The PXL framework is CXL's method for ranking A/B test ideas with mostly yes-or-no questions instead of 1-10 guesses.
Each idea gets points for being visible (above the fold, noticeable within 5 seconds, adding or removing an element), for running on high-traffic pages, for being backed by research (user testing, surveys, analytics, heatmaps), and for being quick to build, for a maximum of 14 points. The highest-scoring ideas get tested first.
It was built to fix one weakness in scoring models like ICE and PIE: when you rate "impact" or "potential" on a sliding scale, you are guessing the result you are about to test.
Below is how the scoring works, a worked example, and what real tests in our database say about each criterion.
How PXL prioritization works
CXL published PXL in September 2016 along with a free Google Sheets template.
Most questions score 0 or 1. Three are weighted because CXL considered them more important, and ease of implementation is scored in time brackets.
The ten PXL criteria and their points
| Criterion | Points | What it stands in for |
|---|---|---|
| Is the change above the fold? | 1 or 0 | How many visitors will see it |
| Is it noticeable within 5 seconds? | 2 or 0 | Whether the change is big enough to move behavior |
| Does it add or remove an element? | 2 or 0 | Size of the change, versus tweaking what's there |
| Is it designed to increase user motivation? | 1 or 0 | Whether it targets a reason to buy, not just layout |
| Is it running on high-traffic page(s)? | 1 or 0 | Absolute value of a relative lift, and time to a result |
| Does it address an issue found in user testing? | 1 or 0 | Evidence |
| Does it address an issue found in qualitative feedback (surveys, polls, interviews)? | 1 or 0 | Evidence |
| Does it address insights from digital analytics? | 1 or 0 | Evidence |
| Is it supported by mouse-tracking heatmaps or eye tracking? | 1 or 0 | Evidence |
| Ease of implementation | 3 (under 4 hours), 2 (up to 8 hours), 1 (under 2 days), 0 (more) | Cost and speed |
Add the column and you get a score from 0 to 14. Four of the 14 points come from research, so an idea that is only someone's opinion starts at a disadvantage.
That is deliberate: CXL's post says the model "asks everyone to bring data."
The "noticeable within 5 seconds" check has a concrete method behind it.
CXL suggests showing people the control and then the variation for 5 seconds and asking whether they can tell the difference.
If they can't, the change probably won't move many visitors.
Our guide to five-second and preference tests covers how to run that check.
A worked example
CXL's template ships with three sample ideas. Here is how they score.
| Idea | Fold | 5 sec | Add/remove | Motivation | Traffic | User test | Qual | Analytics | Heatmap | Ease | Total |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Restructure and rewrite the Tour page copy | 1 | 2 | 2 | 1 | 1 | 1 | 1 | 1 | 0 | 2 | 12 |
| Reverse the order of the home page content blocks | 1 | 2 | 0 | 0 | 1 | 0 | 0 | 1 | 0 | 3 | 8 |
| Increase body font size on mobile | 0 | 2 | 0 | 0 | 1 | 0 | 0 | 0 | 0 | 2 | 5 |
The copy rewrite wins because three separate research sources point to it.
The font-size test scores lowest, mostly because nothing in the research flagged it.
That last row is worth a second look.
In our database, Dave Diederen increased PDP font size by 4px for an older audience and got +4.17% revenue per visitor.
A 5-point idea can still win. PXL ranks the order you test in; it doesn't predict which tests win, and CXL's own FAQ says the same.
Running a PXL scoring session
The model works best as a short, regular meeting rather than a spreadsheet one person fills in alone. A typical routine looks like this:
- Collect ideas with their evidence attached. Each idea in the backlog should link to the survey answers, session recordings, analytics report, or heatmap that prompted it. No link, no research points.
- Bring a developer. CXL recommends having a test developer in the discussion, because the ease brackets depend on a time estimate. Developers tend to underestimate, but a bracketed estimate is still better than a gut "easy."
- Score as a group, out loud. Yes-or-no questions make disagreements quick to settle: either the change is above the fold or it isn't.
- Sort and fill the roadmap. Take the top ideas that don't collide on the same page and slot them into your A/B testing calendar.
Rescore the backlog when new research comes in.
A 6-point idea can jump to 9 after a round of user interviews confirms the problem it addresses.
What our tests say about the criteria
PXL's criteria are reasonable proxies, but they are proxies.
Here is how a few of them hold up against results in the LiftArchive test database.
Above the fold. Position matters, and the effect runs both ways.
Dave Diederen put best sellers above the fold on a homepage and pushed reviews below: +39.03% revenue per visitor.
Going the other way, Patrick McKenzie moved a screenshot below the fold on the Bingo Card Creator home page and saw about 25% fewer downloads.
Research-backed ideas. The evidence questions are where PXL earns its keep.
Dave Diederen changed above-the-fold USPs to the points heatmaps showed buyers engaging with (+7.28% revenue per visitor).
Oliver Kenyon added three survey-based reassurance lines above a collection page grid, worth $44,890/month.
Adding or removing an element. This gets 2 points because bigger changes have bigger effects, but bigger isn't the same as better.
Dave Diederen removed an email popup, a highly noticeable removal, and revenue per visitor fell 20.51%.
Treat a big change as one more likely to produce a clear result, in either direction.
@oliverkenyon
@DaveDiederen@DaveDiederen@DaveDiederen@DaveDiederenCXL's PXL: where it came from and how to adapt it
CXL built PXL after using PIE and ICE with clients.
Their criticism was that "potential," "impact," and "confidence" are open to interpretation, and easy to inflate when someone wants their idea tested.
PXL swaps those for questions that two people should answer the same way.
The binary scoring came from two places, according to CXL: a binary version of ICE that scores items 1 or 2, and a prioritization model Hotwire presented at CXL Live 2016. A yes-or-no scale avoids central tendency, where everyone rates everything a 5 or 6 out of 10.
PXL vs ICE vs PIE
| PXL | ICE | PIE | |
|---|---|---|---|
| Inputs | 10 specific questions | Impact, Confidence, Ease | Potential, Importance, Ease |
| Scale | Mostly 0/1, some weighted (0-14 total) | Usually 1-10 each | Usually 1-10 each |
| Rewards research | Yes, 4 points | Only indirectly, through "confidence" | No explicit criterion |
| Time to score an idea | A minute or two | Seconds | Seconds |
| Main weakness | Fixed weights; many ties | Subjective and easy to inflate | Subjective and easy to inflate |
For the full set of options, see our guide to A/B test prioritization.
Customize it, as CXL intended
CXL says outright that the model should be changed to fit your business.
Their examples: add "conforms to brand guidelines" if a brand team signs off on tests, or "doesn't interfere with SEO" if organic search drives your acquisition.
Other changes teams often make:
- Split traffic into tiers. A single "high traffic" point treats a page with 10,000 visits and one with 500,000 the same. Scoring 0/1/2 by traffic band helps, especially on low-traffic sites where only a few pages can reach a result.
- Add a "closeness to revenue" point. Checkout and product pages usually matter more than a blog template with the same traffic.
- Add a "builds on a past win" point. A tested theme with a follow-up idea is better evidence than a new guess, which fits an iterative testing program.
- Break ties on purpose. With a 0-14 range and a long backlog, many ideas will tie. Pick one rule, such as ease first or revenue proximity first, and stick to it.
When PXL is the wrong choice
- You have no research yet. If nobody has run user tests, surveys, or heatmaps, every idea scores 0 on four criteria and PXL collapses into a visibility-plus-ease score. Do the research first, or use a simpler model until you have it.
- Your backlog is short. With five ideas, a team discussion is faster than a spreadsheet.
- Your tests are not page changes. Pricing, onboarding emails, algorithm changes, and server-side features don't fit "above the fold" or "noticeable within 5 seconds." Rewrite the visibility questions for the surface you test.
- You treat the score as a forecast. PXL is a ranking tool. A 5-point idea can win, as the font-size test above shows, and a big, noticeable change like the popup removal can lose.
Used well, PXL does one useful thing: it makes people show their evidence before an idea takes up traffic.
Pair it with a clear hypothesis for each idea, and use the database to check how similar changes have performed for other teams before you score them.