Which Experiment Areas Look Highest-Impact in Detected A/B Tests?
Across 2,160 area annotations Lazyweb Research scored in July 2026, offer carries the highest average model-assigned impact among high-volume areas at 3.69/5, followed by pricing at 3.53 and hero at 3.28. When you rank by the share of tests scored 4+/5, form (86.5%) and offer (66.4%) lead — but form rests on just 37 web annotations. The most-tested area, hero, sits mid-pack on impact. These are relative model scores on observed diffs, not measured lift. [1]
Offer has the highest average model-scored impact of any high-volume area at 3.69/5, with 66.4% of its tests scored 4+/5 — Lazyweb Research, 301 annotations, July 2026.
Impact by area
Ranking the areas with n≥30 annotations by the share scored 4+/5 (model-assigned 1–5, relative ranking only): [2]
| Area | n | Avg impact | Share 4+/5 |
|---|---|---|---|
| Form (web) | 37 | 3.84 | 86.5% [3] |
| Offer | 301 | 3.69 | 66.4% [4] |
| Pricing | 301 | 3.53 | 53.8% [5] |
| Hero | 488 | 3.28 | 28.1% [6] |
| CTA | 232 | 3.16 | 27.2% [7] |
| Value props | 276 | 3.03 | 15.6% [8] |
| Nav | 144 | 2.74 | 8.3% [9] |
The pattern: money-adjacent surfaces (offer, pricing) score highest; layout/navigation surfaces (nav, announcement bar) score lowest. The most-tested area — hero — is only mid-pack, meaning teams test it heavily but the model rarely flags those changes as high-leverage. [6]
How to apply it
If you're choosing where to spend a scarce experiment slot, the offer and pricing rows are the field's strongest signal that changes to the deal and the price move the needle most often. Form scores highest of all, but its 37 annotations are all web sign-up flows — treat it as 'when you have a form, the form matters,' not a mandate to add one. Nav and announcement-bar changes are common but rarely high-leverage, so reserve them for polish rather than headline bets.
Caveats
Impact is an LLM-assigned 1–5 score on a detected UI diff, usable only for relative ranking with n≥30 — never read it as a measured conversion lift. [10] Form (n=37) and the n≥30 threshold mean small-area rankings are noisier than the offer/pricing/hero rows. HEADER (n=29) falls below threshold and is excluded.
The numbers
| Stat | Computed from |
|---|---|
| 2,160 area annotations scored 1–5 for impact, July 2026 | statpack primaryN + high_impact_share_by_area |
| Impact-4+ share ranking, n≥30 areas only | statpack high_impact_share_by_area |
| Form: n=37, avg impact 3.84, 86.5% scored 4+ (32/37) | statpack area_FORM + high_impact_share_by_area |
| Offer: n=301, avg impact 3.69, 66.4% scored 4+ (200/301) | statpack area_OFFER + high_impact_share_by_area |
| Pricing: n=301, avg impact 3.53, 53.8% scored 4+ (162/301) | statpack area_PRICING + high_impact_share_by_area |
| Hero: n=488, avg impact 3.28, 28.1% scored 4+ (137/488) | statpack area_HERO + high_impact_share_by_area |
| CTA: n=232, avg impact 3.16, 27.2% scored 4+ (63/232) | statpack area_CTA + high_impact_share_by_area |
| Value props: n=276, avg impact 3.03, 15.6% scored 4+ (43/276) | statpack area_VALUE_PROPS + high_impact_share_by_area |
| Nav: n=144, avg impact 2.74, 8.3% scored 4+ (12/144) | statpack area_NAV + high_impact_share_by_area |
| Impact/confidence are LLM-assigned 1–5 scores for relative ranking only (n≥30), not measured lift; HEADER n=29 excluded | statpack smallSampleWarnings |
Sources & citations
- [1] Lazyweb Research analysis of 2,160 area annotations (1,126 experiments), July 2026. Impact is a model-assigned 1–5 score on detected UI diffs; rankings use the n≥30 filter. ↩
- [2] Lazyweb Research analysis of 2,160 area annotations, July 2026. Share of annotations scored impact ≥4/5 per area, n≥30 only. ↩
- [3] Lazyweb Research analysis of 37 form annotations (web), July 2026. Form: 32 of 37 scored 4+; smallest of the ranked areas, all web. ↩
- [4] Lazyweb Research analysis of 301 offer annotations, July 2026. Offer: 200 of 301 scored 4+. ↩
- [5] Lazyweb Research analysis of 301 pricing annotations, July 2026. Pricing: 162 of 301 scored 4+. ↩
- [6] Lazyweb Research analysis of 488 hero annotations, July 2026. Hero: 137 of 488 scored 4+. ↩
- [7] Lazyweb Research analysis of 232 CTA annotations, July 2026. CTA: 63 of 232 scored 4+. ↩
- [8] Lazyweb Research analysis of 276 value-props annotations, July 2026. Value props: 43 of 276 scored 4+. ↩
- [9] Lazyweb Research analysis of 144 nav annotations, July 2026. Nav: 12 of 144 scored 4+. ↩
- [10] Lazyweb Research methodology note, July 2026. Impact/confidence/differentiation are LLM-assigned 1–5 scores usable only for relative rankings with n≥30. ↩
Source: Lazyweb Research — proprietary analysis of real, in-market app screens. Cite as Lazyweb Research, 2026-07-07.