Step One Life  ·  AU / UK / US  ·  Feb – Jun 2026

Re-architecting
discounting and pack pricing

Moving the discount architecture from unit-based tiers to spend thresholds, across three storefronts and two currencies. The commercial finding arrived where the RPV read couldn't see it: over a quarter less promotional cost for the same revenue.

Role

CRO Lead + Ecom Ops — sole owner: hypothesis → build → run → analyse → ship

Programme scale

9 completed experiments, ~780k visitors, 3 storefronts, 2 currencies

Primary metric owned

Revenue Per Visitor (RPV)

Status

AU model shipped and re-confirmed post price-increase. UK concluded with a quantified margin finding. US pack structure tested and held.

01The Problem

Step One's discount architecture was unit-based: buy a 3-pack, save X; buy a 5-pack, save Y. It had three structural weaknesses I wanted to attack:

1. It discounted units, not spend. A customer buying $120 of product across mixed categories got no reward for basket value — only for pack depth on a single style.
2. It was not a traditional ecom ladder. Spend thresholds are the standard mechanic in DTC apparel, and our pricing grid made the jump between pack tiers feel arbitrary rather than aspirational.
3. It was expensive. Unit-based tiers gave away discount on every qualifying order regardless of whether the customer needed the incentive to buy.

The hypothesis I carried into the programme:

If we change the discount structure from a unit-based model to a spend-based model, then we will increase RPV, because it is more of a traditional ecom model and more revenue can be made at the 4- and 15-pack intervals.

I ran this as a multi-market programme, not a single test. AU was the proving ground; UK was the replication with three iterations of root-cause fixes; US tested whether pack structure itself was a lever.

---

02Method

Platform: Shoplift, 50/50 split, storefront-level theme tests. Statistical approach: two-proportion significance test on conversion rate and add-to-cart rate; zero-inflated test on revenue per visitor (per-order revenue padded with zeros to visitor count). Standard 95% confidence threshold, with probability of improvement reported where the test fell short. Segmentation at every read: device, new vs returning, and channel — because the whole question was whether a discount mechanic behaves the same for cold acquisition traffic and warm CRM traffic. Diagnostic layer: SKU-prefix level unit and revenue mix analysis, and pack-tier selection ratios, to explain why a metric moved rather than just that it did.

Critically, I ran the AU test twice — once before and once after a men's price increase landed mid-programme — so the result could not be dismissed as an artefact of old price points.

---

03The AU Result — The Model Worked

Test 1 — AU_Theme_RPV_260305 | Control vs Tiered Pricing D % Off | 5–16 Mar 2026 | 15 days

Mobile carried the volume and led: RPV +6.0%, AOV +3.1%. Desktop followed: RPV +3.8%, ATC +12.0%, CVR +4.7%.

This was the rare outcome where the discount mechanic improved conversion and basket value simultaneously. AOV rose rather than compressed — the opposite of the usual trade-off, and evidence the ladder was pulling people up rather than just letting them in.

Test 2 — AU_Theme_RPV_260317 | Unit vs Spend Based (Post Price Increase) | 16–30 Mar 2026 | 15 days

The model held after the price increase — and CVR crossed into statistical significance. AOV gave back 2.7%, but the RPV gain was larger, which is the trade I would take every time.

Combined AU measured impact: +2.9% incremental revenue across the two test windows, on ~280k visitors.

The AU failure that made the win credible

Before the win, I ran AU_Theme_AOV_260223 (Quantity vs Spend Tiered Pricing v4, 17–20 Feb). It lost:

"Across no-discount customers, the new spend-based discount tiers underperformed the existing units-based 'buy X, save Y' structure, driving materially lower revenue in both Shopify and Shoplift. The variant also reduced average order quantity and AOV, with notable drops in 4- and 7-unit bundles... Recommendation: keep the current units-based discount structure."

I did not ship it. I went back to the threshold design, and AU_Theme_RPV_260305 is what came out of that iteration. An earlier attempt (AU_Theme_RPV_260216) was killed outright as bugged and its results declared invalid rather than reported.

---

04The UK Result — A Replication That Failed, and the Margin Finding That Mattered More

The UK is where this programme earned its keep as a piece of thinking rather than a lucky win. I ran the spend model three times, each iteration fixing the root cause the last one exposed.

V1 — UK_Theme_RPV_260409 | 9–24 Apr 2026 | 15 days | ~80k visitors

Verdict: no rollout. But a clean signal emerged — ATC rate moved significantly while CVR did not. More people were engaging with the offer; the same number were buying. That gap was the problem to solve.

Root cause identified: the tier 1 threshold sat at £75, which a standard 4-pack did not reach. The mechanic was aspirational but not achievable.

V2 — UK_Theme_RPV_260426 | 25 Apr – 21 May 2026 | 27 days | ~290k visitors

Threshold lowered to £65 so a 4-pack triggered tier 1.

This is the finding the case study turns on. On RPV the test was a non-result (65.7% probability of improvement, never significant). On discount efficiency it was emphatic: the spend model gave away 27.3% less in promotional cost — a five-figure sum over 27 days — while generating slightly more revenue (+1.0%). Discount is a direct deduction from revenue, so the contribution improvement was far larger than the revenue line suggests — and it did not require an RPV win to be real.

Two statistical signatures told the story cleanly: - ATC z-score grew monotonically across every read: 3.22 → 3.52 → 3.64 → 4.86 → 5.00. A genuine, compounding effect. - RPV oscillated: +0.57% → +1.18% → +0.58% → +0.89% → +0.95%. Real, but too small to prove at this traffic.

Root cause for V3 identified: because both the 4-pack and 7-pack hit the same 10% tier, they displayed identical per-unit prices. The 7-pack : 4-pack upgrade ratio collapsed from 0.42x (Control) to 0.24x (Variant) — a measurable loss of 7-pack selections and units. Shoppers stopping at 4 were behaving rationally in response to a broken price ladder.

V3 — UK_Theme_RPV_260521 | 21 May – 2 Jun 2026 | 13 days | ~110k visitors

Final iteration: threshold at £119 so the 7-pack Boxer Brief unlocks 15% (vs 4-pack at 10%), corrected PDP button labels ("ADD FOR 15% OFF" on the 7-pack), plus a three-node sidecart progress bar (10% / 15% / 20%).

Control won on RPV. I stopped V3 as a test. Even in defeat the discount-efficiency finding held: discount cost fell 28.0% while revenue fell only 2.5%, a net positive contribution effect.

Where this landed: the spend-based model is now live in the UK with the £65 / £119 / £225 ladder (confirmed in the Step One discount-mechanics KB, updated Aug 2026). The UK adoption did not come from an RPV win — it came from the discount-cost finding across V2 and V3, which showed the spend structure delivering comparable revenue at over a quarter less promotional cost — a five-figure saving — across the two test windows. That is the decision I'd defend: a discount mechanic's job is to protect margin while moving the same revenue, and on that test the spend model won even where the primary metric said it lost.

The honest conclusion: three iterations of progressive UI fixes never recovered pack depth. UPO declined 8.6% and the 7:4 upgrade ratio stayed suppressed. The spend model's real, durable effect is that it lowers the barrier to cart entry (ATC significant in all three iterations) — but a lower barrier without deeper pack selection produces more small orders, not more revenue. That is a structural property of the mechanic, not a fixable implementation detail, and I called it that way.

The one segment that consistently won

SMS outperformed under the spend model in every single iteration — +51.6% (V1), +15.0% (V2), +36.9% (V3 RPV, on ~5k visitors). Warm, purchase-ready audiences respond to a tiered spend message. Every other UK channel was flat or negative, with Paid Search at −26.0% RPV in V3 (AOV −14.6%) — cold, intent-driven traffic is actively damaged by the mechanic.

---

05The US Result — Testing Whether Pack Structure Itself Was the Lever

US_PDP_RPV_260407 | 3,5,10 vs 4,7,15 pack sizes | 7–14 Apr 2026 | ~14k visitors

If bigger pack tiers encourage higher quantity for the discount, the US should reward 4/7/15. It didn't.

Control held. AOV was flat while CVR fell double digits — the structure changed who bought, not how much. SKU analysis confirmed no meaningful mix shift, so the loss was pure conversion drag, worst on desktop (−23.9% CVR).

I also killed UK_Theme_CVR_251118 (Pack Pricing – OOS 10 Pack), a profitability test on striking the 10-pack from the PDP, when it went negative early.

---

06Key Insights

1. Discount architecture is a margin lever, not just a conversion lever. The UK spend model lost on RPV twice and still cut discount cost by over a quarter in both iterations — five-figure savings — for comparable revenue. If the only metric on the dashboard is RPV, that value is invisible. I'd argue this is the single most transferable insight from the programme: instrument discount cost per order as a first-class experiment metric, not a downstream finance artefact.

2. A statistically significant ATC lift is not a win — it's a question. ATC rate was significant in all three UK iterations (z up to 5.00) and converted to revenue in none of them. The gap between "more people engage with the offer" and "more people buy" was the entire substance of the UK programme, and it was never closed.

3. Price parity between tiers destroys the ladder silently. The 7-pack : 4-pack ratio moving 0.42x → 0.24x was invisible in headline AOV. It only surfaced because I tracked pack-tier selection ratios directly. Any tiered-pricing programme should instrument the upgrade ratio between adjacent tiers as a primary diagnostic.

4. Same mechanic, different market, different outcome. AU: +5.5% and +5.4% RPV. UK: +0.47%, +0.95%, −2.34%. The mechanic was identical; the traffic mix was not. UK Paid Search was damaged −18% to −26%, and UK traffic skews colder than AU. Replication across markets is not confirmation — it is a separate test with a separate answer.

5. Channel-segmented rollout beats a binary ship/no-ship. SMS won in every iteration. Paid Search lost in every iteration. The right decision was never "roll out the spend model" or "keep unit tiers" — it was a segmented deployment, and I recommended exactly that rather than forcing a global call.

6. Failures were load-bearing. The Feb AOV_260223 loss is what produced the threshold design that won in March. The V1 threshold miss produced V2. The V2 price-parity diagnosis produced V3. Three of the nine tests were losses or kills, and every one of them changed the next build.

---

07Business Impact Summary

In plain terms: across ~780k visitors and nine experiments I rebuilt Step One's discount architecture in the AU market on a statistically confirmed +5.5% RPV / +2.5% AOV basis, generated measured incremental revenue (+2.9%) in test windows, and identified a five-figure discount-cost saving (−27.6%) in the UK that drove adoption of the spend ladder there despite the primary metric reading as a non-result. The US pack-structure test was a loss, and I reported it as a loss and held the control.

The mechanics now live: AU spend-based ($120 / $200 / $490 → 10% / 15% / 20%) and UK spend-based (£65 / £119 / £225 → 10% / 15% / 20%). US remains on the unit-based model (3-pack save $6, 5-pack save $25, 10-pack save $80) — consistent with the US test result showing the pack restructure did not help.

---

08What I'd Do Differently / Next

• Run the UK test on a contribution-margin primary metric, not RPV. The decision the business actually needed was "which structure makes more money," and discount cost was the swing factor. RPV was the wrong primary metric for a discount-mechanic test, and I'd change that before running it again.
• Instrument the tier upgrade ratio from day one. It was a post-hoc diagnosis in V2. It should have been a pre-registered leading indicator in V1.
• Segment the rollout from the first iteration. SMS was the consistent winner across all three UK reads; a segmented deployment could have captured that value in April rather than leaving it on the table through June.
• Test the "Save £X" framing. Percentage labels require mental arithmetic; pound amounts don't. This was the queued V4 lever when the programme concluded.
• Consider a hybrid model. Women's categories under-indexed on every UK read (Thong index 75, SF Full Brief 77) while Men's over-indexed (Boxer Brief 105, Sport 120, Cloud Mesh 130). The unit model specifically drives women's multi-pack behaviour. Serving the unit model to Women's pages and the spend model to Men's is the untested configuration with the clearest theoretical upside.

---

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_Theme_AOV_260223, AU_Theme_RPV_260216, AU_Theme_RPV_260305, AU_Theme_RPV_260317, UK_Theme_CVR_251118, UK_Theme_RPV_260409, UK_Theme_RPV_260426, UK_Theme_RPV_260521, US_PDP_RPV_260407.