Conversion tracking setup: get it right before a dollar is spent

There's a moment in every marketing engagement where someone asks "so, is it working?" — and the honesty of the answer was determined months earlier, by whether conversion tracking was set up before the spending started. Set up after, you're reconstructing history from fragments. Set up wrong, you're worse off than blind: confidently misinformed.

This is the setup order we use on every project, in plain English.

Step one: decide what a conversion is (harder than it sounds)

A conversion is a visitor doing the thing your business exists for: submitting an enquiry, calling, booking, buying. Write the list down before touching any tool — because every ambiguity here becomes a data dispute later.

Two rules keep the list honest:

• Primary versus secondary. An enquiry is primary. A newsletter signup, a brochure download — secondary at best. Platforms should optimise toward primaries only; blending them teaches Google that a PDF download equals a customer.
• No vanity events. "Time on page", "scrolled 75%", "watched video" — fine as diagnostics, poisonous as conversions. If it doesn't correlate with revenue, it doesn't belong on the list.

Step two: one measurement layer, not four

The classic mess: GA4 tracks the form one way, Google Ads has its own tag doing it another way, Meta's pixel a third, and a developer's leftover code a fourth. Four systems, four answers, weekly arguments about which is "right".

The clean pattern is a single layer — Google Tag Manager — that detects each conversion once, then tells every platform about it. One definition, one place to debug, one source of truth. This is also the foundation that makes server-side tracking straightforward later, instead of a rebuild.

Step three: wire the events that actually matter

• Forms: fire on confirmed submission (the thank-you state), never on button click — clicks include the people the validation rejected.
• Phone calls: tap-to-call tracking on mobile at minimum; call tracking numbers if phone is a primary channel. In many Sydney service businesses, calls are 60% of enquiries and 0% of what's measured.
• Ecommerce: purchase events with real values, deduplicated by transaction ID, so a page refresh doesn't book the revenue twice.
• Offline closes: if deals close by quote or handshake weeks later, feed the outcomes back (even a monthly spreadsheet upload into Google Ads) — otherwise the platform optimises for enquirers who never buy.

Step four: verify before you believe

Every event gets tested three ways before it's trusted: fire it yourself and watch it arrive (Tag Manager's preview mode and GA4's DebugView exist for this); check it arrives once, not twice; and after a fortnight, reconcile the platform's count against reality — actual emails in the inbox, actual calls answered. A 10–15% gap is normal (blockers, consent declines); 40% means something's miswired.

Then leave the definitions alone. Every redefinition breaks your history — if you must change what "enquiry" means, annotate the date and expect a step in the charts.

When to bring someone in

Basic form-and-call tracking on a standard site is genuinely DIY-able with patience. The cases that earn professional help: multiple domains or a separate booking system, ecommerce with refunds and variants, call-heavy businesses, consent banners that must gate tags properly, or an existing setup nobody trusts — where the job starts with an audit of what's really firing, not more wiring on top of the mystery.

Either way, the principle holds across everything we run — sites, SEO, paid media — instrumentation comes first, because it's the only thing that lets you tell the difference between marketing that works and marketing that merely happens.