The real cost of a slow checkout

We have run enough commerce replatforms to recognise the meeting. Conversion has drifted down for three quarters. The team has a list of hypotheses, all about the design. Almost none of them are about the network.

Checkout is the one flow where every millisecond has a commercial owner. It is also the flow where teams most often measure the wrong thing. They track click to next screen. The number that matters is intent to confirmation.

Where the time actually goes

On a typical storefront, the buy button starts a chain: analytics, session refresh, inventory check, promotion lookup, payment intent, address validation, fraud scoring, and a confirmation write. Each step looks reasonable on its own. Together they turn a two-second decision into a nine-second wait.

The fix is rarely a bigger server. Remove steps from the critical path, parallelise the work that does not depend on the previous response, and treat the payment provider as a dependency you can degrade around, so it stops being a single point of failure.

Measure the flow

We ask clients for four numbers before we touch the design: p75 time to first product view, p75 time from add-to-cart to checkout ready, payment authorization latency, and the abandonment rate at each step. If those numbers do not exist, building them is the first engagement.

Once the flow is instrumented, the argument changes. The team can show a before-and-after on the same cohort, with a confidence interval. Finance can engage with that.

The cheap wins are usually architectural

Server-render the product and cart shell. Cache inventory reads for a bounded number of seconds and make the reservation explicit. Move promotion evaluation to the edge where possible. Preconnect to the payment provider. Render a skeleton that matches the final layout so the page does not shift under a spinner.

None of this requires a redesign. It does require someone to own the critical path end to end. That ownership is usually the actual gap.

What we would do first

Start with a two-week performance review of the live flow, on the same network your customers use. Instrument the four numbers above, record a session trace on a mid-range Android device, and list every step between intent and confirmation.

Then fix the two steps with the worst ratio of latency to business value. In our experience that alone buys more conversion than a new hero section, and it keeps buying it every month.