This is the testing plan built from June's storefront audit. It covers the four experiments worth running, the fixes that come first, and, importantly, the honest maths of what a store at Pott's traffic level can and cannot learn from a split test.
Pott is not broken. Around 45% of visitors who reach the cart go on to purchase, which beats the typical DTC store. Nothing in this plan is triage. This is a programme for compounding gains on a store that already works.
An A/B test splits visitors between two versions of a page and waits until the difference is provably real rather than luck. The rarer the action being measured, the more visitors it takes. That single fact shapes this whole plan.
Homepage and collection tests read on clicks into product pages. Plenty of signal, fast answers. These are the workhorses of the plan.
Tests that read on actual purchases only work on the busiest pages, and only when chasing a big lift. Small lifts are mathematically out of reach.
How long the flagship test would take to give a trustworthy verdict, depending on how big an improvement we ask it to detect:
This is why the plan is honest about three things: we read clicks wherever we can, we run purchase-based tests only on the highest-traffic pages while chasing big lifts, and some improvements simply ship without a test because pretending to test them would produce noise, not knowledge.
The single most important rule in this framework. Pott's subscription programme is where long-term value lives, and one of our tests deliberately touches it.
A subscriber is worth several times a one-time buyer. Any test that touches the subscription offer is judged on total value, not just the first order.
A page variant can lift one-time sales while quietly cutting subscription sign-ups, and look like a winner on raw conversion while losing money overall. So the verdict metric for subscription-touching tests is revenue per visitor including subscription value: each new subscriber is counted at expected lifetime value at the end of the test, and the result is re-confirmed after a full subscription cycle. A variant that wins on conversion but craters sign-ups is a loser, full stop.
Every finding from the audit was sorted into one of three buckets before anything else.
Clear bugs and unambiguous improvements: broken overlaps, duplicate URLs, contradictory copy. Testing these wastes months proving the obvious. They ship, and we monitor before and after.
Genuine trade-offs where reasonable people could argue either way and being wrong is expensive. These get a real experiment with a pre-agreed verdict rule.
Real signals sitting on thin or noisy data. We capture proper evidence before designing anything, so we never build a test on a mirage.
Each one either removes a real hazard or makes the tests possible in the first place.
Paid traffic is currently split across duplicate copies of the Starter Pack page. Roughly half of Google product spend lands on the duplicates. We consolidate to one page and repoint the ads. Until this is done, the flagship test cannot produce a valid answer.
Unblocks Test 1The floating rewards widget overlaps the Checkout button on mobile cart, plus option cards on product and collection pages. It sits on top of the single most important tap on the site. A small CSS fix, shipped immediately, no test needed.
Straight fixFirst: is the current homepage permanent, or a temporary launch takeover? Second: was £82.50 a genuine recent selling price for the strike-through "Save 16%"? Both answers change what we build. Details in the decisions section below.
Needs your answerA dedicated Shopify testing tool, a revenue metric that counts subscription value (not just the first order), and a clean 30-day pull of on-site behaviour data. Every test below reads through this layer, so it gets built first.
Unblocks all testsEach test changes exactly one thing, holds everything else constant, and has its verdict rule agreed before launch.
The Starter Pack page currently pre-selects the £65 subscription over the £69 one-time price, inside a busy first screen showing four prices and a low-contrast buy button. We believe defaulting to the one-time pack with a cleaner price stack will earn more per visitor and cut accidental subscriptions.
Version B defaults to the one-time pack with a simplified price stack and a high-contrast add-to-cart button. Everything else (product, price, photography, shipping copy) stays identical.
Revenue per visitor including subscription value. Each new subscriber is counted at expected lifetime value (£725), then the result is confirmed after a full subscription cycle. Raw conversion alone is never the verdict.
Subscription sign-up rate, refunds and chargebacks, accidental-subscription support tickets. If the variant wins on one-time sales but craters subscriptions, it loses.
On mobile, the Outdoor collection opens with a full screen of editorial copy and zero products. The Tomato Vine collection leads with products and holds attention noticeably longer. We believe showing product sooner moves more visitors from the collection into a product page.
Version B applies the product-first layout to the Outdoor collection. The copy is relocated, not rewritten; the products themselves do not change.
The rate at which collection visitors click into a product page, then add to cart.
Scroll depth, bounce rate, and text readability over photo backgrounds.
The homepage currently renders as a published article: headline, byline, body text, comment form, and no shop button, product grid, or price above the fold. Around £119k of tracked revenue landed on this page in the audit window. We believe a real storefront homepage will move more visitors into products.
Version B is a storefront homepage: hero with a clear Shop button, a collection grid, and visible pricing.
The rate at which homepage visitors click into a product page, plus revenue per visitor.
Bounce rate and scroll depth.
The cart already converts at roughly 45%, which beats the typical DTC store, so this is polish rather than repair. We believe surfacing express payment buttons (Shop Pay, Apple Pay, PayPal) at the cart nudges that rate higher.
Version B surfaces express payment buttons prominently at the cart. Cart logic and shipping thresholds stay identical.
Cart-to-purchase conversion.
Average order value (express checkouts can shrink baskets).
URL consolidation, widget fix, and your two answers. Same engineering pass, no waiting.
Testing tool, subscription-aware revenue metric, clean 30-day behaviour data. Nothing launches without it.
The moment its two prerequisites are live. A 13-week commitment, read on subscription-aware revenue.
Test 2 on the collection page and, if the homepage is permanent, Test 3. Different pages, so they cannot contaminate each other.
Test 4 after the widget fix. In parallel, we capture and review the two "options" router pages where visitors bounce back unusually often, so a future test is designed on real evidence.
None of these are big time commitments, but each one changes what we build or when.
It renders as an article rather than a storefront. If it is a temporary launch takeover, we drop Test 3 entirely. If it is permanent, Test 3 is the fastest and one of the most valuable tests we can run.
The Starter Pack shows "£82.50, Save 16%" as a permanent strike-through. Under UK pricing rules, a "was" price that was never genuinely charged is a legal liability. If it was real, nothing changes. If not, we fix the label before it becomes a problem.
Test 1 needs a pre-agreed line: the maximum acceptable dip in subscription sign-up rate before a variant is rejected, whatever it does for one-time sales. Our suggested starting point is a 20% relative drop.
Test 1 is a 13-week commitment. Starting mid-July means it reads in mid-October, just as Q4 traffic ramps and the audience mix shifts. Starting now also freezes the product page against other changes through that window. The alternative is running the fast tests now and starting Test 1 in January on stable traffic.
If two things change, we cannot know which one worked. Each test isolates a single treatment.
Success metric, guardrails, and the shipping decision rule are written down first, so the result cannot be argued with after the fact.
Checking results early and stopping at a good moment produces false winners. Every test runs its planned course, minimum two full weeks covering both weekends.
A win on the main metric that trips a guardrail (subscriptions, refunds, order value) is not a win.
While a test runs on a page, that page does not change. Photography updates and restructures are scheduled around tests, never during them.
If a page does not have the traffic to give a trustworthy answer, we ship the improvement and monitor it honestly instead of dressing it up as a test.
Figures and revenue estimates come from the June 2026 storefront audit and 90-day trading data; monetary opportunity sizes are estimates, not guarantees. Test durations are calculated from current traffic levels and standard statistical practice (95% confidence, 80% power) and will be re-confirmed once the measurement layer is live. Where behaviour data was thin (the option router pages), this plan says so and gathers better evidence before acting.