A checkout that differs per shopper is only better if revenue per started checkout goes up
Most checkout work is judged on one number: the share of shoppers who start the checkout and complete the order. That works as long as everyone gets the same checkout. Once the checkout starts to differ by how the shopper arrived, which device they use or what they bought before, the number stops telling the whole story. A variant that takes steps away can raise completion and leave the basket the same. A variant that adds offers can raise order value and lose orders. Both can look like progress on the chart the team happens to watch.
The case for differentiating the checkout was made in July 2023 in an opinion piece in Chain Store Age by Jay Myers, co-founder of the checkout software company Bold Commerce. He argues that the usual fixes still send every shopper into the same generic checkout, while shoppers differ in where they come from, which device they use and what they buy, and he notes in passing that a checkout which drives conversions does not necessarily drive revenue. The piece comes from a company that sells customisable checkouts, and its figures carry no source, so read it as an argument rather than as evidence. That passing remark is the part worth building on, because it decides whether any tailored variant is worth keeping.
Set the measure before you build a variant
For each entry route you want to treat differently, track three figures side by side: the share of started checkouts that complete, the average order value, and the revenue per started checkout. Compare them with a control group from the same route that receives the standard checkout. The third figure is the one that decides. A variant that adds 5% to order value but loses more than 5% of completed orders earns less, however good the order value chart looks.
Fix that rule before anyone designs a variant, and test one route at a time. Otherwise every team reads the result on the figure that flatters its own change, and the checkout slowly fills with variants nobody can switch off.
Taking steps away is the test most likely to pass
The cheapest gains come from knowing what a shopper has already done before reaching the checkout. Myers gives the example of a QR code on a package: whoever scans it already knows the product and has no need for the product page. A link that puts the product straight into the basket and opens the checkout removes a step without asking the shopper to give anything up. A logged-in repeat customer should find address and delivery preference already filled in.
Shoppers who come from social media are the reverse case. The article points out that they have seen less of a brand’s own content than visitors on its site, so the checkout has to carry the information they would otherwise have found there. In practice that means delivery cost, delivery time and return terms visible in the checkout itself, next to the total, rather than behind a link to a page the shopper never visited.
Changes like these remove questions instead of adding them. Because they take nothing out of the order, they put little at risk on the order value side, which makes them the easiest variants to keep on the measure above, and the right place to start.
Adding offers is where the measure earns its keep
The other direction, using the checkout to raise order value for shoppers who are not in a hurry, is where the trade-off becomes real. Myers suggests upsells and cross-sells for shoppers who arrive through a search for that specific product and for logged-in repeat customers, whom he sees as less pressed for time, and he wants the checkout to carry loyalty points, subscriptions and in-store pickup as well. Baymard Institute’s research shows how narrow the margin is. It describes cross-sells in the cart and checkout as a delicate balance, and reports that an injected cross-sell which forces a decision before the shopper can continue is often perceived as highly aggressive. In its checkout testing, 66% of users who had started a checkout and had to go through a separate cross-sell step at Amazon showed extreme frustration.
Relevance is the second condition. In a January 2021 study of cart cross-sells, Baymard found that 52% of the desktop sites in its benchmark showed suggestions that were either completely irrelevant or based only on what other customers bought, and observed that a single doubtful suggestion was enough for test users to stop trusting the rest.
If you test offers, keep them inside an existing step, optional, and complementary to what is already in the basket: the charger for the device, the refill for the product. Never add a separate step that has to be declined. Show loyalty points to members who have points to redeem, not to a first-time visitor. Then keep the offer only if revenue per started checkout goes up, not because the average basket grew.
The payment step should vary least
Payment methods are the one area where differentiation needs a firm limit. Changing the order by context helps: a returning customer can find the method they used last time preselected, and a wallet button that does not work on the shopper’s device does not need to be shown. Removing methods based on a segment guess does not. A shopper whose method is missing has no way to ask for it, and the checkout gives no sign that it exists. Change the order, never the list.
In the Netherlands one method needs a fixed, visible place on every route: iDEAL, which shoppers have seen as the combined iDEAL | Wero logo in the checkout since early 2026. Whatever else varies, that one should be where a Dutch shopper expects it.
Start with one route where shoppers clearly differ from the average, such as reorder traffic from a QR code or visitors from a social campaign. Build one variant that removes a step for that route, run it against the standard checkout until both groups have a meaningful number of orders, and keep it only if revenue per started checkout rises.
Sources: chainstoreage.com, baymard.com, checkout cross-sells, baymard.com, cart cross-sells and ideal.nl
