Does conversion tracking match reality?

Findings and recommendations

Keep the working primary import. Fix the interpretation and QA separation. September 23 has five Ads-linked GA4 phone events across four sessions, four imported primary conversions and five parallel secondary signals. Never add these together as callers.

Daily evidence, Budapest time

Paid sessions use a linked Ads campaign ID. Phone events are interactions, not calls. All counts are the September 25 early-morning snapshot.

Sep21–24: 29 phone events = 23 known QA + 5 Ads-linked + 1 unverified agency referral. The emails support at least five calls, with incomplete day coverage. The earlier email covers Sep18–21, the later one Sep22–24. They are separate periods, not cumulative totals.

Timing: Sep21's two QA phone events occurred after the email cutoff. Sep24's hourly Ads data cannot determine which 08:00-hour clicks preceded 08:44. September 25 is before the current serving start and remains incomplete. Recent GA4 data may still mature.

Why five events become four primary conversions

Generic, Érd and Budakeszi each show 1 GA4 event, 1 primary and 1 secondary signal. Csepel shows 2 GA4 events, 1 primary and 2 secondary signals. GA4 counts every event, while the import counts One per ad interaction. The difference is consistent with that rule, but a per-click ledger is unavailable. Both Ads date views agree. The secondary action also uses One, so its extra signal needs identity-level evidence before claiming a duplicate-tag defect.

Five checkpoints

• Ad clicks: proven. Traffic exists.
• Measured paid visits: partial. Clicks and sessions differ. Consent, blockers, early exits and attribution loss remain alternative explanations.
• Phone events and primary import: observed working. The old zero resolved after the partial snapshot.
• Actual calls: not proven by tracking. Reveal copies the number without dialling. A call button still does not confirm a connected call.
• Qualified work and revenue: unknown. The emails are not a paid-job ledger.

Exact proposed settings and process

1. P1 · Rename, keep the primary: “Phone click proxy (GA4) - Secondary” → “Phone intent (GA4) - Primary”. Keep Primary, One, 90-day window and the same event. Keep the direct Ads action Secondary. Risk if unchanged: reports imply calls. Verify same action and role after renaming, revert the label if needed.
2. P1 · Separate reveal from dial: add phone_action=reveal or phone_action=call_intent to the existing tel_szam_katt event and register that custom dimension. One event per interaction, no extra duplicate trigger. Verify both tagged QA paths without placing a call. Roll back tag/site versions if event counts or dial behavior regress.
3. P1 · Exclude explicit QA: current Internal Traffic filter is Testing. Add dedicated QA marker traffic_type=qa and QA exclusion, Testing first, Active after verifying only QA matches. Suppress direct Ads conversion tags for the same QA flag. Leave the existing internal rule unchanged. Incorrect exclusion can lose real data, so validate controls and retain the baseline. Historical data still needs report exclusions.
4. P1 · Record outcomes: one call-log row with timestamp, answered/missed, new/repeat, service need, qualified status and actual job outcome. Source stays Unknown unless supported. Reconcile against call history. This addresses business truth without pretending emails can be uploaded as attributed conversions.
5. P2 · Compare settled data: repeat conversion-time action counts after 48 hours, filtered GA4 events and the dated call ledger. Preserve recent-day caveats. Investigate persistent same-definition gaps, not raw total inequality.
6. Keep consent and bidding: denied-default behavior, auto-tagging, link configuration and Manual CPC remain unchanged. Manual CPC does not optimize bids for conversions. No evidence supports an import rebuild.
7. Optional future actual-call measurement: keep the current forwarding-number action Secondary until a legitimate paid call, number route, provider receipt and business outcome are validated. Its two-second threshold is not a qualified lead. Only then consider replacing the proxy as Primary, never count both as separate leads.

Separate serving request: surrounding correspondence requests weekdays 06:00–22:00 and a 50 km radius centered on central Budapest. Current readback remains 07:45–19:00 and the Budakeszi center. Flag for a separate decision within the budget, not as a tracking repair. No changes made.

Evidence strength and limits

High confidence: current import receipts, QA counts, action settings and original email wording. Unknown: unique callers, answered status, exact event-to-call matches, capture completeness and paid jobs. API reports show no sampling or threshold warnings. Both systems use Budapest time. Raw emails, private IDs and telemetry stay in the local evidence package.

Google: Analytics imports · Google: conversion columns and dates · Google: data freshness

Progress and limits

All five investigation criteria verified. Measurement still has stated business-outcome limits. Recommendations have not been applied. No calls, campaign edits or client messages were made.