Every team has more ideas than testing capacity. Without a shared method, the loudest or most senior opinion wins. PXL, a framework created by CXL, replaces opinion with specific questions.
The ten PXL criteria
| Group | Question | Score |
|---|---|---|
| Impact | Is the change above the fold? | 1 / 0 |
| Impact | Is it noticeable within 5 seconds? | 2 / 0 |
| Impact | Does it add or remove an element? | 2 / 0 |
| Impact | Is it designed to increase user motivation? | 1 / 0 |
| Impact | Does it run on high-traffic page(s)? | 1 / 0 |
| Evidence | Does it address an issue found in user testing? | 1 / 0 |
| Evidence | …found in qualitative feedback (surveys, polls, interviews)? | 1 / 0 |
| Evidence | …found in digital analytics? | 1 / 0 |
| Evidence | Is it supported by heatmaps or eye tracking? | 1 / 0 |
| Ease | Under 4 hours = 3, up to 8 hours = 2, under 2 days = 1, more = 0 | 3–0 |
The maximum is 14. Binary answers stop people hedging with a “6 out of 10”, and the evidence questions mean an idea with no research behind it cannot score highly.
A worked example
Hypothesis: Because session recordings and an exit survey show shoppers leaving checkout when shipping appears, showing the delivery cost on the product page will reduce checkout abandonment.
Above the fold (1) · noticeable (2) · adds information (2) · motivation (1) · high traffic (1) · user testing (0) · qualitative (1) · analytics (1) · heatmaps (0) · ease, about 6 hours (2) = 11 / 14.
Score your own backlog with the PXL calculator: it ranks ideas and exports a CSV.
PXL versus ICE, PIE and RICE
- ICE (impact, confidence, ease) and PIE (potential, importance, ease) are quick but rely on subjective 1–10 ratings that vary between people.
- RICE adds reach and is popular in product teams; it still needs an impact estimate.
- PXL trades nuance for consistency: concrete questions, mostly binary, weighted towards evidence.
Customize it, as CXL intends
CXL explicitly recommends adapting PXL. Common additions:
- “Complies with brand and accessibility guidelines” (1/0)
- “No SEO risk” for content-heavy sites (1/0)
- “Affects the primary outcome directly” (1/0)
Keep the changes written down so scores stay comparable over time.
Limits of any scoring model
- Scores rank candidates; they don’t forecast results.
- Criteria are heuristics, not empirically calibrated weights.
- Ties are common: break them with strategic fit and dependencies.
- Re-score when new evidence arrives in the next Wonder phase.