthHackers community), its whole value is being fast and good-enough — a structured estimate that beats arguing, not a precise forecast that pretends to certainty.

How it works and why the three factors are the right three: Impact and Confidence together encode expected value (a high-impact idea you're unsure of and a modest idea you're certain of can score alike — which is correct), and Ease divides by cost so cheap fast wins rise above expensive maybe-bigger ones (the bias toward momentum that early-stage growth programs want). The honest limitations that mature teams price in: the scores are subjective (Impact and Confidence especially are educated guesses, and the same idea scores differently across people — calibration and a shared rubric reduce but never remove this), it's gameable (whoever scores can inflate their pet idea — the GOODHART risk when ICE becomes the political instrument rather than the thinking aid), and the multiply-versus-average choice changes the rankings (multiplication punishes any low factor hard — a sensible default for experiments where a fatal weakness should sink the idea). Where ICE sits among its cousins: it's the lightweight option — RICE adds Reach (and a real denominator), PIE and PXL formalize CRO test scoring, and full expected-value modeling is the heavyweight — so ICE is for speed and volume (ranking many small bets fast, the HYPOTHESIS-TESTING pipeline's intake filter), and the heavier frameworks earn their overhead only on bigger, costlier decisions. The discipline that makes it work: shared rubrics so scores mean the same thing across raters, ICE as an input to judgment rather than an autopilot (the score starts the conversation; it doesn't end it), post-hoc calibration (compare predicted Impact and Confidence to what actually happened — the loop that turns guessing into estimating), and the humility to treat it as the rough triage it is.

ICE matters most where there are more ideas than capacity and the decisions are individually small and reversible — growth experiment backlogs, CRO test queues, quick-win prioritization — where its speed and good-enough structure beat both gut-feel chaos and analysis-paralysis. It matters less, and misleads more, on large irreversible bets that deserve real expected-value work, and wherever it's been captured as a political weapon for justifying predetermined choices. The discipline is shared rubrics, multiply-to-punish-fatal-flaws for experiments, the score as conversation-starter not verdict, calibration against outcomes, and the standing honesty that ICE is a fast rough triage — useful exactly because it doesn't pretend to be more.

ICE comes from the growth-hacking movement - Sean Ellis and the GrowthHackers community popularized it as a way to rank high-volume experiment backlogs fast - and it spread because it formalized just enough (three honest factors) without the overhead of full expected-value modeling; RICE and other cousins later added rigor for bigger decisions.

Etymology: source.

Search interest for this term over the last five years:

View interest-over-time on Google Trends →

Areas of marketing where ice score is a core concern: