Did the website really get faster?
Compare the same pages and tasks under similar conditions. Otherwise a better result may come from a different test rather than the repair.
Record the starting conditions
Note the exact URL, test tool, device and network settings, login state and time. Include whether it was a first or repeat visit. Record any important third-party content or unusual site condition.
A fictional store's empty cart and a logged-in cart with several products aren’t the same test. A cached homepage and a first uncached request also answer different questions. Decide which conditions matter to the real task.
Repeat enough to understand variation
One run can be affected by transient network or server conditions. Use an appropriate repeated comparison rather than selecting the most flattering result. Keep raw information and explain the method in the report.
Lab and field information should remain distinguishable. The web.dev lab and field guide explains why they may differ. A lab improvement can be verified immediately while real-user reports still reflect an earlier period.
Would these two measurements stand up in a conversation with the shop owner?
A shop asks for help because customers on phones wait too long to see its product photos. Take a product that will stay in stock for the duration of the work, save the address and note which picture is shown first. Make your first visit with a fresh browser profile. Then leave the product page and come back in the same profile. Record the two experiences separately. A returning shopper may already have the picture and the shop's fonts on their device. Someone following a link in a promotion may have neither.
Suppose the team delivers a smaller first image on Tuesday. You repeat the test on Wednesday and the page appears to have slowed down. Before anyone rolls back the change, open the two recordings side by side. Is it the same picture? Did the merchandising team add a sale banner overnight? Does the Wednesday page ask for a delivery location while the Tuesday page already knows yours? Those are ordinary changes in a working shop. They also make it hard to say what the new image alone did.
Look at the actual image request. Which file did the browser choose at the screen size you tested, how large was the transfer, and when did the request start? You might find that the new file really is smaller but the site is waiting for an offer banner before showing it. That gives the owner two useful findings. Customers download less data to see that particular photograph, and there's a separate delay that still keeps them waiting. Neither finding requires you to pretend the overall page feels quicker than it does.
A warm visit needs the same care. The browser can reuse a script file without downloading it again and still spend time running the script. A returning visitor may also be recognized by the shop and see different content. I like to keep a screen recording beside the timing record. It's much easier to explain a change in the numbers when we can point to what the customer was looking at during those seconds.
What to keep with the result
For a small change, a short note can be plenty. Give the product address, the date and time, the device or test profile, the connection conditions and the change you're checking. Say what counted as “ready.” The first photo becoming visible is different from being able to choose a size and add the item to the basket. Include the repeated runs, particularly the awkward one that took longer than the others. If the variation between runs is bigger than the apparent improvement, say so.
If the old version is still available in a suitable test environment, you may be able to compare both versions with the same promotion and the same product pictures. If it isn't, see whether there's a comparable product that missed the banner or wait until the promotion ends and take a new baseline. There are times when the honest answer is simply that two things changed together. You can still report which files got smaller, which request arrived late and what you'd measure on the next release.
When you ask someone to investigate a speed change, send the customer complaint with the numbers. “The size selector won't open while the main photo loads” is enough to shape a useful test. It gives you something you can personally try after the work is finished, on the phone your customers use, even if you never open a performance report yourself.
Check that the page still works
List the functions affected by the change and test them. Script deferral may alter menus, forms, product options or tracking. Caching can affect content freshness and personalization. An improved timing with a broken action isn’t a successful outcome.
Report what changed and what did not. If one page type improved while another remains slow, say so. If physical devices were not tested, don’t imply that a browser simulation established their behavior.
A performance comparison doesn’t prove increased sales or better rankings. Those outcomes need their own data and context over time. The immediate claim should stay close to the measured change.
Our performance approach uses defined tasks and conditions so the result is reviewable by the business owner as well as the developer.
