What is conversion tracking?

What counts as a conversion

The definition is a business decision, not a technical one, and getting it wrong invalidates everything downstream. A conversion should be an event that reliably indicates value, close enough to the marketing to be measurable and far enough down the funnel to be worth optimising toward.

The common failure is optimising to a shallow event because it is plentiful. Optimising to form fills produces form fills, including worthless ones. If the shallow event is the only one with volume, the honest approach is to optimise to it while monitoring the deeper one, and to re-baseline when the relationship between them drifts.

Client-side and server-side

• Client-side. A tag or SDK fires in the browser or app. Simple to deploy, and subject to blockers, consent gating, and the browser's own restrictions.
• Server-side. Your backend reports the event directly to the platform's API. More reliable, harder to set up, and gives you control over what leaves your systems.
• Both, de-duplicated. The common production pattern: client-side for coverage and richness, server-side for reliability, with a shared event ID so nothing is counted twice.

What routinely breaks it

Four causes account for most broken conversion tracking, and all four are silent — the number simply drops or drifts.

Consent. Where a user declines tracking, client-side events do not fire. This is correct behaviour and it means your tracked total is a subset of your real total.

Redirect and parameter loss. A tracking parameter stripped by a redirect takes the source attribution with it.

Deployment drift. A checkout template ships without the tag. Nothing errors; the number just falls.

Double counting. The same event reported twice through two paths with no shared identifier.

Tracking conversions that arrive without a session

An agentic conversion arrives as an order with no preceding web behaviour — no page, no tag, no session. Client-side tracking sees nothing at all.

The workable approach is server-side, keyed on how the order arrived: which integration, which API credential, which agent identity. That gives you an exact count of agent-driven volume, which is more than click-based tracking ever delivered. What it will not give you is the upstream cause, and for that the instrument remains a holdout.