Step One Life · AU + UK · Jan – Jun 2026
Six iterations across two markets to fix three structural problems on the product detail page. The UK rollout returned a definitive +17.6% conversion rate on flat order value — and the programme's other result was a conversion “win” I refused to ship.
UK conversion rate lift, >99.99% confidence
UK revenue uplift in a six-day window
iterations across 2 markets, ~85k visitors
revenue lost by the “winning” iteration I refused to ship
Contents
Step One's PDP had three structural problems that no amount of merchandising could fix:
The hypothesis I carried through the programme:
If we redesign the PDP with a cleaner look and feel along with much quicker product switching, then we will increase conversion rate, because the new design is quicker and more user friendly enabling users to browse and ATC more efficiently.
The design concept was "Islands" — modular content blocks that persist in the viewport, with a stepped selection process (fit → size → pack) and a pop-out colour selector showing larger product images. The critical build decision was that USPs and the pack selector stay in view while the customer browses, so the value proposition never scrolls away.
Platform: Shoplift, storefront theme tests. Statistics: two-proportion significance test on conversion rate and add-to-cart rate; zero-inflated test on revenue per visitor. 95% confidence threshold. Segmentation every read: device (mobile carries the large majority of traffic), new vs returning, channel. Diagnostic layer: SKU-prefix unit mix and pack-tier analysis, to explain why a metric moved.
I ran this as an iterative programme, not a single test. Each round kept what worked and rebuilt what didn't — and one round was killed outright rather than reported.
AU_PDP_CVR_260107 | Islands, 4 Products | 7–15 Jan 2026| Metric | Change | Significance |
|---|---|---|
| Visitors | ~46k total | balanced split |
| Conversion Rate | +2.1% | not significant |
| Revenue Per Visitor | −3.9% | — |
| Average Order Value | −6.8% | — |
| Revenue | −4.8% | — |
| Add to Cart Rate | +3.3% | — |
This is the classic trap, and I did not ship it. The variant produced more orders at a marginally better conversion rate — and 4.8% less revenue, because average order value fell 6.8%. More, smaller baskets.
The read: the Islands layout improved browsing and cart entry but degraded the pack-selection step. Customers were reaching cart with shallower packs. A conversion win that destroys order value is a loss, and I reported it as one.
AU_Theme_CVR_260116 | Islands, Most PDPs | 16–19 Jan 2026Killed. The experiment log records: "Bugged Quick Add - needs to be retested." Status inconclusive, result declared invalid rather than reported. A quick-add bug in the roll-out to broader PDPs made the data unusable, so I stopped it and fixed the build.
AU_Theme_CVR_260121 | Round 4 | 21–23 Jan 2026Rebuilt with the quick-add fixed and the layout tightened.
| Metric | Change | Significance |
|---|---|---|
| Visitors | ~7.8k total | balanced split |
| Conversion Rate | +5.1% | → 93.2% confidence |
| Revenue Per Visitor | +2.7% | — |
| Average Order Value | −1.7% | — |
| Add to Cart Rate | +4.7% | — |
RPV went positive. The CVR lift survived and the value leak was largely closed — AOV was down only 1.7% versus 6.8% in iteration 1, and RPV moved from −3.9% to +2.7%.
The standout was desktop:
| Device | Metric | Change |
|---|---|---|
| Desktop | CVR | +34.7% |
| Desktop | RPV | +28.8% |
| Mobile | CVR | flat |
| Mobile | RPV | −2.0% |
Desktop converted dramatically better; mobile was flat. That split told me the layout's benefits — persistent USPs, bigger product imagery, in-viewport pack selection — were being realised where there was screen real estate, and lost on small viewports. Mobile needed its own round.
UK_PDP_CVR_260204 | UK Version | 4–9 Feb 2026 ★The UK test is the strongest result in the entire CRO programme.
| Metric | Change | Significance |
|---|---|---|
| Visitors | ~32k total | clean 50/50 split |
| Conversion Rate | +17.6% | >99.99% confidence |
| Revenue | +18.8% | — |
| Revenue Per Visitor | +19.8% | — |
| Average Order Value | +0.7% | — |
| Add to Cart Rate | +0.7% | — |
| Orders | +18.0% | — |
The important detail is what moved and what didn't. CVR jumped 17.6% and revenue 18.8%, but ATC rate was flat (+0.7%). The redesign did not push more people into the cart — it made the people already in the funnel convert. AOV was also flat, so the entire gain was clean: more orders at the same value, not the same orders at inflated value.
Device segmentation held the pattern:
| Device | Metric | Change |
|---|---|---|
| Mobile | CVR | +18.8% |
| Mobile | RPV | +20.4% |
| Desktop | CVR | +12.2% |
| Desktop | RPV | +18.7% |
Unlike AU Round 4, mobile now carried the gain — the volume driver, moved with desktop rather than lagging it. That was the fix the AU read had demanded.
AU_Theme_CVR_260526 | PDP v3 | from 26 May 2026With the UK confirmed, AU went back for a v3 targeting the elements the earlier rounds had exposed. The build included:
The hypothesis states the goal directly: "stepped process — allows users to select their preferred fit easily (people weren't using the top nav), reversed pack buttons to focus on higher value packs." This is the round that attacks the AOV leak at its source.
AU_Theme_CVR_260616 | PDP v3, Paid Search Target | from 16 Jun 2026A segment-targeted follow-up, and the detail I'd point to as the mark of a mature programme. The experiment log reads:
"Continues from AU_Theme_CVR_260526 - specifically targets paid as this channel suffered during the last test."
Rather than accepting a global result or abandoning a channel, I isolated paid search — the channel that regressed — into its own conditional test. That is the difference between running tests and running a programme.
1. A conversion win that lowers AOV is a loss. Iteration 1 delivered more orders and a better CVR and 4.8% less revenue. If I had reported CVR as the headline, it would have read as a win. The Islands layout was genuinely improving the browsing experience while quietly degrading the pack-selection step — and only the revenue view exposed it. This is why I treat RPV, not CVR, as the commercial guardrail even on a CVR-primary test.
2. The gap between ATC and CVR tells you which half of the funnel you fixed. Iteration 4 lifted CVR 17.6% with flat ATC rate. The redesign was not adding people to the cart — it was converting the people already there. That distinction changes the follow-up: you don't need more traffic, you need the same traffic better served.
3. Desktop and mobile fail for different reasons, and you have to read them separately. AU Round 4 was desktop-only in effect (RPV +28.8% desktop vs −2.0% mobile). The UK version fixed it — mobile moved +18.8% CVR. Had I read only the aggregate in January, I would have missed that the layout's persistent-USP advantage depends on viewport, and shipped something that helped a small minority of traffic and hurt the rest.
4. Killing a test is a result. Iteration 2 was stopped on a quick-add bug and its data declared invalid rather than published with a caveat. A bugged test reported as a directional result contaminates every decision built on it.
5. Replication is a separate test, not a formality. The same design shipped +2.7% RPV in AU and +19.8% RPV in the UK. UK traffic skews different, the price points differ (£15–17 vs A$29–33), and the mobile/desktop mix differs. The UK result is not confirmation of the AU result — it is a stronger, separate finding that happened to agree.
| Iteration | Market | Dates | Visitors | Best metric | Outcome |
|---|---|---|---|---|---|
| Islands 4-Product | AU | 7–15 Jan | ~46k | CVR +2.1% | RPV −3.9%, AOV −6.8% — not shipped |
| Islands Most PDPs | AU | 16–19 Jan | — | — | Killed — quick-add bug |
| Round 4 | AU | 21–23 Jan | ~7.8k | CVR +5.1%, RPV +2.7% | → Directional; desktop RPV +28.8% |
| UK Version | UK | 4–9 Feb | ~32k | CVR +17.6%, >99.99% confidence | Shipped — +18.8% revenue |
| PDP v3 | AU | from 26 May | running | — | Reversed pack hierarchy, stepped process |
| PDP v3 Paid Search | AU | from 16 Jun | running | — | Segment-targeted follow-up |
In plain terms: across six iterations and two markets I rebuilt the product detail page from a static template into a stepped, modular experience with persistent value messaging and an in-viewport pack selector. The UK rollout delivered a statistically definitive +17.6% conversion rate and +18.8% revenue uplift (>99.99% confidence) with flat AOV — a clean, non-inflated gain in a six-day window on ~32k visitors.
The programme's more important output was the discipline it demonstrated: I caught and refused a conversion "win" that was costing 4.8% of revenue, killed a bugged test rather than reporting it, diagnosed the device split that explained why the first layout failed on mobile, and then built a follow-up that targets the single channel that regressed.
AU_Theme_CVR_260616 is the right follow-up; the priority is getting it to significance so the channel decision is settled rather than parked.Sources: the Step One experiment log and Shoplift test exports. Revenue, traffic and baseline figures have been generalised; relative changes and test statistics are as measured. Programme IDs: AU_PDP_CVR_260107 · AU_Theme_CVR_260116 · AU_Theme_CVR_260121 · UK_PDP_CVR_260204 · AU_Theme_CVR_260526 · AU_Theme_CVR_260616.