The Attribution Black Hole: $41K in Google Ads With No Way to Tell What Worked
Metrics Comparison
Timeline
30 days
No server-side conversion tracking; Google Ads conversion tag firing on page view instead of actual wallet creation; cross-channel deduplication absent; deep link attribution broken on Android
Implemented server-side conversion import via Google Ads API; fixed deep link attribution with AppsFlyer; built unified attribution dashboard mapping ad click to wallet creation
Attribution accuracy improved from 38% to 94%; identified YouTube as 3.2x more efficient than Search; reallocated budget for 41% CPA reduction (21 days)
The Situation
A non-custodial crypto wallet provider was running $41,000/month across two Google channels:
- Google Search: $25,000/month targeting keywords like "best crypto wallet," "secure bitcoin wallet," "defi wallet download"
- YouTube: $16,000/month with pre-roll ads targeting crypto content viewers
Google Ads reported 340 total conversions for the month. The backend database showed 890 new wallet creations attributed to paid channels (via UTM parameters). The 550-conversion discrepancy meant the team was making budget decisions based on data that captured only 38% of actual results.
What Went Wrong
The tracking infrastructure had four compounding failures:
Failure 1: Wrong Conversion Event
The Google Ads conversion tag was firing on the "Download Page" view — not on actual wallet creation. Users who visited the download page but never installed the app were counted as conversions. Simultaneously, users who installed via direct app store links (bypassing the download page) were not counted at all.
Failure 2: Deep Link Attribution Broken
The Android deep link configuration was malformed. When a user clicked a Google ad, arrived at the download page, then opened the Play Store and installed the app, the attribution chain broke. The app could not trace the install back to the original ad click. This affected approximately 60% of Android conversions.
Failure 3: Cross-Channel Cannibalization
Users often saw a YouTube ad, then searched for the wallet name on Google and clicked a branded Search ad. Google attributed the conversion to Search (last-click), but the actual awareness driver was YouTube. Without multi-touch attribution, YouTube appeared to have a $96 CPA while Search showed $52 — when the true relationship was reversed.
Failure 4: Cookie Consent Impact
The wallet's website used a cookie consent banner that 44% of users dismissed without accepting. The Google Ads tag only fired for users who accepted cookies, creating a systematic undercount of all web-based conversions.
Diagnosis
RedClaw's tracking audit compared three data sources for the same 30-day period:
| Source | Conversions | CPA | |--------|------------|-----| | Google Ads Dashboard | 340 | $120 | | Backend (UTM-attributed) | 890 | $46 | | Backend (all new wallets) | 1,420 | $29 |
The gap between Google-reported (340) and backend-UTM (890) represented tracking failures. The gap between UTM-attributed (890) and total new wallets (1,420) represented organic and unmeasurable touchpoints.
The Fix
We rebuilt attribution from the conversion event up:
-
Server-side conversion import: Replaced the browser-based conversion tag with server-side conversion uploads via Google Ads API. When a wallet is created, the backend sends the conversion (with GCLID) directly to Google — bypassing cookie consent and browser restrictions entirely.
-
AppsFlyer integration: Deployed AppsFlyer for mobile attribution with:
- Properly configured deep links for Android and iOS
- Fingerprint matching as fallback when deep links break
- Deferred deep linking for users who install via app store browse
-
Cross-channel attribution model: Built a data-driven attribution model in Google Analytics 4 that distributed credit across touchpoints. This revealed YouTube's true contribution: 62% of converting users had a YouTube touchpoint in their path.
-
Consent-mode implementation: Replaced the blocking cookie banner with Google Consent Mode v2, which provides modeled conversions for users who decline cookies.
Results
Within 21 days of implementation:
- Attribution accuracy: 38% to 94% (measured by server-reported vs backend variance)
- Identified true channel efficiency:
- YouTube actual CPA: $31 (previously reported as $96)
- Search actual CPA: $58 (previously reported as $52)
- Budget reallocation: Shifted from 60/40 Search/YouTube to 35/65 Search/YouTube
- Blended CPA: Dropped from $68 to $40 (41% reduction) purely from allocation improvement
- ROAS: Improved from 1.1 to 3.1
The critical insight: the team had been underinvesting in their best channel (YouTube) and overinvesting in their mediocre channel (Search) because broken tracking inverted the performance signals. Fixing tracking did not change the ads — it changed the decisions.