Agentic IntentPlatform · 07
Performance
On a phone, over the public internet, three runs, median reported. Not a laptop on office wifi.
Almost no enterprise site is slow because of one decision. It is slow from years of additions, each defensible alone: a tag manager, a consent sheet, a chat widget, an accessibility overlay, three analytics tags from three agencies. Nobody owns the total, so nobody can fix it.
We audit the total, name every third-party script and what it costs in milliseconds, then hold the number with a gate that blocks any deploy which gives it back.
What is in it · 15 deliverables
Everything this layer covers.
01Measured honestly
- On a phone, over the public internet, three runs, median reported. Never a laptop on office wifi
- Core Web Vitals tracked over time rather than screenshotted once
- Field data reconciled against lab data, so you know which one is lying
- Every number published with the method beside it
02Where the seconds actually go
- Every third-party script named individually, with what it costs in milliseconds and bytes
- Render-blocking resources, unused JavaScript and unused CSS itemised
- Tag managers opened up: what is in them, who added it, and what still fires
- Chat widgets, consent sheets and accessibility overlays measured separately, usually the largest single thing on the page
03The build
- Images re-encoded to WebP and AVIF at the widths the layout uses, with dimensions in the markup
- Fonts self-hosted and subset; nothing render-blocking from a third party
- Critical path inlined, the rest deferred
- Immutable fingerprints on every asset, so a return visit is near-instant
04Holding it
- A performance budget written down and agreed
- A gate that fails the deploy when the budget is exceeded
- Monthly measurement, with any regression attributed to the change that caused it
Scope is written into the proposal in your words before anything starts. Anything outside it is quoted in writing first. See the engagement model.
Want this one looked at on your site?
Send the address. The first reply has findings in it, not a calendar link.