checkout_conversion_audit v1.0.0

name: "checkout_conversion_audit" description: "Audit a checkout for lost sales with a scored sheet: total cost shown early, guest checkout, form field count and order, payment and delivery options, trust signals at the payment step, error and decline handling, and recovery messages, each fix ranked by evidence weight, the store's own step drop-off, and effort. Use when a store loses buyers at checkout, before a checkout redesign or platform change, or when cart abandonment is high and nobody knows which fix comes first."

Checkout Conversion Audit

Purpose

Find where a checkout loses buyers who already decided to buy, score it, and hand back a fix order the store can work through. Baymard Institute puts the average cart abandonment rate at 70.19 percent, and 42 percent of US shoppers in its survey said they left because they were just browsing or not ready to buy. That segment is largely out of reach. The rest left for reasons a store controls. Shopify, reporting the same Baymard survey with the browsing segment removed, lists extra costs too high at 40 percent, delivery too slow at 20 percent, distrust with card information at 19 percent, forced account creation at 18 percent, a too long or complicated checkout at 17 percent, and site errors at 17 percent. This audit checks each of those causes in the live checkout, in that evidence order.

Scope line: this skill audits and ranks. It does not build the recovery flow. Recovery message design, discount discipline, and holdout measurement belong to the abandoned cart recovery skill. Section G of the audit only checks that a sane recovery flow exists and stops on purchase.

Workflow

1. Get the store's own numbers first. Pull the last 30 to 90 days: carts created, checkouts started, orders completed, and drop-off at each checkout step, split by phone and desktop. Calculate cart abandonment as (abandoned carts / carts created) x 100, and keep one definition for the whole audit, because platforms count carts, checkouts, and bot traffic differently. Record the step drop-off percentages in the audit sheet. External benchmarks set the checklist. The store's drop-off sets the priority.

1. Walk the checkout as a first time buyer. Place test orders on a phone and on desktop, in a private window, with no saved account. Cover a normal order, an order with bad data (bad postcode, expired card date), and, where the store and processor allow a test mode, a declined card. Screenshot or record every step. Never use a real customer's card or data for any test.

1. Score all 20 checks in references/audit-sheet.md. Score 0 for fail, 1 for partial, 2 for pass, and attach evidence to every score. The sheet covers total cost early, guest checkout, form count and order, payment and delivery options, trust at the payment step, errors and declines, and recovery. Count form elements the Baymard way: every field, checkbox, drop-down, and radio group shown by default. Their usability testing found a flow can work with as few as 12 elements, against an average US checkout of 23.48 elements by default, so treat 20 or more as a fail.

1. Run the scorer and rank the fixes. Fill examples/harbor-lane-audit.csv format with the store's scores, weights from the sheet, step drop-off, and a rough fix effort in hours, then run bin/audit_scorer.py <sheet>.csv. It returns a weighted score out of 100, a score per section, and fixes ranked by weight times gap times step drop-off, divided by effort. Read the ranking critically: if guest checkout, total cost, or form count failed and the overall score is under 40, fix those first even when a cheaper partial outranks them, because they gate every later read.

1. Write each fix as a testable change. One change, the page or step it touches, the expected effect stated as a step conversion read, and the guardrail (overall conversion and average order value must not fall). Ship one change at a time where traffic allows, and hold each read for a full buying cycle that includes weekends. Where traffic is too low for a split test, ship in ranked order and compare step conversion against the pre-audit baseline from step 1, and say plainly that the read is before and after, not a controlled test.

1. Check the recovery basics, then stop. Confirm one recovery flow exists, the first touch lands about 1 hour after abandonment with later touches at about 24 hours and 48 to 72 hours, any incentive waits for the last touch, and the sequence stops the moment an order lands. Prove the stop with a test order. Report gaps and hand the build or rebuild to the abandoned cart recovery skill. Do not design messages inside this audit.

1. Re-audit after the top fixes ship. Re-score only the changed checks plus form count, re-run the scorer, and compare section scores and step drop-off against the first audit. Baymard estimates the average large site could gain a 35.26 percent conversion increase from checkout design improvements alone. Quote that as their estimate for large sites, never as a forecast for this store. The store's forecast is its own step data.

Worked Example

Harbor Lane Home Goods is a fictional home goods store used here as a worked example. Its audit inputs sit in examples/harbor-lane-audit.csv: 20 checks scored from test orders, with step drop-off from its analytics (a made up but realistic set: 31 percent leave at the account step, 27 percent at the address form, 22 percent at shipping cost).

Running python3 bin/audit_scorer.py examples/harbor-lane-audit.csv gives:

• Score 25.2 out of 100, graded critical. 9 fails, 9 partials, 2 passes.
• Weakest sections: total cost 12.5, guest checkout 15.4, errors 16.7, forms 19.6. Recovery is strongest at 72.2, which fits a store that bought a recovery app before fixing the checkout itself.
• Top ranked fix: B1, guest checkout as the default, priority 11.79. It failed, the account step has the highest drop-off at 31 percent, and the fix is a settings change estimated at 2 hours.
• Next: a security note beside the card fields (E1, 1 hour), returns terms visible before payment (E2), wallets on (D1), and the full total in the cart (A1).
• Form count (C1) ranks only eighth despite a fail, because cutting 23 elements to 14 is an 8 hour job. The audit still flags it as a gate fix under the under 40 rule in step 4, so the fix order the store receives is: guest checkout, total in cart, form count, then the ranked list from E1 down.

That last judgement is the point of the audit. The scorer sorts. The auditor checks whether a low ranked fail blocks the fixes above it, and says so in the report.

Output Contract

A completed audit delivers:

• The filled audit sheet (CSV in the scorer format) with a score and an evidence note, screenshot, or recording reference for all 20 checks, plus the phone and desktop test order records.
• The scorer output: total score out of 100, section scores, and the ranked fix list, with any gate fix overrides from step 4 named and explained.
• A fix plan: each fix as one testable change with owner, effort estimate, the step conversion read that will judge it, and its guardrail metrics.
• A baseline sheet: the store's abandonment rate and step drop-off for the audit period, the definition used, and the date range, so the re-audit compares like with like.
• A recovery gap note: what section G found, and a clear handoff line when the recovery flow needs building or rebuilding under the abandoned cart recovery skill.

Operating Rules

• Evidence or it did not happen. No check gets a pass from the store's say so or from a settings screenshot alone. The score comes from walking the live checkout.
• Keep benchmarks in their place. Baymard and Shopify figures explain why a check exists. They never substitute for the store's data, and never get quoted to a store owner as a promised lift.
• Test orders only, in test mode where the platform offers it, and cancel or refund every test order the same day. Never test with a real customer's payment details.
• Respect platform limits. Some hosted checkouts cannot change field order or remove certain fields. Score what the shopper sees, record the platform limit in the evidence note, and rank the closest allowed fix, such as collapsing optional fields, instead of pretending the limit is a choice.
• Accessibility is part of the score, not an extra. Redundant entry, autocomplete values, and text errors are scored checks (C2, C3, F1) because WCAG 2.2 criteria 3.3.7, 1.3.5, and 3.3.1 with 3.3.3 cover them, and every shopper pays for the friction they create.
• Report a bad score plainly. A critical score with a clear fix order is a useful audit. A softened score helps nobody and breaks trust in the re-audit.