Twelve months, fourteen experiments, six real wins, three flops. The Netorigo Storefront reference store conversion went from 2.1% to 3.6% (+71%). This article is the full list with percentages, plus the lessons.
The six wins
1. Clearer price display (+8.4% conversion)
The biggest single-change win. Previously the PDP showed the price at 36 px with a 12 px "VAT included" under it. VAT was already in the figure but you couldn't tell at a glance. Change: "VAT included" went up to 14 px with medium weight, plus a faint "installments available from X" anchor for high-ticket items. Conversion: +8.4% (n=11,400, p<0.01).
2. Free-shipping threshold visible (+6.1%)
The cart page now always shows a progress bar: "X HUF away from free shipping!". Auto-applies above 25,000 HUF. Visibility nudges buyers toward the "just one more item" pattern. AOV: +12.3%, conversion: +6.1%.
3. Persistent cart (+4.7%)
Cart contents persist 30 days via cookie (anonymous user) or in the account (logged-in). Start shopping on mobile, finish on desktop — one of the longest-tail saves. Measurement: cross-device session pairing on the cart_id cookie correlation. +4.7% conversion, plus return-visit rates (+18%).
4. Faster PDP loading (+3.9%)
PDP LCP (Largest Contentful Paint) went from 2.4s to 0.9s. Two main steps: (a) next/image priority hint on the hero plus fetchpriority="high" HTTP/2 hint, (b) above-the-fold CSS critical-extracted from the main bundle. Conversion: +3.9%, bounce: -11%.
5. Guest checkout (+5.2%)
Account registration used to be mandatory before purchase. Guest checkout moved to the first step, with account creation optional. Conversion: +5.2%, mobile +9.1%.
The internal sales-rep pushback: fewer registered buyers, fewer email-marketing targets. Mitigation: a post-success "save a profile in 5 seconds for faster next checkout" CTA. Account registration didn't crater the feared 50%, only 18% — net to revenue, guest checkout won.
6. Mobile autofill (+3.1%)
The mobile checkout form got autocomplete="shipping street-address", autocomplete="cc-number", plus iOS Safari <input mode="tel"> polish. Mobile checkout conversion: +3.1%, restart-after-validation-error: -34%.
The three flops
1. Gamified checkout (-0.2% conversion, killed)
Experiment: instead of a 3-step checkout, a 7-step one with progress bar, "50% done!" reward strings, the works. Hypothesis: smaller steps reduce cognitive load, progress motivates.
Reality: the 3-step did 2.4% conversion, the 7-step did 2.2%. Not statistically significant, but no improvement either. Plus user feedback: "feels long". Killed.
2. Video PDP carousel (-1.8% conversion, killed)
Experiment: the PDP hero image replaced with an 8-second autoplay video (360-degree product spin). Hypothesis: more info, more buyer confidence.
Reality: the video was ~60 KB heavier, LCP went from 0.9s to 1.6s, and conversion fell 1.8%. The video was prettier, but the speed loss dominated. Killed.
The lesson: speed almost always wins over add-on value.
3. ML-driven personalised coupon (+0.4% conversion, didn't earn ROI)
Experiment: every visitor gets an LLM-generated personal coupon code ("NORBI10" — 10% off). Cost: GPT-4 call per visitor, ~$0.02 per coupon code. Conversion: +0.4%, AOV: -3.1% (the coupon discounts).
Net: the coupon did win some volume, but the -3.1% AOV ate the +0.4% volume. Plus the $0.02 / visitor is a direct cost. Killed.
The end numbers
- Start (May 2025): 2.1% conversion, 22,400 HUF AOV.
- End (May 2026): 3.6% conversion (+71%), 27,800 HUF AOV (+24%).
- Effective revenue index (conversion × AOV): +112%.
The lesson
The revenue-moving changes were boring. Price clarity, cart persistence, speed, guest checkout — none of them are flashy AI or gamification. The magic experiments (video, ML coupon, gamified checkout) all flopped.
The engineering impulse goes to magic. The user data goes to boring. Listen to the data.