Recover 40–80% of your Tracking QA & Monitoring ad revenue.

Preview mode tests what you remembered to test, at the one moment you tested it, on the browser and consent state you happened to be using. It's the right tool during a build. It tells you nothing about the Tuesday six weeks later when a plugin update changes the SKU format on one product type.

GA4 has its own gaps, its own modeling and its own outages, so a dip there might just be GA4. And a broken purchase parameter often doesn't change GA4's event count at all — the events arrive, they're just worth nothing. Comparison against the backend is the only check that can't be fooled by the thing it's checking.

Money events stopping or dropping sharply is an incident. Match quality drifting down over weeks is a ticket. We agree the response window in writing during onboarding and hold to it — we don't advertise round-the-clock coverage we don't staff, and you should be suspicious of anyone who does at this price point.

Publish rights in GTM, access to the server container and its hosting, read access to the ad accounts and GA4, and some way to see order or lead counts — an admin export, a read-only database user, or a scheduled report is enough. Read-only wherever the job allows it.

Small fixes are part of the retainer — a trigger that stopped matching, a parameter that vanished, a token that needs rotating. New scope is quoted separately: a new platform, a replatformed checkout, a first-time CAPI build. We'll tell you which bucket something falls in before we start, not after.

No, and it's a common setup. We monitor and reconcile, they build. When a check fails we send them the specifics rather than opening a debate about whose fault it is. We're not looking to take over the implementation, and we'll say so on the call if that's the concern.

If your site rarely changes, you run one platform, spend is modest, and someone in-house actually opens the numbers weekly and knows what normal looks like, a retainer is overkill. Monitoring earns its cost when deploys are frequent, spend is large enough that a week of bad data is expensive, or nobody owns the numbers day to day.