Why checkouts leak
The checkout's job is to take an intention and turn it into an order with as little work and as little doubt as possible. Each step adds work — fields to complete, choices to make, pages to wait for — and each piece of unexpected information adds doubt: a delivery cost, a delivery date, a request for a phone number nobody explained. People abandon when the work or the doubt outweighs an intention that was, for some, only moderate to begin with.
The Baymard Institute's long-running research on checkout usability has repeatedly found the same reasons near the top of the list when shoppers explain why they abandoned: extra costs that appeared late, being required to create an account, a checkout that was too long or complicated, and doubts about payment security, alongside slow delivery and a lack of payment options. Nielsen Norman Group's form research adds the mechanics: short forms, clear labels, sensible defaults and specific error messages are the difference between a form that gets completed and one that does not.
Two things follow for the practitioner. Most checkout problems are known and fixable without original research. And the fixes are testable, because a checkout has a measurable rate at every step — which is what the funnel analysis tutorial measures, and what the conversion rate calculator turns into comparable rates. The funnel and checkout hub explains why the late steps are usually the best place to work.
Friction points
| Friction point | Where it shows in the funnel | What shoppers experience | Fix |
|---|---|---|---|
| Forced account creation | First checkout step | A wall between the basket and buying | Guest checkout by default; offer the account after the order |
| Long or multi-page forms | Each form step | Many fields, unclear progress | Remove fields; show progress; one column; sensible defaults |
| Address entry | Delivery step | Typing a full address on a phone | Address lookup; autofill support; postcode first where the market uses it |
| Late delivery cost or date | Delivery step, basket | A total that grew | Show delivery cost and estimated date on the product and basket pages |
| Limited payment methods | Payment step | Their usual method is missing | Offer the market's common methods, including wallets |
| Security doubt | Payment step | An unfamiliar page asking for a card | Consistent branding; visible security cues; no redirect surprises |
| Validation errors | Any form step | A red message with no explanation, or the form cleared | Inline, specific, with the input preserved |
| Discount code field | Basket, payment | A reason to leave and search for a code | Collapse it; show applied promotions clearly |
| Distractions | Every step | Navigation, promotions, cross-sells pulling attention | Strip the checkout chrome; keep one path forward |
| Mobile layout | Every step on phones | Small targets, wrong keyboards, buttons below the fold | Input types that bring up the right keyboard; a sticky continue button; test on real phones |
Each fix is a hypothesis for a test with the step's rate as the primary metric and order value (AOVAverage revenue per order.) and refund rate as guardrails; the table is where the research and the testing programme meet.
Guest checkout and accounts
A forced account is the single largest avoidable wall in most checkouts. The shopper came to buy, and is asked instead to invent a password, confirm an email and remember a login they did not want. Baymard's research places forced account creation among the most-cited reasons for abandonment, and the fix is well understood: let the shopper buy as a guest by default, and offer the account afterwards, on the confirmation page, when the email and address are already captured and the account is a single click rather than a form.
The business case for accounts — repeat purchase, order history, marketing consent — is real and is not served by the wall. Accounts created under duress are abandoned; accounts offered after a successful order, with a stated benefit (track this order, reorder faster), are used. Where a login exists, make the path for returning customers fast: an email field that recognises the address and offers the password or a one-time code, rather than a separate login page that sends the shopper away from the basket.
Watch the first checkout step rate by new against returning visitors after any change here, because the two groups respond differently and an improvement for new visitors can hide a regression for returning ones who liked the old login.
Address and form entry
Address entry is where the checkout becomes a form, and the general rules of form design apply; the forms guide covers field count, labels, input types and validation in full, and this section takes only what is specific to checkout. Addresses are long, structured and tedious to type on a phone, so a lookup that completes the address from a postcode or a few characters removes most of the typing, provided it handles the addresses it cannot find gracefully and lets the shopper type the whole thing when they prefer. Browser autofill works only when the fields carry the standard autocomplete attributes and are ordered as browsers expect, which is a one-off engineering task with a lasting effect. The billing address should default to the delivery address with a single control to change it. Phone number requests need a stated reason (delivery updates) or shoppers assume marketing. Fields the business does not use should go: every field has a completion cost and a privacy cost.
Nielsen Norman Group's recommendations — a single column, labels above fields, no placeholder text as the only label, optional fields marked as such, field length matched to the expected input — all apply, and the mobile version matters more than the desktop one because that is where the typing hurts. Test on the slowest common phone, with the on-screen keyboard open, and watch where the continue button goes.
Costs, shipping and surprises
The most-cited reason for abandonment in Baymard's research is costs that appeared late: delivery, taxes and fees that the shopper discovers at the delivery step after committing effort. The behaviour is rational. A total that grows feels like a trick, and a shopper who has not yet entered payment details has little to lose by leaving to compare.
The fix is to move the information earlier. Delivery cost and the estimated delivery date belong on the product page and in the basket, calculated for the shopper's likely location where that can be inferred, with a clear statement of the threshold for free delivery if one exists and how far the basket is from it. Taxes should be included in displayed prices where the market expects it. Where cost genuinely depends on the address, show the range early and the exact figure as soon as the postcode is known, before the shopper has typed the rest. Delivery options at the delivery step should be few, clearly priced, and ordered with the default most people choose first; a long list of carriers and speeds is a decision the shopper did not want.
Delivery timing deserves equal attention. A date — "arrives Thursday" — converts where a duration — "a few working days" — does not, because it answers the shopper's actual question. Where the date cannot be promised, say what can be, and never let the checkout be the first place the shopper learns that the item is out of stock or back-ordered; that belongs on the product page, as the category and product page guide discusses under availability handling. A date shown at checkout is only worth showing if the store keeps it, and the guide to setting and keeping delivery promises covers setting that promise with operations and measuring how often it is kept.
Payment and trust
The payment step is where doubt peaks, because it asks for the most sensitive information on a page the shopper may be seeing for the first time. Two kinds of problem live here: the shopper's preferred method is missing, and the shopper does not trust the page.
On methods, the rule is to offer what the market uses: cards, the digital wallets common on the shopper's device, bank transfer or direct debit where those are normal, and pay-later options where they fit the price point and the business is comfortable with their terms. Wallets matter on mobile because they remove the typing entirely — address, card and confirmation in one gesture — and a wallet button on the basket page lets some shoppers skip the form altogether. Each additional method is also a maintenance and reconciliation cost, so the choice is deliberate rather than exhaustive.
On trust, the cues are mostly about consistency. A payment page that looks like the rest of the site, with the same branding and no unexplained redirect to a third-party domain, is trusted; one that suddenly changes appearance is not. Security signals — the padlock, a short statement of how card details are handled, recognised payment logos — help at the point of entry, not scattered across the page. The order summary, delivery address and total should be visible on the payment step, so the shopper confirms what they are paying for without going back. Error handling on card entry is the sharpest trust test of all: a declined card needs a plain message and an immediate alternative, not a cleared form.
Errors and recovery
Errors happen at every step, and how the checkout handles them decides whether a stumble becomes an abandonment. The rules from form research are specific: validate inline as the shopper leaves a field, say exactly what is wrong and how to fix it, keep everything the shopper has typed, and never show a generic failure for a specific problem. Address lookup that finds nothing should fall back to manual entry without a dead end. A payment that fails should say why in plain language where the provider allows, offer a retry and another method, and preserve the basket and the delivery details. Timeouts and session expiry should be long enough for a slow shopper and should restore the basket when the shopper returns. To see how often each of these failures happens, record them as events when the site confirms them, as the tutorial on tracking errors and failed actions describes.
Recovery is the second line. Some shoppers will leave anyway — interrupted, comparing, not yet sure — and a recovery message that reminds them of the basket, answers the doubt that most likely stopped them at that step, and makes returning a single click brings a share of them back. The abandoned cart flows tutorial builds the flow; the checkout's job is to capture the email early enough to send it, which argues for the email field at the top of the first step, and to make the return link land on a restored basket rather than an empty one. Incentives in recovery messages are a policy decision with a known trap: they teach abandonment and erode margin, and they should be tested against a no-incentive flow before becoming the default.
Measure recovery alongside the fixes, with the step rates, the conversion rate (CVRShare of visits or clicks that convert.) and order value as the KPIA metric tied directly to an objective. set, and read a recovery programme that lifts orders while shrinking baskets with suspicion.
A garden supplies shop's checkout rebuild
A garden supplies retailer measures its checkout for a month: 18,400 checkout starts, 11,900 past the account step, 9,700 past delivery, 8,600 past payment and 8,300 orders, with an average order value of £64. Recordings show the account wall and the delivery step as the stall points; the delivery page is the first place shoppers see the £7.95 delivery charge, and a third of mobile shoppers struggle with the address form. The team tests three changes in sequence over the quarter: guest checkout by default, which lifts the account step from 65% to 81%; delivery cost and date on the product page and basket, which lifts the delivery step from 82% to 88% and slightly reduces checkout starts as some shoppers now leave at the basket instead; and postcode lookup with autofill attributes, which lifts the mobile delivery step rate by a further 4 points. Orders rise to about 11,100 a month on the same traffic, with average order value flat at £63 and the refund rate unchanged.
If the delivery and payment steps lose far more shoppers on phones than on computers, the page on a low mobile conversion rate shows how to compare devices within the same source before deciding which step to fix.
Shoppers who leave checkout often receive a recovery email, and the test design for discounts in abandoned cart emails shows whether an incentive there earns more margin than a plain reminder.
Common mistakes
- Treating the checkout as a form problem only. Costs, delivery dates and payment methods leak more than field count does.
- Hiding delivery cost until the delivery step. The shopper finds it anyway, and leaves.
- Forcing the account before the order. Offer it after, when it is a click.
- Measuring the checkout as one rate. A single number hides which step leaks; measure every step.
- Testing on a desktop. The typing, the keyboard and the button position are mobile problems.
- Clearing the form on error. The shopper does not type it twice.
- Adding payment methods without a reason. Each one is maintenance; add what the market uses.
- Recovering with discounts by default. Test the no-incentive flow first; discounts teach abandonment.
- Redesigning instead of fixing. A full checkout redesign tests everything at once and learns nothing; fix the biggest leak, test it, then the next.
