Growth experiment template with ICE scoring, a weekly cadence and a backlog you can copy

A growth experiment template is a ranked backlog: every idea gets a hypothesis, an ICE score, one owner and one target metric, and the list is reviewed every week. Score Impact, Confidence and Ease from 1 to 10, average the three and sort. In Growth Method's example, a homepage headline rewrite scored 8.0 and beat a referral program at 6.0, because the team had interview evidence behind it and it was easy to ship.

What a growth experiment template is for, and the short answer

A growth experiment template turns a loose pile of ideas into a ranked queue. The short answer: every idea gets a hypothesis, a score and one owner. The list is sorted by ICE score and reviewed every week, and every idea is judged against the same single metric.

In this worked example, the homepage headline rewrite scores 8.0 and the other three ideas score between 6.0 and 6.7.

ICE scores for four example growth experimentsICE score (1 to 10)

1. Homepage headline rewrite8 ICE score (1 to 10)
2. Exit-intent popup6.7 ICE score (1 to 10)
3. Double-sided referral program6 ICE score (1 to 10)
4. Personalized onboarding rebuild6 ICE score (1 to 10)

ICE scores for four example growth experiments

Item	Value, ICE score (1 to 10)
Homepage headline rewrite	8 ICE score (1 to 10)
Exit-intent popup	6.7 ICE score (1 to 10)
Double-sided referral program	6 ICE score (1 to 10)
Personalized onboarding rebuild	6 ICE score (1 to 10)

ICE stands for impact, confidence and ease. Growth Method describes them as "three factors used to score and rank experiment ideas on a scale of 1 to 10" and combines them as ICE Score = (Impact + Confidence + Ease) / 3. Some templates rate each criterion from 1 to 5 and add them up. In this piece we use the 1 to 10 average, so scores stay on one familiar scale. See our ICE score entry for the definitions.

The weekly review is the core of the method. The Hacking Growth summary advises teams to "Set a weekly heartbeat to encourage experiment velocity, discuss results and what to test next." The same summary notes that "Most experiments fail to produce results, so volume is key."

It also says to choose "a single metric of success that all growth activity is geared toward." In our reading, for most teams that is their north star metric or one step that feeds it.

The four fields every idea needs

• Hypothesis: one sentence that opens with "We believe that", naming the problem, the change and the expected outcome.
• Score: impact, confidence and ease, then the average.
• Owner: one name, the person who ships it and reports the result.
• Metric: the single number the test should move.

A worked ranking

Growth Method scores four common ideas this way:

Experiment	Impact	Confidence	Ease	ICE score
Rewrite homepage headline from customer interviews	8	7	9	8.0
Add exit-intent popup offering a lead magnet	6	6	8	6.7
Launch double-sided referral program	9	5	4	6.0
Rebuild onboarding with personalized steps	9	6	3	6.0

The headline rewrite comes first even though two ideas score higher on impact. The source explains it wins because the team has high confidence (backed by customer interview data) and it is easy to execute. The referral program still has a place in the backlog. It waits until the evidence behind it is stronger.

This week, list every open idea in one sheet. Have each person score independently first, then average the scores, sort the list and put one owner on each of the top three. Treat the result as an order of work: Use ICE for ranking, not precision.

What is an ICE score and how is it calculated?

ICE asks three questions of every idea in your backlog. In the Growth Method definition, Impact is how much the idea will move your target metric, Confidence is how sure you are it will work, and Ease is how quickly you can ship it. You score each factor, combine the three into one number, and sort. Our ICE score glossary entry keeps the short version.

Five example ideas scored on Impact, Confidence and Ease give the same order whether you average the scores or multiply them, from the shorter signup form at 7.3 or 378 down to the referral credit at 5.3 or 108.

ICE scores for five example ideas (average of impact, confidence, ease)ICE score

1. Shorter signup form7.3 ICE score
2. Exit survey on cancellation7 ICE score
3. Annual plan discount on the pricing page6.7 ICE score
4. Onboarding checklist in the product6 ICE score
5. Referral credit for both sides5.3 ICE score

ICE scores for five example ideas (average of impact, confidence, ease)

Item	Value, ICE score
Shorter signup form	7.3 ICE score
Exit survey on cancellation	7 ICE score
Annual plan discount on the pricing page	6.7 ICE score
Onboarding checklist in the product	6 ICE score
Referral credit for both sides	5.3 ICE score

Two ways to combine the scores

The first method is the plain average covered above. It keeps the result on the same scale as the inputs, so a 7.3 reads like a 7.

The second method multiplies the three scores. Savio scores each factor on a scale from 0 to 10 and then applies ICE Score = (Impact * Confidence * Ease).

Here are the same five ideas scored once and combined both ways:

Idea (example)	Impact	Confidence	Ease	ICE (average)	ICE (product)
Shorter signup form	6	7	9	7.3	378
Exit survey on cancellation	4	8	9	7.0	288
Annual plan discount on the pricing page	8	5	7	6.7	280
Onboarding checklist in the product	7	6	5	6.0	210
Referral credit for both sides	9	4	3	5.3	108

Where the rankings agree and where they differ

The order is identical. Both methods put the signup form first and the referral credit last, so this week's first test is the same either way.

The gaps are what change. Under the average, the signup form leads the exit survey by 0.3 and the whole list spans 2.0 points. Under the product, the lead is 90 points and the top idea scores 3.5 times the bottom one.

In our reading, multiplying punishes a single weak factor much harder. The referral credit has the highest Impact on the list, yet its Confidence of 4 and Ease of 3 cut its product to about half of the onboarding checklist's. The average leaves it 0.7 behind. If your team cares about bigger bets, the product can bury them. That makes a referral idea a good one to shrink into a cheaper first test. Our referral program playbook covers how.

The middle of the list is close under both methods: 6.7 against 7.0, or 280 against 288. We would treat ideas that close as a tie and break it on which one teaches you more.

What to do before your first session

1. Pick one method and keep it. Mixing them across weeks makes old scores useless.
2. Agree what Ease means. Growth Method flags that it is unclear whether Ease means time, cost, or complexity.
3. Write both formulas into your spreadsheet as extra columns. If the two orders ever disagree, talk about that idea before you commit.

What are the problems with ICE scoring?

ICE breaks in three predictable places. Each has a calibration fix you can add to the template before your next scoring session.

Subjective inputs

Two people can look at the same idea and land on different numbers. One practitioner guide calls the method still a bit subjective and gives the argument every team has had: “Is it a 6 or an 8?” CXL makes the same point about the whole family of frameworks: broad criteria such as potential, impact, and confidence can leave too much room for subjective scoring.

Fix: write sub-questions for each factor. That guide suggests asking “How many visitors will this affect?” when scoring Impact. In your template, add one line under each factor that describes what a low, middle and high score means for your target metric. Scorers then point to the line instead of debating the number.

Confidence scored on gut feel

CXL's critique of ICE is that 2 scores out of the 3 are a lot about “gut feeling.” It adds that it’s easy to skew, too, if you really want to pursue the idea. In our reading, the person who pitched the idea is the one most likely to inflate it.

Fix: score evidence, not belief. Change the question to “How much evidence do you have that makes you confident?” Then tie each score to a type of evidence. Savio uses Itamar Gilad's Confidence Meter for this:

Evidence behind the idea	Confidence score (Savio's scale)
You think it will work	0.01
Opinions of coworkers and managers	0.1
A few customers asking for it	0.5
Market research	1 to 3
Rigorous longitudinal user studies	3 to 7
Actual launch data	up to 10

Add an Evidence column next to Confidence and require a link to the interview notes, survey or dashboard. No link, no score above the bottom row.

Ease bias that buries big ideas

Because Ease carries the same weight as Impact, cheap tweaks float to the top. In the guide's worked example, the exit-intent popup wins out, despite having a lower potential impact than the referral program. The referral idea in our own example earlier in this piece fell behind the headline rewrite for the same reason.

Fix: protect big swings by quota. The guide advises to put more weight on ‘Impact’ and less on ‘Ease’ and to keep an absolute minimum of 20% of your experiments dedicated on big swings. In the backlog, tag high-impact, low-ease ideas as big swings and rank them in their own list. Fill the reserved slots from that list before the main ranking takes the rest. If a big swing is hard to test whole, our referral program playbook shows how to cut one into a smaller first test.

What is the difference between ICE, RICE, PIE and PXL?

All four frameworks turn a backlog into a ranked list. They differ in what they ask you to score, how fine the scale is, and how the factors combine. The table sets out what each framework's authors published, so you can match one to the backlog you actually have.

Framework	Factors	Scale	How factors combine	Developed by	Backlog it suits
ICE	Impact, Confidence, Ease	As covered above	Multiplied, as above	Sean Ellis at GrowthHackers	Fast-moving growth experiment backlogs
RICE	Reach, Impact, Confidence, Effort	Reach in people per period; Impact 3, 2, 1, 0.5 or 0.25; Confidence 100%, 80% or 50%; Effort in person-months	Reach × Impact × Confidence ÷ Effort	Sean McBride at Intercom	Product and roadmap prioritization at scale
PIE	Potential, Importance, Ease	1 to 10 per factor	Average of the three	Chris Goward at WiderFunnel	Conversion rate optimization (CRO) and landing page tests
PXL	10 specific questions, such as above the fold or high traffic page	1 or 0 for most questions, 2 or 0 for some	Points for each yes answer	Peep Laja at CXL	Underperforming pages for CRO

A RICE worked example

In our reading, RICE is the one framework here that weighs ideas against actual effort. The arithmetic is ours, the inputs are Intercom's:

• Project 2: 2,000 × 1 × 80% ÷ 4 = 400
• Project 3: 800 × 2 × 50% ÷ 1 = 800

The project with the lowest confidence ranks first, because it takes a quarter of Project 2's effort. In our reading, that is the pattern RICE surfaces and ICE tends to hide: a cheap, uncertain idea can beat a large, safe one.

Lesser-used variants

• BRASS: Blink, Relevance, Availability, Scalability, Score, from David Arnoux at Growth Tribe.
• HIPE: Hypothesis, Investment, Precedent, Experience, from Jeff Chang at Pinterest.
• DICET: Dollars (or revenue) generated, Impact, Confidence, Ease, Time-to-money, from Jeff Mignon at Pentalog.
• HiPPO: the highest paid person's opinion, listed as a framework mostly as a warning.

How to pick this week

Growth Method's view is that the specifics of each framework matter far less than picking one and implementing it within your team. We think the backlog decides it. If most ideas are page tests backed by research, PXL's yes/no questions cut the arguments. If ideas span channels and effort varies by months, RICE is the one that divides by effort. If you run a weekly cadence with small tests, keep ICE and add the ICE score evidence rules from the earlier section.

How RICE and PXL take some of the guesswork out of scoring

In our reading, RICE and PXL swap some of those judgments for things you can count or check, which makes a score harder to inflate in a backlog review.

RICE: reach and effort are counts

In RICE, reach measures "the number of people or users affected by a project within a specific time frame". You pick the event and the window, then look it up in your analytics. ProductPlan's example: if a project sends 1,200 prospects to a trial page in a month and 30% sign up, reach is 360. That is a number someone else can check.

Effort is also a count of person-months, and Intercom scores anything less than a month as a.5. Confidence then discounts the parts you guessed. ProductPlan adds that a score below 50% should be treated as a "moonshot".

Here is Product School's three-project example, worked through the formula from the earlier section.

Project	Reach (users/month)	Impact	Confidence	Effort (person-months)	RICE score
A	1,000	2	80%	3	533.33
B	500	3	70%	2	525
C	2,000	1	90%	4	450

Project A: 1,000 × 2 × 0.8 = 1,600, divided by 3 = 533.33. Project B carries the top impact rating and still lands second. In our reading, a gap that small means one revised input could flip the order, so write down how you counted reach and effort next to each score.

PXL: yes or no instead of a confidence guess

CXL built PXL because a confidence score is hard to justify in advance. As the CXL post puts it: "But if I could guess what the impact would be, why would I even test it?" PXL makes a yes or no, binary decision on each item. It gives more weight to ideas supported by user research, qualitative feedback, analytics, heatmaps, or eye tracking, and noticeability, reach, and implementation effort also count.

Questions in that style that you can paste into your backlog this week:

• Is the change visible without scrolling?
• Did user research or customer feedback point to this problem?
• Do analytics show a drop-off on this step?
• Does a heatmap or session recording show the issue?
• Can it ship without engineering help?

Every yes adds a point. A teammate cannot nudge a yes upward the way they can nudge a 7 to an 8.

CXL says PXL should be customized to reflect the constraints and priorities of your business, so swap in questions that match your funnel. If traffic limits how many tests you can run, check the numbers with the A/B test sample size calculator before an idea earns its slot.

The experiment backlog template you can copy

Copy the table below into a spreadsheet or a Notion database, one row per idea.

The hypothesis column uses two fixed openers from that framework: "We believe that" and "To verify that, we will". The first states the problem, the fix and the outcome you expect. The second states what you will change and how you will test it. In that framework, every experiment starts with a problem to solve found in qualitative or quantitative data.

Column	How to fill it	Example row (our illustration)
Idea	One line, plain words.	Shorter signup form
Hypothesis	We believe that + problem, fix, expected outcome. "To verify that, we will" + what changes and the test type.	We believe that the phone field causes drop-off, so removing it will lift completed signups. To verify that, we will A/B test the form with and without it.
Target metric	One metric only: a volume, a conversion rate or an amount.	Signup completion rate
Success criteria and damage control	Minimum uplift that counts as a win, plus one metric that must not get worse.	Win only above the uplift threshold set in advance. Guard: sales-qualified leads
ICE inputs with evidence notes	Each score next to the evidence behind it: a recording, a survey quote, a past test. A score with no note stays provisional.	Impact: funnel data. Confidence: session recordings. Ease: one front-end change
Owner	One named person who ships and reports.	Growth marketer
Status	Backlog, scoped, running, analyzing, done.	Running
Result	Metric before and after, and whether it cleared the threshold.	Filled at the end
Learning	What you learned, what surprised you, what you would change.	Filled at the end
Next step	Scale, iterate or kill.	Filled at the end

Notes on the columns that go wrong

Write the test description for an engineer. In our reading, the framework warns that a vague description can lead to misinterpretation, which can make results unusable.

Keep the target metric singular. The same source says having a single, clearly identified metric is the way to go, and defines damage control as the part of the experience you must not degrade. Set the uplift threshold before launch and check the volume you need with the sample size calculator.

For ICE, that framework rates each criterion from 1 to 5 and adds them. Whichever arithmetic you use (see our ICE score entry), keep it fixed across the sheet. Tie Impact to your north star metric.

Fill the learning column even when the test fails. One of the questions the framework asks is whether you discovered anything not expected. For a failed test, its next step is to kill it or try it differently. In our reading, the learning column is what stops the team from re-running the same idea next quarter.

How many experiments should a growth team run each week?

Tempo is the point of the whole system. The summary of Hacking Growth by Sean Ellis and Morgan Brown puts it plainly: "The faster you learn the faster you can grow." It adds that "Most experiments fail to produce results, so volume is key." The same summary advises teams to "Set a minimum number of tests a week."

That minimum is a floor you commit to, not a target to beat. A small team that promises more tests than it can build or analyze ends up with half-finished tests and no learnings.

The weekly heartbeat

The book's advice is to "Set a weekly heartbeat to encourage experiment velocity, discuss results and what to test next." We split that heartbeat into four slots.

Slot	What happens	Output	Our suggested day
Idea review	Anyone adds ideas to the backlog, each tied to a problem in the data	New backlog rows	Monday
Scoring session	Score new ideas with Impact, Confidence, Ease (ICE) and pick the next launches	A ranked shortlist	Tuesday
Launch slot	Ship the chosen tests, with metric and damage control set	Live experiments	Wednesday
Learnings review	Read finished tests, write learnings, decide next steps	Updated backlog	Friday

Picking a launch count you can sustain

Your weekly count is the lower of two limits: what the team can build and what your traffic can support.

1. Build limit. The template's estimated time field covers "the time to build the experiment;" the run, and the analysis. Add up build and analysis hours for your shortlist and stop when you hit the hours you have.
2. Traffic limit. Run each test idea through the sample size calculator. Two tests on the same page split its traffic, so count tests per surface, not per team.
3. Take the smaller number and commit to it for a month.

A worked example. A two-person team has about 10 hours a week for experiments, and each test takes roughly 5 hours to build and analyze, so the build limit is two. Their pricing page needs a full two weeks of traffic for one test, so it carries one test at a time. Their onboarding emails and signup page each support one more. The traffic limit is three, the build limit is two, and they commit to two launches a week.

Volume should not mean scattershot. The summary warns that "Scattershot experimentation is a waste of time and effort - instead focus on growth levers." Keep each week's launches pointed at one stage, such as activation.

When a test ends flat, the learnings review has three options: "Kill it, continue running it to gather more learnings or change the test methodology." If flat results keep coming from one surface, we think that surface is short on traffic, and you should lower its count.

Worked example and the steps to set this up this week

Here is one week of the template in use, with five ideas pulled from the backlog.

Idea	Impact	Confidence	Ease	ICE total	Week 1 status (our example)
Rewrite pricing page headline	4	4	5	13	Launched Tuesday
Onboarding checklist email	4	3	4	11	Launched Tuesday
Cancellation exit survey	2	4	5	11	Launched Wednesday
Referral prompt after first project	5	2	3	10	Spec written, build next week
Annual plan discount banner	3	2	4	9	Parked

Re-scoring the page test with PIE

The headline rewrite is a page-level test, so we ran it through PIE (Potential, Importance, Ease) as a second check. Growth Method lists PIE as a fit for conversion rate optimization (CRO) teams prioritizing page-level A/B tests. Chris Goward created it for conversion optimization. You score each criterion from 1 to 10 and multiply them.

The headline stays first. In our reading, PIE earns its place for page tests because Importance asks how valuable the page's traffic is, which ICE never asks directly.

Calibrate before you score. Statsig gives one anchor: a 10 for Potential could mean doubling a key metric, a 5 a 20% improvement.

One result, one learning

Result. By Friday the headline variant showed a 4.1% signup rate against 3.8% for control, but traffic was short of the sample the A/B test sample size calculator gave us. Decision: keep running, review next Friday.

Learning. The exit survey surfaced setup time as the most common reason for cancelling. That became a new backlog idea on time to value, scored at the next heartbeat. Statsig's advice applies here: document why you scored things the way you did.

Set this up this week

1. Monday: copy the backlog template and enter every idea you have, each with a hypothesis and one metric.
2. Monday: write down what a 1 and a 5 mean for each ICE factor. Growth Method's tip is to score independently first, then discuss gaps.
3. Tuesday: score the backlog and pick the top three that fit your weekly test minimum.
4. Tuesday: re-score any page-level idea with PIE before it ships.
5. Wednesday: launch, and log the start date and required sample size.
6. Friday: add new ideas from the learnings and re-score anything that changed.

What this page does not cover

This page does not cover how to analyze test results statistically beyond checking the required sample size, or how to choose software to host the backlog. It also leaves out the lesser-used frameworks such as BRASS, HIPE and DICET beyond naming them. It is not financial or legal advice.

Related on Race to Scale: A/B test sample size and duration calculator, with the formula; What is a good activation rate? Benchmarks by product type and how to define yours; Free-to-paid conversion rate benchmarks for freemium, opt-in trials and card-required trials.