Building Checkout around the Customer

The platform served 4M monthly active users and processed roughly $60M in annual GMV across a catalogue of mid-market consumer goods. Mobile drove 72% of all traffic, yet accounted for only 38% of completed conversions. That parity gap between desktop and mobile was the single largest unaddressed lever in the business, and it had been treated as a fixed cost rather than a design problem.

I framed the engagement around one number: the desktop-to-mobile conversion delta. If mobile converted at even half the desktop rate, the revenue implications dwarfed every other initiative on the roadmap. That reframing got the work prioritised over a quarter of feature requests.

There was a hard constraint. We could not modify the backend payment processor or the underlying order model, so every gain had to come from the front-end flow, validation logic, and information architecture rather than from new capabilities.

I started quantitative. Funnel analysis in Mixpanel paired with session replay in FullStory showed a 68% drop at step 3, the address-entry screen. Within that step, 54% of sessions contained rage-clicks against form-validation errors, and 31% of all abandonments happened within eight seconds of the screen loading. The pattern was unmistakable: users were hitting a wall, not drifting away.

Session replay told the qualitative half of the story. On mobile, the address step asked for fourteen fields above the fold with aggressive on-blur validation, producing a sense of cognitive overload and distrust. To turn that observation into evidence, I ran 18 moderated user interviews and fielded intercept surveys to 430 abandoning users.

The key insight reframed the whole problem. Users did not abandon because the form was long; they abandoned because they were asked to create an account before they could see shipping costs. Perceived risk was high before any commitment had been earned. The form was a symptom; the sequencing was the disease.

I put three options in front of leadership rather than a single recommendation, because the decision was as much about appetite for risk as it was about design. Option A patched individual form-validation rules. Option B added a guest-checkout bypass. Option C was a full progressive-disclosure redesign of the entire flow.

I sized each against revenue ceiling, technical feasibility, reversibility, and downstream retention impact. Option A was cheap but capped at marginal gains. Option B was a meaningful single step. Option C carried roughly ten times the revenue ceiling of A because it addressed the sequencing problem at its root rather than its surface.

We sequenced toward C but de-risked it by shipping B first. That gave us a reversible early win, validated the guest-first hypothesis with real money, and bought the team confidence to invest in the larger redesign.

The redesign collapsed seven steps into a progressive single-page flow. Contextual validation replaced on-blur error storms, surfacing guidance at the moment of need rather than punishing intermediate states. The flow became guest-first: account creation moved to after purchase, framed as saving your details for next time rather than a gate in front of value.

I introduced an inline shipping-cost preview the moment a postcode was entered, killing the primary source of perceived risk. A persistent order summary rode along the top of the experience on every step, using a mental model of receipt at the top, fields below to anchor the user in what they were paying for.

Address entry was the highest-friction surface, so I designed and prototyped four variants of address autocomplete before settling on a single-field predictive lookup with a manual-entry escape hatch for edge cases.

Nothing shipped on conviction alone. We ran a three-round A/B programme, each round live for a minimum of two weeks to reach 95% statistical significance before a decision.

Round 1 isolated guest checkout and returned +11% completion. Round 2 layered in the persistent order summary and reached +19%. Round 3 introduced the full progressive-disclosure layout and landed at +31%. The mobile subset, the segment the whole project existed to fix, showed +44%.

Sequencing the rollout as additive rounds meant every component earned its place independently, and we always knew which change drove which gain.