Crestfield Capital
Investment services · Conversion tracking rebuild
Crestfield Capital is a Cyprus-based investment services firm running Google Ads to drive qualified booking completions — consultations, account opens, and service inquiries — from high-intent audiences in target markets.
At audit, four conversion actions were active — all set to primary. The bidding strategy saw no distinction between a completed booking form and a user tapping the phone number. 75% of reported conversions were low-value local interactions. Campaigns appeared to be performing. The actual booking pipeline was not.
The four conversion actions active before the rebuild — and what they actually measured:
| ACTION | TYPE | STATUS |
|---|---|---|
| Booking form completed | Real intent signal | ✓ Keep — set as sole primary |
| Phone number click | Low-value interaction | → Secondary only |
| Get directions tap | Local action — not a client | → Secondary only |
| Service page view (3s+) | Engagement proxy | → Remove from bidding |
The fix was not a creative refresh or bid adjustment. It was a structural rebuild of what the account was optimizing toward.
- →Primary conversion isolatedWeb booking completion only
All other actions demoted to secondary — visible in reporting, invisible to bidding algorithm
- →tCPA reconfiguredTarget based on real booking cost
Previous tCPA was set against a blended average of all four actions, making the target meaningless
- →Server-side GA4 deployedRevenue attribution corrected
Browser-based tracking was losing attribution in Safari and cookie-restricted environments — server-side closed the gap
- →Conversion window adjusted30-day booking window
Financial services decisions take longer — extending the attribution window recovered conversions previously credited to last-click only
Smart Bidding doesn't care what you want — it optimizes on what you tell it. When 75% of conversions are low-value local actions, the algorithm actively selects for audiences that trigger those actions. Clicks, direction requests, phone taps: all cheap, none of them bookings.
Four conversion actions, all set to primary. The algorithm averaged signals across booking completions, page views, phone taps, and map directions. The result: a bidding strategy simultaneously chasing four incompatible user behaviors — and succeeding at none of them.
Browser-based GA4 in financial services loses meaningful attribution to Safari ITP, consent drop-off, and cookie blockers. Moving to server-side collection recovered sessions that were being counted as organic or direct — and gave the model accurate revenue data to optimize against.
Reported CPA looked acceptable because it averaged across all four conversion types. The actual cost per completed booking — once isolated — was 2.5× higher than the headline figure. The fix wasn't better creative or lower bids: it was removing the noise that masked the real number.
After rebuilding the conversion layer, the account began optimizing on real booking completions for the first time. tCPA dropped from €94+ to €38 per booked consultation. Over the following 12 months, 312 bookings were tracked and attributed correctly — a volume the previous setup could not have measured, let alone driven.
Fix: 4 actions → 1 primary · server-side GA4 · tCPA recalibrated
Result: €38 booking CPA · 312 bookings tracked · 12-month window
Client name is a pseudonym. Report ID is internal. Metrics reflect specific account conditions — results vary by market, budget, and funnel maturity.