The checkout friction audit: 40 checks, one afternoon

Every e-commerce team we've worked with has a number they know too well: cart abandonment somewhere between 65 and 80 percent. What they rarely have is a breakdown of which part of checkout is doing the damage, because abandonment dashboards measure the whole funnel and blame no one. "Checkout" is not one problem. It's a dozen small debts — a field that didn't need to exist, a postcode lookup that trusts no one, an error message written by a payment API — and each collects interest on every session.

Before we redesign a checkout, we audit it. The instrument is forty checks, run in one afternoon with a laptop, a phone, a stopwatch and a test card. We've run it on a skincare brand with nothing to hide, a butcher-turned-providore, and a wine label that sells out of a shed in the Adelaide Hills — Fernleigh Wines. The same checks catch the same failures, every time. Here it is, organised into the five debts that matter most.

Debt one: field math

Every field in a checkout is a micro-tax, paid in attention and compounding in abandonment. The arithmetic is unforgiving: going from fourteen fields to eight isn't a 43% tidier form, it's the difference between a task and a chore.

1. Count every field, including the hidden ones. Most teams guess nine; the real number, once you count "address line 2", "company (optional)" and the marketing checkbox, is usually thirteen to sixteen. Count what's actually rendered, on mobile, with autofill disabled.
2. Kill the second address line. Address line 2 exists for databases, not humans. Merge to one line with a "Add apartment/suite" reveal. We've never seen delivery failure rates move; we've always seen completion move.
3. One name field. First name / last name splits inherited from the CRM. "Your name" handles mononyms, multi-part surnames and orders you haven't imagined — and deletes a field.
4. Interrogate every "optional" field. Optional fields failed some previous meeting. If it's optional, it ships hidden behind an explicit add-control, or it doesn't ship.
5. Price the phone field. Phone is the single most-abandoned field in checkouts we audit. If carriers need it, label it "for delivery updates only" — the qualifier recovers most of the drop. If nobody calls it, delete it and watch.
6. Defer account creation. "Create an account to continue" is a paywall with extra steps. Capture email for the order, then offer account creation on the confirmation screen — "You're done. Want a password so next time takes half as long?" Post-purchase conversion on that offer routinely beats mid-checkout forced registration by a wide margin.
7. Autofill must actually work. Run autocomplete attribute coverage as a checklist item — shipping street-address, cc-number, the lot. On mobile, a working autofill stack is the difference between checkout and thumb-gymnastics.
8. Email is the first field, or near it. It's the cheapest ask, it seeds autofill, and it gives you a recovery address if the session dies. Put it early.

Debt two: wallet placement and payment rhythm

On phone-first checkouts, the express wallet button is the checkout for a large share of buyers. Everything else is a fallback.

1. Wallets appear above the card form, not after it. Apple Pay, Google Pay, PayPal — whichever the audience uses — go at the top of payment and again in the cart drawer. Buried-after-card-fields placement is the single most common unforced error we find; moving wallets up was worth a meaningful, measurable lift in completed mobile checkouts at Fernleigh Wines within the first month. (Illustrative figure; every store's mix differs. Measure yours.)
2. The express path skips the duplicate asks. If the wallet carries the address, don't re-render address fields beneath it "for confirmation". Trust the token.
3. Card inputs brand themselves. Detect the card type from the BIN and show it. Asking "card type?" in a dropdown is 2009 talking.
4. Never clear the card on another field's error. A checkout that wipes the card number because the postcode failed has told the customer you have never watched a human buy anything. Errors affect their own field. Full stop.
5. The last button says what it does. "Pay $84.00" beats "Place order" beats "Submit". "Continue" as a final label is a lie that costs money.

Debt three: address lookup integrity

Address lookup is the section where audits get quiet, because everyone has one and almost nobody has tested it against reality.

1. Test with an address that isn't in the database. New estates, rural routes, units in old buildings. If lookup fails, the fallback must be a graceful manual form — not a dead end with a red box. The Tallow & Co. checkout treated lookup failure as a first-class flow, because a providore shipping to regional NSW meets a lot of addresses the lookups have never heard of.
2. Let people override the suggestion. "Use suggested address" must always lose to "no, I know where I live". Forced normalisation is how parcels tour the state.
3. PO Boxes and parcel lockers work or are honestly refused. Nothing erodes trust like accepting an address you'll later reject by phone.
4. Postcode/suburb state machine matches the shipping country. Shipping to New Zealand with an Australian-state dropdown hard-coded is a classic.
5. Address errors name the fix, not the fault. "We couldn't find that address — check the suburb spelling or enter it manually" beats "Invalid address". See our piece on error messages that de-escalate for the full writing pattern; it applies to checkout hardest of all.

Debt four: error recovery and failure states

Assume failure. Cards decline, sessions expire, couriers' postcode rules change mid-checkout. The audit asks: what does each failure feel like?

1. Declined cards get a human script. "Your bank declined this card — no charge was made. Try again, or use another card." Not "ERROR 5110". The words "no charge was made" do the heavy lifting; fear of double-charging is what sends people to support or away entirely.
2. The summary comes with the error. On submit failure, a summary at the top lists what's wrong, each line a link that focuses the offending field. Scrolling a long form hunting for one pale-red outline is a mobile nightmare.
3. Session expiry preserves the cart. Twenty minutes of inactivity should cost nothing. Cart survives; re-entry lands where you left.
4. Payment spinner has a floor and a ceiling. Instant spinners read as glitches; endless ones read as doom. Minimum 600ms so the state change reads as intentional, honest timeout message with a support path if the processor hangs.
5. Reload-safety. Refresh the page mid-checkout during the audit. If the cart dies, you've found your leak.
6. Stock loss mid-checkout is explained, not silent. "The 250g tins sold out while you were checking out — we've removed them" beats a mysteriously smaller order total.

Debt five: trust proximity

Trust isn't a badge wall in the footer. It's the presence of reassurance at the exact pixel where doubt occurs.

1. Total cost is visible before payment is asked. Shipping surprises at the payment step are the number-one abandonment cause in every study worth citing for two decades, and we still find it weekly. Show the real total — shipped, taxed — on the cart step.
2. Delivery estimate sits next to the pay button. "Arrives Tue 30 Sep – Thu 2 Oct" is a trust element, not a nice-to-have.
3. Returns policy in one line, one click from payment. "Free returns within 30 days" beside the button. The full policy can be a link; the promise can't.
4. Security reassurance near the card field, worded like a human. "Encrypted and processed by Stripe — we never see or store your card number." Specificity outperforms padlock clip-art.
5. Support is reachable from checkout. A help link that opens chat or a phone number — staffed hours stated — rescues orders daily. At GLADE it rescues ingredient questions that would otherwise become abandoned carts.
6. Order review before the final commit on high-consideration purchases. Under $80 impulse? One step. $400 hamper for a client? Give them the review screen.

The second ten: mobile, speed and the small stuff

The remaining ten checks are faster to list than to fail:

1. One-thumb reachability of the pay button in the drawer. 32. Numeric keyboards for numeric fields (inputmode, not just type). 33. No marketing popups inside checkout — ever, no exceptions, fight the meeting. 34. Checkout page weight under control; third-party scripts audited — the script-audit discipline from technical SEO checklists applies here too. 35. Taxes and duties computed, not TBD, for cross-border. 36. Gift options available pre-payment, not post-purchase email. 37. Multi-quantity discount clarity: the line item shows the saving. 38. Confirmation screen carries delivery window, support path and referral prompt. 39. The confirmation email arrives within a minute and renders on a phone. 40. Analytics events on every step and every error — because next quarter's audit should start from evidence, not vibes. Tracking plans before tools is the discipline that makes this data trustworthy.

Running the audit

Book three hours. Two people minimum: one driving, one scoring. Every check gets pass, fail, or "can't tell" — and "can't tell" is a fail with a deadline. Run it on the slowest Android you can find, on hotel wifi, because that's where your Friday-night customers live. Then sort fails into two columns: fixes this sprint (labels, input modes, error copy) and fixes the quarter (field math, wallet architecture). The first column alone usually pays for the afternoon.

And re-run it quarterly. Checkout decays. Campaigns bolt things on; plugins update; someone adds a birthday field "for a surprise". Friction is not a bug you fix once — it's entropy you manage.

Key takeaways

• Abandonment is a dozen small debts, not one big problem — audit, don't redesign blind.
• Field math is arithmetic: every field deleted buys completion. Run the audit before any layout work.
• Wallets belong above the card form on mobile. The express path must trust its own tokens.
• Error copy is checkout UX: name the fix, promise no double-charge, never clear entered data.
• Trust lives next to the doubt: total cost, delivery window and returns beside the pay button.
• Audit quarterly. Friction is entropy.

FAQ

How much abandonment is "fixable"? Some abandonment is window-shopping and comparison — you'll never recover it, and that's fine. In our experience the fixable slice is roughly a third of abandonment: invited-surprise costs, form friction and failure states. That's still an enormous number at most stores' volumes.

Should we A/B test every fix? Test the structural changes (wallet placement, guest checkout, shipping-cost reveal timing). Don't test fixing a bug — a checkout that clears card numbers isn't an experiment, it's an injury. Our CRO experiment design piece covers where the line sits.

Single-page checkout or multi-step? The honest answer is that it depends on basket complexity and device mix, and the audit data tells you which. What we will say: multi-step with an excellent order-summary rail and honest progress beats a long single page on mobile almost every time.

Where does checkout UX sit in a wider engagement? It's typically week one of an e-commerce engagement — highest leverage per hour of anything we do — then feeds the merchandising, PDP and retention work after. Arrive with your abandonment numbers, your field count and one screen recording of checkout on a phone, and the first conversation gets very practical, very fast.