The checkout-to-purchase rate was sitting at 83–87% — seemingly healthy, but hiding a specific, fixable problem: users were abandoning right at the address form. I led the UX research and redesign, projecting +5% conversion and ~$700/day in additional revenue.
Sheet Music Plus's checkout conversion was 83–87% — good by industry standards, but our data suggested hidden friction. Users were reaching the checkout page and not making it to payment.
The business question was simple: why are users who clearly intend to buy failing to complete the address step?
Two compounding issues were known going in:
UX friction in the address form
Unclear required vs. optional fields, no inline validation, and generic server errors that gave users no actionable guidance.
Technical blockers
Users switching between EU and non-EU warehouses got stuck in infinite loading states. Saved non-EU addresses broke checkout for EU-located users entirely.
Both were silently costing conversions without appearing in headline metrics.
Before designing anything, I ran a structured data study to answer five specific questions about where and why users were dropping off.
Finding 1 — Users start the form, but don't finish it
Over 93% of users with a blank form start filling it. But only 72% of those who begin the Full Address form actually reach the payment page — compared to 92% for the Country-only flow. That's a 20-point gap created entirely by form complexity.
Finding 2 — The disabled Continue button was a major confusion signal
In the Full Address flow without prefill, 52% of users hovered over the disabled Continue button and 36% clicked it — the highest confusion signal across all segments.
Finding 3 — Phone was the most-missed field in EU; ZIP in non-EU
For EU users, phone was missing in 33% of cases — but 73% still recovered and proceeded, suggesting the requirement wasn't communicated clearly rather than users refusing to provide it.
Finding 4 — Errors were generic and unhelpful
The most impactful error was a bare {"name":"Not Found","code":404} — 77% of users who encountered it did not proceed. Neither error gave users any indication of what to do next.
Finding 5 — Delivery choice architecture affected conversion more than price
Conversion peaked at 83.8% with exactly 2 delivery options, dropped to 79.9% with 1 option, and fell again to 81.9% with 3+.
The redesign addressed three layers of friction simultaneously.
Layer 1 — Reduce required fields to the legal minimum by region
After mapping legal requirements for digital vs. physical products across EU, US, and rest-of-world, we stripped each form to its true minimum. For US digital purchases: country and ZIP only. For EU digital: country, name, postal code.
Layer 2 — Fix validation feedback at every point of confusion
Every field now validates inline on blur, not on submission. Required fields are clearly marked. The Continue button, when disabled, shows a tooltip explaining exactly what's missing. Errors were rewritten from internal server codes to human-readable messages with clear recovery actions.
Layer 3 — Restructure the page layout to reduce cognitive load
CTA buttons moved to the top of the page. Order summary moved below the CTA. Delivery options positioned under the address block — visible only after the address is complete. Address autocomplete (Google Places API) was added to reduce manual input errors.
The change was shipped as an A/B test with two primary metrics:
Primary — Checkout → Payment
Baseline 83.85%, MDE 5%. Sample 1,079, duration ~3 days.
Secondary — ARPU
Baseline $0.44, MDE 10%. Sample 114,000, duration ~7 days.
The checkout → payment metric required only 3 days to reach significance — meaning we could make a fast decision without waiting for a full revenue signal.
The experiment launched June 8, 2026. Results are pending — final analysis scheduled June 15.
+5% — Projected checkout → payment conversion. Based on the pre-experiment model.
+$700 — Projected daily revenue. ARPU projected to increase from $0.44 to $0.47.
The pre-experiment data gave strong confidence in the direction: the problems were measurable, the solutions mapped directly to identified failure modes, and the non-EU validation bug fix alone was expected to recover a meaningful segment that was being blocked despite completing the form correctly.
This project was as much a data investigation as a design project. The headline conversion number looked acceptable — it took a structured breakdown of the funnel by form type, prefill state, and error type to find where real money was being lost.
The most surprising finding: prefilling the address form barely improved conversion (81.3% vs 74.4% for Full Address). Users weren't dropping because of effort alone — they were dropping because they didn't understand what the form wanted from them. That distinction changed the design direction from "reduce fields" to "clarify requirements at every step."
The next priority would be the 3DS authentication failure for EU users whose banks don't support it — a 57% churn rate on a recoverable error that currently has no user-facing guidance at all.