Capture the identifier before anything else
- Store the GCLID (and GBRAID/WBRAID for app and iOS traffic) as a hidden field on every form and as a first-party cookie with a lifetime that matches your sales cycle.
- Write the identifier into the CRM record at creation, not at close — a lead that changes owner or gets merged will lose an identifier attached late.
- Capture the same field on phone paths via your call tracking provider, so calls and forms land in one model.
- Layer enhanced conversions with hashed email or phone to recover clicks where the identifier is missing.
Map CRM stages to conversion actions
One static 'lead' event teaches the algorithm that all leads are equal. They are not. Define a small set of stages that reflect real economics — qualified, opportunity, closed-won — and import each as its own conversion action with its own value.
Then decide deliberately which action bidding optimizes toward. Optimizing to closed-won is ideal but requires enough monthly volume to learn; many accounts do better bidding to the qualified stage while still importing closed-won for reporting and periodic value calibration.
Where the sale value varies, send the actual value rather than an average. Value-based bidding on a real distribution behaves very differently from target CPA on a flat assumption.
Legal and healthcare change the requirements
In legal, the cheapest lead is usually the least likely to sign. Intake qualification is the entire game: a firm importing signed matters instead of contact forms will often see reported CPA rise and actual cost per signed matter fall at the same time. That gap is the point.
In healthcare, the relevant outcome is a qualified intake or an approved order, and the compliance boundary is strict. The design constraint is to move an outcome and a value — never clinical detail — from the CRM into the ads platform, with the mapping reviewed by the client's compliance team before anything is sent.
Why Smart Bidding fails without it
Smart Bidding is a prediction engine trained on the conversions you feed it. Feed it form fills and it will find the audiences most likely to submit a form — a population that overlaps only partially with the population likely to buy.
The failure is invisible on the platform dashboard, where volume and CPA both look healthy. It shows up in the CRM as declining lead quality, longer sales cycles, and a growing gap between reported and actual cost per acquisition.
Operate it like a pipeline, not a project
- Audit match rate weekly; treat a five-point drop as an incident, not a metric fluctuation.
- Re-import value adjustments as deals progress so long-cycle revenue is credited to the click that produced it.
- Reconcile imported conversions against CRM totals monthly to catch silent pipeline breaks.
- Keep a documented mapping between CRM stages and conversion actions so a personnel change does not reset the measurement model.
Questions we get
- What match rate should we expect?
- A healthy GCLID-based import typically lands in the 70–90% range depending on form capture quality and CRM discipline. What matters more than the absolute number is stability: a sudden five-point drop usually means a broken form field, not a market change.
- Can we do this without a data warehouse?
- Yes for simple setups — a scheduled CSV or a native CRM integration works. BigQuery becomes necessary when you need multiple stages, multiple business units, or value adjustments over a long sales cycle.
- Does this conflict with HIPAA or attorney-client confidentiality?
- Import sends a click identifier and an outcome value, not clinical or matter detail. In regulated verticals the mapping is designed with the client's compliance team so no protected information leaves the CRM.