Choose a page and a user state
For example, a mobile visitor needs to read a service and open the quotation form. Record URL, time, device, network profile, browser and sign-in state. Note the version being tested and whether the page contains personalised content.
Do not compare a signed-in editor visit with a cached public page as though the difference alone proves a fault. They may use different delivery paths. An empty browser cache is also different from an empty server or CDN cache; identify the layer being tested.
Read the loading sequence in order
Inspect the document request in the network panel first. A long wait before the first response byte can include connection setup and server work. TTFB is not a pure measurement of PHP execution, and one local value does not describe every visitor.
If HTML arrives early but the main content appears late, follow images, stylesheets, fonts and JavaScript instead. A late-discovered image does not necessarily call for more server capacity. If the page reacts slowly after loading, inspect main-thread work and the particular interaction.
| Observation | Hypothesis | Next evidence |
|---|---|---|
| Document starts late | Connection, cache or server work | Request phases and authorised server logs |
| Early HTML, late image | Late discovery or heavy file | Image request, variant and priority |
| Menu reacts late | Work in the browser | Recording of the menu interaction |
| Only the signed-in view is slow | A different delivery path | Separate test cases |
Try one change in a safe copy
Fictional example: a service page contains a large introductory image and a booking plugin. The team suspects the plugin, but the network chain shows the image being discovered late. The first experiment therefore corrects image delivery in an isolated copy rather than disabling live bookings.
If a theme or plugin needs isolating, use a backup, appropriate permissions and a documented retest. Change one factor at a time where practical. Afterwards check quotations, bookings, editing and consent journeys that may be affected. Cache rules must not expose private content.
Choose improvement or investigate a move
Keep the baseline, change and comparable follow-up runs. State what actually improved and what remains unknown. Relevant field data can complement the test; say when none is available. A better local score does not prove more enquiries.
An existing WordPress solution can often improve without a platform change. If editing, operations and functions need reassessment anyway, a scoped migration discussion may be appropriate. Bring the plugins' real tasks, URL inventory and enquiry journey. Astro does not automatically remove oversized images or unnecessary browser work.