Measuring store speed properly: field data, not screenshots
Why a lab score screenshot proves little, which measurements reflect what customers experience, and a monthly routine for a store.
The short answer
Store speed is usually discussed with a screenshot: a lab score from one run of one page on one machine. It proves almost nothing about what customers experience, because customers are on phones, on variable connections, with caches in every state, across many pages. The measurements that matter are field data: aggregated from real visitors, by page type, on mobile, which is also what search engines use. Lab tools have a different job: diagnosis. Their waterfall shows exactly which images, scripts and apps cost the time on a slow template. Use field data to judge and lab tools to find out why, measure by template monthly, and measure before and after every app install, theme change and banner, because that is when stores get slower without anyone deciding they should.
Judgement versus diagnosis
| Question | Use | Because |
|---|---|---|
| Are we fast for customers? | Field data by page group, mobile | It is what customers and rankings experience |
| Which templates are slow? | Field data by template | Homepage, collections, products, cart differ |
| Why is this template slow? | A lab run on a mobile profile with throttling, reading the waterfall | Shows the images, scripts and apps responsible |
| Did that app make it worse? | Lab before and after; field data a few weeks later | Regressions are tied to changes |
| Are we improving? | Field data monthly, same templates | Trend, not a single number |
The monthly routine
- Open the real-user reports by page group on mobile: homepage, collections, products, cart.
- Note the three measurements per group against last month.
- For any group that worsened, check the change log: apps installed or updated, theme changes, banners, pixels.
- Run a lab diagnosis on that template with a mobile profile and read the waterfall.
- Fix the cause: an image, an app, a script, a missing dimension.
- Record the numbers and the fix; re-check next month.
- Gate changes: no app or banner without a before-and-after lab check.
Reading the store’s own reports
The platform’s speed reports and the search console’s page experience reports both draw on real-user data and both break it down usefully. Look for the templates that carry the most traffic, the mobile numbers, and the trend. A store that watches those three things monthly notices regressions within weeks; a store that looks at a lab score once a year notices them when sales fall.
What this means for you
Judge store speed by field data from real visitors, by template, on mobile, tracked monthly. Diagnose with lab tools and their waterfalls. Measure around every app, theme and banner change. Stop accepting screenshots as evidence. The store that measures this way gets faster over time; the one that collects scores gets slower one harmless-looking change at a time.
Frequently asked questions
Our agency sent a screenshot with a high score. Are we fast?
You were fast on one page, on their machine, on their connection, at that moment, possibly with a cleared cache and a desktop profile. Ask for the real-user data by page type on mobile from the search console or the platform's speed report. That is what customers experience and what rankings use, and it frequently tells a different story.
Which tool should we use?
For judgement: the real-user field reports in the search console and the platform's own speed reporting, by page group, mobile. For diagnosis: a lab tool run on a mobile profile with throttling, reading the waterfall to see which images, scripts and apps cost the time. Both, for different questions.
How often should we measure?
Field data monthly by template, plus a lab check before and after every app install, theme change or new banner. Speed regressions are almost always caused by a specific change on a specific date; measuring around changes is how you catch them while the cause is obvious.
Sources
- web.dev: Why lab and field data can be different (accessed 2026-09-12)