A commerce platform's field performance metrics are poor. Which improvements have the largest effect on conversion?

The order of effect

• Largest contentful paint, which is what the user experiences as "did it load". Driven by server response time, render-blocking resources and the loading of the hero image — and the hero image is frequently the single largest win, through correct sizing, format and priority.
• Interaction responsiveness, which is what the user experiences as "does it work". On a low-end device this is dominated by JavaScript execution, not by the network — and it is the metric most likely to be poor while lab tests pass.
• Layout stability, which produces mis-taps and abandonment and is usually caused by images without dimensions, injected content and late-loading fonts. Cheap to fix and frequently ignored because it does not appear as slowness.

Why field data rather than lab data

The lab measures one device on one network; the field measures your actual users. For a commerce platform serving a wide device range, the lab number reflects the experience of nobody in particular — and the distribution matters more than the median, since the slow tail is where the abandonment is.

Segment the field data by device class, connection type, region and app version. An aggregate that looks acceptable frequently hides a segment where the experience is unusable, and that segment is often growing.

The improvements with the best ratio

• Correctly sized and formatted images, which for a visually-driven commerce catalogue is usually the dominant payload.
• Removing or deferring third-party scripts, which are frequently the majority of the JavaScript and are added by people who never see the build.
• Reserving space for anything that loads late, which fixes layout stability at almost no cost.
• Reducing JavaScript rather than delivering it faster, since execution is the constraint on the devices that matter.

The connection to the business

Performance improvements on the slow tail affect conversion more than improvements on the median, because the median user was already completing the journey. That reframing is what gets the work prioritised — it is a conversion project rather than an engineering one.