Checkout page optimization: the last hundred pixels.

Every field has to be re-justified at the moment of payment.

A form field is cheap to add on a Tuesday and expensive forever afterwards. The test is not “is this useful?” — almost everything is useful. The test is whether it has to be answered before money moves, or whether it could be asked five minutes later when the cost of asking has collapsed.

Run the whole form through that single question and most checkouts lose half their fields. What survives is what the payment genuinely requires, what the law genuinely requires, and what a human will genuinely act on. Everything else moves to the thank-you page or the first onboarding email, where an unanswered question costs you an answer rather than a customer.

The mechanics that matter more than the count

A short form entered badly is still a bad form. One column, not two. Real autocomplete attributes so a browser can fill it in one tap. Numeric keypads on card and postcode fields. Labels that stay visible while somebody types, rather than placeholder text that disappears exactly when it is needed.

And errors that behave: shown beside the field rather than at the top, phrased as what to do rather than what went wrong, and never — under any circumstances — clearing what the buyer already typed. Most checkout forms are tested by their author on a desktop with a saved card. Almost none of the people who use it are in that situation.

Trust is answered questions, not downloaded graphics.

A buyer at a checkout has four questions running at once: what exactly am I paying, what exactly do I get, what happens if this is wrong, and who do I talk to. Everything that answers one of those is a trust element. Everything else is decoration, and some of it is a liability.

The decorative kind is tempting because it is easy: a badge is a five-minute job and a plain-English refund policy is an afternoon of deciding what you actually mean. But a seal for a certification you do not hold, a live-purchase ticker counting nothing, or a timer that resets on reload are all claims somebody can check — and the buyers most likely to check are the ones spending the most.

Write the summary to prevent the refund, not to look tidy.

Most refund requests are not complaints about quality. They are the discovery of a gap between what somebody thought they were buying and what turned up. The order summary is the last surface where that gap can be closed for free.

1. 01Name the thing the way the buyer would. Your internal product name, SKU or cohort code means nothing to them and will mean nothing on a bank statement either. Use the words from the page they just read.
2. 02State the boundary that causes refunds. Every product has one line that separates what people assume from what they get — live calls or no live calls, one seat or the team, twelve months or forever. Put that line in the summary rather than in the FAQ.
3. 03Write the billing shape in words. “$X today, then $X on the same date each month until you cancel” is understood by everyone. A price with a small “/mo” beside it is understood by people who already knew.
4. 04Show the plan total, not just the instalment. If there is a payment plan, the total collected across it belongs on the page. Hiding it is the single fastest way to turn a buyer into a dispute.
5. 05Say when access starts and how long it lasts. Immediately, on a date, or when the cohort opens — and for how long. “Lifetime” is a promise about your own future hosting costs; only write it if you mean it.
6. 06Be explicit about currency and tax. The currency you will actually charge in, and whether the number shown includes tax. A total that changes at the final step is the most reliable abandonment cause on this list.

The billing descriptor, which nobody remembers

A large share of “I do not recognise this charge” disputes are recognition failures rather than fraud: the buyer bought The Practical Systems Course and the statement says something like SP* HLDGS LTD. Set a descriptor that contains a name they have seen, then print it on the confirmation page and in the receipt: “this will appear on your statement as [descriptor]”. It is ten minutes of work against a category of dispute that costs you the money and the fee.

One checkbox, placed after the decision and before the button.

An order bump is the cheapest additional revenue in the whole funnel, and the easiest thing to break a working checkout with. The difference is entirely in placement, quantity and whether it asks the buyer to think again.

• →Below the summary, above the button. The decision has been made; the bump is an accessory to it. Anywhere earlier and it competes with the thing you actually wanted to sell.
• →One, not three. Two bumps do not double anything. They convert a single click into a comparison, and a comparison at the payment step is a reason to come back later.
• →Never pre-ticked. A pre-selected add-on produces revenue this week and refunds, disputes and a reputation next month. It is also the kind of thing card networks and consumer regulators take an interest in.
• →Priced well under the main item. The bump is a rounding decision, not a second purchase. If it needs its own argument, it is not a bump — it belongs after payment as an upsell.
• →Obviously the accessory to what they chose. The transcript pack for the course, the templates for the system, the extra seat for the team. Something a buyer can evaluate in one line without leaving the page.

Everything that needs more than one line belongs on the other side of the payment — a one-click upsell after the card has cleared, where the buyer is already a customer and a “no” costs you nothing. Order bumps and upsells covers what to put in each slot and how to tell which is which.

A declined card is a customer who already said yes.

Every other section of this guide is about persuading somebody to decide. This one is about people who already decided, told you so with their card, and then hit something neither of you controls. They are the most recoverable group in the entire funnel and the one most businesses do nothing about.

Cards decline for ordinary reasons: an expired card, a mistyped security code or postcode, a limit, a bank’s routine check on a payment to an unfamiliar business, an abandoned two-factor step. Almost none of it is fraud and almost none of it is a verdict on your offer.

On the page

1. 01Never show the raw gateway error. “do_not_honor” tells a buyer nothing except that something technical went wrong and it might be their fault. Translate it into a sentence a person can act on.
2. 02Give an ordered list of what to try. Check the number, expiry and security code; try a different card; wait a few minutes and try again; contact us and we will help. In that order, because that is the order of likelihood.
3. 03Keep everything they typed. A form that empties itself after a decline turns a two-minute retry into a decision about whether this is worth doing twice.
4. 04Offer a second payment method if you have one. A second route past a single bank’s risk rules is often the whole fix, and it costs a configuration change rather than a redesign.

In the inbox, within the hour

If they got as far as entering an email, you can follow up — and should, once, quickly, while the intention is still live. The email has one job: make it obvious this was routine and hand back a working link.

Subject: Your card was declined — here is the way back

[First name] — the payment for [product name] did not go through.
Your bank declined it, which is usually one of four ordinary
things: an expired card, a mistyped security code or postcode, a
limit, or a routine check on a payment to a new business.

Nothing was charged and nothing was lost. This link goes straight
back to the same checkout:

[Checkout link]

If the same card declines again, a different one normally works.
If it does not, reply to this email and [name] will sort it out
with you directly.

Two emails, then stop — and mean it.

Somebody who reached your checkout and stopped is not the same as somebody who read your page. They stated an intention and then hit something. Follow-up is legitimate precisely because of that — and stops being legitimate the moment it becomes volume.

Subject: Your checkout for [product name] is still open

[First name] — you got as far as the checkout for [product name]
and stopped there, which in my experience usually means a question
rather than a change of heart.

The one people ask most often is [the question you actually get
asked most], and the answer is [the real answer, in one or two
sentences].

If that was yours, the checkout is still where you left it:

[Checkout link]

If it was not, reply and tell me what it was — I would rather
answer it than sell you the wrong thing. And if the answer is
simply no, that is a completely reasonable answer. This is the
second of two emails, and then I stop.

What makes it respectful rather than grubby

• →A cap you actually enforce. Two sends, or three where a real deadline exists. The third email in a sequence with no deadline is the one that turns a warm prospect into an unsubscribe.
• →No manufactured scarcity. “Your cart expires in twelve minutes” when nothing expires is a lie the buyer can test by waiting thirteen minutes. Some of them will.
• →Suppression the instant they buy. The most visible failure in this entire system is a “you left something behind” email arriving after the receipt. It tells the customer the two halves of your business do not talk to each other.
• →A real way to say no. Reply, or a one-click way out of this specific sequence rather than out of everything you send. People who decline politely today buy the next thing.
• →Behaviour, not a broadcast. This is triggered by one person doing one thing, which is what makes it welcome. Treat it as a segment to blast and it becomes exactly the thing everybody complains about.

Where it sits in the wider system is covered in the follow-up funnel, and the mechanics of firing it from a single action are in behavioural trigger workflows.

Find out where they stop before you change anything.

“Checkout optimization” is three unrelated problems wearing one name. Which one you have is answered by a single question: at what point do people disappear?

1. 01They never reach the checkout. Then the checkout is not your problem and optimising it will change nothing. The argument, the price presentation or the traffic is what needs work — start with funnel diagnostics and fix the stage that is actually leaking before you touch this one.
2. 02They reach it, start, and leave. This is the checkout, and it is almost always one of three things: a cost that appeared later than the buyer expected, a field they did not want to answer, or a question the page in front of it never answered. In that order of likelihood, and all three are fixable in an afternoon.
3. 03They finish, and the payment fails. A design problem this is not. These people decided to buy and told you so with their card. Everything they are worth now depends on the recovery path — the message on the page, the retry email, and for subscriptions the dunning sequence behind them.

Then change one thing at a time and give it a full cycle. A checkout is the one page where a change you cannot explain is worse than no change: if revenue moves and you altered four things, you have learned nothing and you cannot undo the one that hurt.