The PIE framework is a way to rank A/B testing opportunities by scoring each one from 1 to 10 on three criteria: Potential (how much the page can improve), Importance (how valuable its traffic is) and Ease (how hard the test is to build and get approved).
You average the three scores, and the highest PIE score gets tested first.
Chris Goward created PIE at the agency WiderFunnel, which is now Conversion, to answer one question: where should we test first?
Below is how the scoring works, a worked example, and what real tests in our database say about each of the three factors.
What PIE prioritization is
PIE was built to rank places to test, such as pages, templates and funnel steps, not individual ideas.
In Goward's 2013 Practical Ecommerce article, the example table lists the home page, category template, product detail template, shopping cart and blog template, not button colors or headlines.
That matters because a lot of advice now uses PIE to rank single test ideas.
That works, but the criteria were written with pages in mind.
Importance especially is a property of a page's traffic, not of an idea.
Potential: how much can this page improve?
Goward's definition is "How much improvement can be made on the pages?"
Conversion's current framework page says to look at your worst performers using analytics, customer data and heuristic analysis.
A page with an obvious problem scores high.
One store tested by Dave Diederen had a conversion rate below 1%, and enabling Shop Pay with buy now, pay later moved it above 1%.
A page that is already well optimized has less room to gain, so it should score lower.
Importance: how valuable is the traffic?
Goward's rule is that "your most important pages are the ones with the highest volume and the costliest traffic."
Given two pages with similar problems, the one fed by paid traffic gives a better return on the testing effort.
Importance also decides whether a test can finish at all.
A page with little traffic needs a very large effect to reach significance in a reasonable time. Our low-traffic testing guide explains why.
The cart is a good example of a high-importance step, because nearly every buyer passes through it.
Oliver Kenyon added tiered spend rewards to a cart (free shipping plus bonus items) and reported +$57k/month.
Dave Diederen showed customer reviews on the checkout page and got +23.13% revenue per visitor.
Ease: how hard is it to test here?
Ease covers both technical work and internal approval.
Goward warns: "Don't underestimate the 'political' ease of a test opportunity," because a page that is technically simple "may have many stakeholders or vested interests that can cause barriers."
A high Ease score does not mean a small payoff.
Oliver Kenyon removed a single line of copy from a cart page and reported +$40,413 per month, about +$1.50 revenue per session. Building that test takes minutes.
How to use the PIE prioritization framework
The method has four steps. It fits in a spreadsheet, and our prioritization spreadsheet template has the columns set up.
- List the opportunities. Group pages that share a template into one row. Goward's team rated template pages, such as all product pages, as a single page rather than testing thousands of product pages one by one.
- Score each factor from 1 to 10. Use data where you have it: conversion rate by step for Potential, sessions and traffic cost for Importance, a developer's estimate for Ease.
- Average the three scores. PIE score = (Potential + Importance + Ease) / 3.
- Sort by PIE score and test from the top. Write a hypothesis for each top row and add it to your testing roadmap.
How to score more consistently
PIE scores are only as good as the evidence behind them.
A few habits make them less of a gut call.
- Anchor Potential to data. Use drop-off rates by funnel step, session recordings and customer survey answers. A step where half of visitors leave has more room than one where 5% leave.
- Anchor Importance to money. Rank areas by revenue or paid sessions passing through them each month, then convert that ranking to 1-10.
- Ask the person who will build it. Ease scored by a marketer alone tends to be optimistic. Include approvals, legal review and app limits, not just coding time.
- Score as a group, then compare. Have two or three people score on their own and discuss any factor where they differ by more than 3 points.
- Rescore every quarter. Traffic mix, site changes and past results all move the numbers.
The original scoring table
Goward's published example uses the 1-10 scale and a plain average. A few of its rows:
| Test area | Potential | Importance | Ease | PIE |
|---|---|---|---|---|
| Home A/B cluster | 10 | 10 | 8 | 9.3 |
| Category template A/B cluster | 10 | 9 | 7 | 8.7 |
| Product detail template A/B cluster | 10 | 10 | 4 | 8.0 |
| Shopping cart A/B cluster | 9 | 9 | 3 | 7.0 |
| Blog template | 8 | 2 | 7 | 5.7 |
The blog template has plenty of room to improve (Potential 8) but drops to the bottom because so few buyers reach it (Importance 2).
The cart scores 9 on both Potential and Importance but loses ground because the test is hard to build (Ease 3).
Worked example: a small Shopify store
Here is a made-up store with four candidate areas, scored the PIE way.
The numbers are illustrative, not from a real client.
| Area | Potential | Importance | Ease | Calculation | PIE |
|---|---|---|---|---|---|
| Cart drawer | 7 | 9 | 8 | (7 + 9 + 8) / 3 | 8.0 |
| Product page template | 8 | 10 | 5 | (8 + 10 + 5) / 3 | 7.7 |
| Checkout | 6 | 9 | 4 | (6 + 9 + 4) / 3 | 6.3 |
| About page | 9 | 2 | 9 | (9 + 2 + 9) / 3 | 6.7 |
The cart drawer goes first. Notice the About page beats checkout, because its very high Potential and Ease make up for almost no traffic.
That is a known weakness of plain averaging, covered below.
Many teams fix it by dropping any row with Importance under 4 before sorting.
What our tests say about scoring
The scores are guesses, and the database shows how often guesses miss.
Clean Commit publishes its full test library, and of its 173 tests in our database, 16 won, 35 lost and 122 were flat.
Microsoft's experimentation team found that only about 1/3 of ideas improve the metrics they were designed to improve.
Big changes are where Potential gets overrated most.
Clean Commit's product page redesign with new copy lost 28.3% conversion rate, and a hockey landing page rebuilt on another sport's layout lost 38.5% of conversions.
Dave Diederen ran a direct comparison. A full checkout update dropped conversion 1.18%, while adding urgency messaging alone lifted revenue per visitor 6.59%.
The easier, narrower test won and was easier to learn from.
@oliverkenyon@oliverkenyon
@DaveDiederen@DaveDiederen
Clean CommitClean CommitPIE vs ICE vs PXL
PIE has two close relatives. All three rank test opportunities, but they ask different questions.
| PIE | ICE | PXL | |
|---|---|---|---|
| Criteria | Potential, Importance, Ease | Impact, Confidence, Ease | Specific yes/no questions on evidence, noticeability, traffic and effort |
| Scale | 1-10 each, averaged | A subjective score for each factor | Mostly 0 or 1, with extra weight on some items |
| Origin | Chris Goward, WiderFunnel | Popularized by Sean Ellis, GrowthHackers | CXL, 2016 |
| Best for | Choosing which pages or funnel steps to work on | Ranking growth ideas quickly | Ranking many individual test ideas with less guessing |
CXL created PXL because broad criteria leave too much room for opinion.
Its PXL write-up points out that "if we'd know in advance how much potential an idea has, we wouldn't need prioritization models," and that someone pushing an idea could just add a couple of points to Potential.
See our ICE framework and PXL framework pages for details.
When PIE is the wrong choice
- You are ranking dozens of small ideas on the same page. Importance will be the same for all of them, so PIE becomes a two-factor guess. PXL's specific questions do better here.
- Your team argues over scores. A 1-10 scale with no definitions invites inflation. Write down what a 3, 6 and 9 mean for each factor before scoring, or switch to binary criteria.
- Traffic is the real constraint. Averaging lets a page with no traffic rank highly. If a test can't reach its minimum detectable effect within a few weeks, its PIE score doesn't matter.
- You have past results. Once you have run tests, results on similar pages are better evidence for Potential than opinion. Keep them in one searchable place and check them when scoring.
For a broader view of the options, start with our guide to A/B test prioritization.
To check your Potential guesses against real data, browse all tests in the database by page type.