Faith Forge LabsWebsite Speed Optimization
5 MIN READ

The speed test looks fine. Why are customers still waiting?

Your laptop, the test tool, and a customer's phone can experience different things. Compare the page and action as well as the score.

A lab test is a controlled example

A test tool uses a particular device profile, network model and page state. That makes it useful for diagnosis and repeatable comparisons. It doesn’t reproduce every visitor's phone, connection, location or interaction.

A fictional store might test its homepage from a fast desktop while customers struggle with a product-options panel on older phones. The homepage result doesn’t answer the interaction question.

Real-user information has coverage limits

Field information describes actual visits under the collection system's rules. It may represent a period, a page group or a broader origin. Some sites or pages don’t have enough information available. Missing data shouldn’t be replaced with an assumed good or bad result.

The web.dev guide to lab and field differences explains these categories. Check the report's label, date window and scope before comparing numbers.

The report and the person on the train

A customer has already booked a seat. She’s on the train, trying to open her confirmation so she can show the ticket number when she arrives. If she says the site took forever, the most useful question you can ask is what she was waiting to see. The home page could be quick while the booking history takes ages. Or the confirmation could appear promptly, with the ticket number a little square of gray that sits there until another request finishes. Running a test on the home page won’t tell you much about either wait.

Write down what you know without asking for her private booking details. Which screen? Phone or laptop? Had she already signed in? Did the page eventually work or did she give up? A slow connection on the train is one possible contributor. You haven't established that the train caused the problem. Perhaps customers who never leave the office wait just as long for their booking histories and only one person has mentioned it.

Now try the journey with a test account that contains a few past bookings. An empty account might skip the very work you need to measure. Open the confirmation from a link, as a customer would, and watch for the ticket number. In a controlled test you can choose a slower network profile, a smaller screen or a less powerful device. You can also run the same sequence again after a change. That repeatability is the reason I'd keep the test, even if its timings differ from what the customer experienced.

Find out who else has to wait

If the site already collects appropriate performance measurements, look at the relevant page type over a reasonable period. How many visits went into the figures? Do they include signed-in pages? Was the result for the entire site combined because too few visits were available for that one address? In a small business, the route that matters most to a paying customer may have very little reported traffic. An empty chart for ticket confirmations doesn't clear the page of suspicion.

Be careful around release day. An email campaign can send a wave of first-time visitors just when the developer ships a fix. If the overall numbers move, you may be seeing the different mix of visitors as well as the change in the site. Compare similar visits where the data allows it. Keep your private test booking handy so you can also check the change under the same conditions you used yesterday.

What I'd put in the work order depends on what we actually find. If the number is already in the first response but stays hidden behind a loading message while the site fetches unrelated recommendations, there's a specific change to discuss. If getting the booking details accounts for most of the wait, investigate that request instead. If we still can't reproduce the delay and only have a handful of visits, I’d record that and agree what to observe next. There’s no need to collect customers’ names or ticket numbers simply to measure when a screen becomes usable.

Bring the speed report to that conversation, but bring the customer's words too. “I was standing at the door and couldn't pull up my ticket” gives the team a much more useful finish line than turning an already green indicator a darker shade of green.

Use the difference to ask a better question

Choose a representative page and task. Record the test settings, login state and whether the cache was already warm. Repeat comparable checks after a change, and keep the real-user review separate where it needs more time.

Ask customers for a plain description of the moment: opening the page, selecting an option or sending a form. A reproducible action can reveal a problem that a general load test misses.

Don’t promise that one passing run means the site is fast for everyone. A useful report explains what was measured, what improved and what remains untested. It also checks that the customer task still works.

Our performance audit service starts with those conditions rather than a universal score target.