Site Care / Performance

10 Client sites fixed in one afternoon

What Site Care Looks Like When It Is Measured: Ten Sites to 90+ in an Afternoon, Then Checked Again a Month On

On 4 October 2026 we ran every client site through Google’s PageSpeed API, fixed what the numbers pointed at, and ran it again. On 3 November we measured the same ten a third time, with no work in between, to see what held. Three columns, and a correction a week later where a single run had misled us.

Ten clients, nine industries New Zealand 2026
Site care: the speed pass website preview

10

Client sites fixed in one afternoon

8 of 10

Still 87 to 99 on mobile a month on, untouched

3

Measurements: before, after, one month later

> THE PROBLEM

Every site we build ships fast. Then it lives: analytics tags get added, photos get uploaded at full size, a plugin update changes how a page loads, a hero animation that looked great in review turns out to hold the page blank until the JavaScript arrives on a phone.

We check the fleet monthly. On 4 October 2026 the check showed ten sites under 90 on mobile that had all launched above it. None of the owners had noticed; the sites looked fine on an office connection. Google notices, and so does a customer on a phone in a car park.

The usual answer is a vague "performance review" and an invoice. We wanted the number before, the number after, and the cause of each gap in one sentence.

• • • • • • • • • • • • • • • • • • • • • • • • •

> WHAT WE BUILT

We pulled Google’s PageSpeed Insights API for every client homepage, mobile and desktop, and read what each report pointed at rather than guessing. Five patterns covered almost everything: hero text animated in from invisible so the biggest element waited for the JavaScript; Google Fonts loaded from a third party in front of the first paint; analytics and session-recording scripts fighting the hero image for the first seconds of a phone connection; a 230 KB logo where a 21 KB one would do; and one e-commerce homepage running five database queries on every visit.

Each fix was small and specific. Hero entrances became CSS animations that paint on the first frame. Fonts were self-hosted. Analytics moved to load after the page is idle. Images were resized to what is actually displayed. The e-commerce homepage caches its data for ten minutes and renders its hero on the server. Every change was deployed and re-measured the same afternoon.

Two sites were left alone on purpose: one whose live domain still points at an old site awaiting the client’s go-live, and one where the remaining fix needs the client’s approval because it touches how their analytics loads. Site care is knowing where the line is as much as moving the number.

On 3 November we ran the same ten homepages again without touching anything first. Eight held within the normal run-to-run noise. Two did not on the day, Solis at 73 and our own site at 78, and we wrote them up as slips. On 8 November we ran each three more times and took the median: Solis 87, our site 97. The single run on 3 November was noise, not a regression, and PageSpeed on a phone profile can swing twenty points between runs on the same page. The lesson stays in the write-up: one run is a reading, three runs is a measurement, and a speed pass that only reports the good month is a brochure.

TECH STACK:

PageSpeed Insights APINext.jsAstroTailwind CSSNetlifyVercelCloudflare
• • • • • • • • • • • • • • • • • • • • • • • • •

> THE RESULTS

  • Mobile, before / 4 October / 3 November. Flyinn 68 / 96 / 97. Malama Birth and Being 68 / 98 / 99. Arborist Auckland 75 / 93 / 97. Covermania 83 / 99 / 94. Tommy Paint Redo 83 / 98 / 92. cre8 Collective 85 / 97 / 98. TLC 86 / 99 / 99. Tridence 87 / 98 / 98.
  • OPTC (e-commerce, 2,600 products): 74 / 82 / 87 on mobile, desktop 98. Still climbing a month on as the cached home page and server-rendered hero settle in; the rest is a server in another country.
  • Solis Panelbeaters: 92 / 97 / 87 on mobile (the 87 is the median of three runs on 8 November; a single run on 3 November read 73), desktop 98. Largest paint 3.2 seconds, with analytics, call tracking and a review widget added in October still loading after the page is idle.
  • Our own site: 63 before the October rebuild, 92 after, 97 now on mobile (median of three runs on 8 November; a single run on 3 November read 78), largest paint 2.4 seconds. We hold ourselves to the same rule, which is why the 78 is still here.
  • Desktop on 3 November: 94 to 100 on every site measured.
  • Four sites built or moved since the October pass, measured on 3 November for the record: PixelWheels 94 (was 32 on the old site), Glasshouse Consulting 98 (new), Wizard IT 99 (was 53), TLC 99.

Measured with Google’s own API: before, after, and again a month later. No agency screenshots. The scores are from the PageSpeed Insights API, the same numbers Google shows anyone who looks a site up, and the fleet is re-checked every month as part of site care.

Visit Site care: the speed pass
★ ═══════════════════════════════════════════ ★

> QUESTIONS THIS PROJECT ANSWERS

• • • • • • • • • • • • • • • • • • • • • • • • •
★ ═══════════════════════════════════════════ ★

> GOT A SIMILAR PROBLEM?

Tell us what's broken, slow, or missing. We'll scope it and send you a fixed quote.
Websites from $3,000 NZD. Stores and apps priced on scope.

> INITIATE CONTACT

Tell us what you need. You get a fixed quote back, usually within two working days, and no sales call unless you want one.

What happens next: a reply within a working day, a fixed quote within two, and if it is a yes, your home page on a private link inside the first week.

No newsletter, no follow-up sequence. One person reads this and replies.