Why this matters
A healthy growth team has far more ideas than test slots. A typical page can only run one or two clean tests at a time, and each needs one to four weeks. That means a D2C brand might run 30 to 50 tests a year on its key pages, and a lower-traffic SaaS company perhaps 10 to 20. If half of those slots go to the wrong ideas, you lose half a year of learning.
Prioritisation frameworks do not make the decision for you. They make the trade-offs visible and comparable. Without them, the backlog gets sorted by who argues loudest, what a competitor just shipped, or what is most fun to build. With them, you can explain to a founder in one table why the boring free-shipping bar goes before the exciting redesign.
The framework
ICE
ICE was popularised by Sean Ellis, who helped coin the term “growth hacking” and used the method to run high-tempo testing at GrowthHackers. In a talk transcribed by SaaStr, Ellis said that team had a backlog of more than 700 untested ideas and shortlisted three to five to test each week. ICE is how you get from 700 to five without a fight. Each idea gets three scores from 1 to 10.
- Impact: if this works, how much will it move the primary metric?
- Confidence: how much evidence do we have that it will work?
- Ease: how cheap and quick is it to build and test?
ICE is quick. A team can score 30 ideas in an hour. That speed is its strength and its weakness: it has no explicit term for how many people an idea touches.
RICE
RICE was developed by Intercom’s product team to compare very different projects consistently. It adds Reach and turns Ease into Effort.
- Reach: people or events affected per time period, from real data where possible. For example, customers per quarter.
- Impact: effect on each person, on a fixed scale: 3 massive, 2 high, 1 medium, 0.5 low, 0.25 minimal.
- Confidence: 100% high, 80% medium, 50% low. Below that is a moonshot.
- Effort: total person-months across product, design and engineering.
| ICE | RICE | |
|---|---|---|
| Best for | Fast growth experiments on a single funnel | Product and growth roadmaps with ideas of different scope |
| Speed | Minutes per idea | Needs reach data and effort estimates |
| Handles reach | No, unless folded into Impact | Yes, explicitly |
| Biggest risk | Inflated, uncalibrated scores | False precision from made-up reach numbers |
Step-by-step playbook
- Set one goal per backlog. Score against a single primary metric, such as orders per visitor or activated accounts. Ideas aimed at different goals cannot be compared fairly.
- Anchor the scales. Write a one-page scoring guide. For ICE Impact, for example: 3 means under 2% expected lift on the primary metric, 6 means 2 to 10%, 9 means over 10% or a change to a core step. For Ease: 9 means under a day of work, 6 means under a week, 3 means a sprint or more.
- Tie confidence to evidence. Use a ladder. A prior winning test on a similar page caps at 9 or 100%. Strong analytics plus user research caps at 7 or 80%. Best practice or competitor copying caps at 5 or 50%. Gut feel caps at 3.
- Get effort from builders. The engineer or designer who will do the work estimates Ease or Effort. Proposers always underestimate.
- Score as a group, quickly. Each person scores silently, then compare. Discuss only scores that differ by more than 3 points. Take the median.
- Sort, then sanity-check. Look at the top five. If something feels wrong, challenge the inputs, not the formula. Dependencies and must-have features can jump the queue, but write down why.
- Re-score after every result. A test result on page A changes confidence for related ideas on page B. Update them the same week.
- Prune monthly. Archive ideas untouched for two quarters, merge duplicates and cap the active backlog at about 30 items so it stays readable.
Worked example
ICE for a D2C skincare brand
Say a D2C skincare brand in India has five ideas aimed at orders per visitor. Try changing the scores below and watch the ranking move.
Score = Impact × Confidence × Ease
| Idea | Impact 1–10 | Confidence 1–10 | Ease 1–10 | Score | Rank | Remove |
|---|---|---|---|---|---|---|
| 441 | #1 | |||||
| 320 | #2 | |||||
| 294 | #3 | |||||
| 120 | #4 | |||||
| 72 | #5 |
Ranking (highest score first)
- Free-shipping bar at ₹499 on cart 441
- One-tap UPI at checkout 320
- Photo reviews on product page 294
- Skin-type quiz funnel 120
- Homepage hero video 72
The arithmetic:
- Free-shipping bar:
7 × 7 × 9 = 441 - One-tap UPI:
8 × 8 × 5 = 320 - Photo reviews:
6 × 7 × 7 = 294 - Skin quiz:
8 × 5 × 3 = 120 - Hero video:
4 × 3 × 6 = 72
The quiz has high potential impact but low confidence and low ease, so it lands fourth. That is not a verdict that the quiz is bad. It is a signal to raise its confidence cheaply first, for example with a survey asking customers how they choose products.
RICE for a B2B SaaS team
Now a B2B SaaS team with roughly 1,200 new signups a quarter scores five ideas against paid conversions.
Score = Reach × Impact × Confidence% ÷ Effort
| Idea | Reach people / quarter | Impact 0.25–3 | Confidence % | Effort person-months | Score | Rank | Remove |
|---|---|---|---|---|---|---|---|
| 1,440 | #1 | ||||||
| 1,120 | #2 | ||||||
| 960 | #3 | ||||||
| 250 | #4 | ||||||
| 150 | #5 |
Ranking (highest score first)
- Default pricing toggle to annual 1,440
- Upgrade prompts at usage limits 1,120
- Onboarding checklist 960
- Pricing page redesign 250
- Slack integration 150
The arithmetic:
- Annual default:
900 × 1 × 0.8 ÷ 0.5 = 1,440 - Usage-limit prompts:
700 × 2 × 0.8 ÷ 1 = 1,120 - Onboarding checklist:
1,200 × 2 × 0.8 ÷ 2 = 960 - Pricing page redesign:
3,000 × 0.5 × 0.5 ÷ 3 = 250 - Slack integration:
400 × 3 × 0.5 ÷ 4 = 150
The pricing page has the highest reach but scores low because nobody can say what the redesign will change, so impact and confidence are low and effort is high. The fix is to break it into specific hypotheses, each with its own score.
B2B SaaS vs D2C: what changes
| Factor | B2B SaaS | D2C / e-commerce |
|---|---|---|
| Better default | RICE, because ideas range from copy tweaks to integrations | ICE for on-site CRO, RICE for bigger bets like a loyalty programme |
| Reach unit | Accounts or signups per quarter | Sessions or orders per month |
| Impact anchor | Activation or paid conversion | Orders or revenue per visitor |
| Effort reality | Engineering heavy, often weeks | Many tests are theme or app changes, often days |
| Confidence sources | Sales calls, product analytics, churn reasons | Reviews, returns, exit surveys, heatmaps |
Running the scoring meeting
A good scoring system fails if the meeting around it is loose. Use this format every two weeks.
- Pre-score alone. Each person scores new ideas in the sheet before the meeting. Scoring in the room anchors everyone to the loudest voice.
- Discuss only the gaps. Talk only about ideas where scores differ by 3 or more points on any input. Agreement needs no airtime.
- Make people show evidence. Anyone arguing for a higher Confidence score must point to the data, research clip or past test behind it. Opinions keep the default score of 3 to 5.
- Cap the meeting at 45 minutes. Ideas not reached go to the next session with their pre-scores intact.
- Commit the top slots. Fill only as many test slots as your traffic allows. A page that supports one test at a time gets one winner, not three half-powered tests.
Say a team has 40 ideas and traffic for four tests a month. Pre-scoring and gap-only discussion usually leaves eight or ten ideas worth talking about. The rest are either clearly high, clearly low or waiting on evidence. The meeting then decides four slots, not forty rankings.
Re-score the whole backlog once a quarter. Effort estimates go stale as the codebase changes, and Reach shifts with traffic mix. An idea that scored 4 on Ease last spring may now be a one-day job.
How to measure it
- Experiment velocity: tests launched per month. Count only tests with a written hypothesis and a pre-set sample size.
- Win rate: clear wins divided by concluded tests, per quarter.
- Impact per test slot: sum of measured lifts on the primary metric from shipped wins divided by number of tests run. This shows whether prioritisation is choosing high-value bets.
- Score accuracy: for each concluded test, compare predicted Impact band with the measured result. Track the share that landed in or above the predicted band. If it is under a third, your scoring is not predictive yet.
- Backlog age: median days since ideas in the active backlog were last scored. Keep it under 60.