The page speed playbook
Speed work fails when it optimises a lab score once and declares victory. The durable version measures real users, fixes by layer, and treats regression as the default to defend against.
Store speed is a revenue input — conversion, ad efficiency, and rankings all price it in — and it decays by default as apps, scripts, and images accumulate. This playbook runs the discipline: establish field-data baselines per template, fix in impact order at the correct layer (edge configuration versus page structure), and install the regression tripwires that make the improvement permanent instead of a quarterly heroic effort.
Stage 1: Baseline on field data, per template
Lab scores mislead in both directions. Collect real-user LCP, INP, and CLS segmented by template — product, collection, home, checkout — and by device. The revenue-weighted view tells you where to start: a slow product template on mobile usually outranks everything else in impact.
Real-user Core Web Vitals per template and device class
Template × revenue weighting to order the work
Full third-party script inventory with per-script cost
The lab-versus-field gap, which itself points at causes
Stage 2: Fix at the layer that owns the problem
Edge-layer problems — caching rules, image formats and sizing, compression, delivery — get fixed in configuration and pay off store-wide at once. Page-layer problems — render-blocking scripts, oversized hero media, layout shift from late-loading widgets, app bloat — get fixed in the template. Doing edge work first often halves the problem before any template is touched.
Stage 3: Make it stay fixed
Every future app install and campaign script is a speed regression waiting to land. Keep the per-template baseline live, review script additions against their measured cost, and re-audit the third-party inventory on a schedule — the script that earned its place last year may not still earn it.
How it works
Baseline and rank
Field metrics per template, weighted by revenue, with the script inventory costed.
Edge first, then templates
Configuration-level fixes ship store-wide immediately; template fixes follow in impact order.
Defend the gains
Live baselines, script cost review, and scheduled re-audits keep the store from re-decaying.
What you get
Speed measured on the users who actually pay you
Fixes applied at the layer that owns each problem
Third-party scripts individually accountable for their cost
Regressions caught the week they land, not the quarter after
Handled by these agents
Frequently asked questions
What Core Web Vitals targets should I aim for?
The published good thresholds — LCP under 2.5s, INP under 200ms, CLS under 0.1 — at the 75th percentile of real users, per template. But direction matters more than thresholds: a store moving from terrible to mediocre captures most of the conversion gain.
How much speed comes from just the edge layer?
On stores that have never tuned it, edge work — proper caching, modern image formats, compression — commonly delivers the larger half of the improvement without touching a template. That is why it goes first.
My theme is the problem. Rebuild or patch?
Measure first: if the template fixes list is short and bounded, patch. If the theme’s architecture fights every fix, the rebuild pays back — but that is a decision to make on evidence, not frustration.
Which agents run this?
Ops owns the baselines, edge layer, and regression watch; Builder ships template fixes with the evidence attached. The audit log records what changed and what it did to the metrics.
Keep reading
Site speed optimisation
Find what actually slows real users down, fix it, and stop the regressions that undo it.
Improve mobile conversion
Close the mobile gap: thumb-reach reality, speed on real devices, and checkout friction.
Cloudflare
Edge performance, caching, and bot protection — managed continuously by the Ops agent.
Rank in AI search
A programme for getting retrieved, cited, and recommended by AI answer engines.
Put the team to work on your store
Connect your storefront, analytics, and email stack, set a goal, and let the agents run the work end to end. Start on the free plan with your own model key.