About 25% cheaper registrations by fixing structure, not creative
Consolidating fragmented ad sets cut cost per registration by roughly a quarter, and server-side custom events kept optimisation alive after Meta restricted health events.
Client and context
ReDial Clinic's main acquisition campaign on Meta, targeting people looking for metabolic health support across Indian cities.
The problem
"We keep testing new creatives, but costs stay high and every ad set is stuck in learning."
Four ad sets split one budget. None reached the conversions Meta needs to exit learning, so all four stayed learning-limited. Later, Meta categorised the clinic's domains as health and wellness and blocked the standard conversion events the campaigns optimised on.
What we built
- 4 ad setsstuck in learning
- 1 campaignthe winning ads only
- CRMqualified, paid
- n8nserver-side events
- Metalearns from real data
- Diagnosed budget fragmentation, not creative, as the real constraint.
- Rebuilt into one budget-optimised campaign, keeping only ads that had produced at least one registration: 15 ads that had driven over 90% of all registrations.
- Cut ads that had spent without converting.
- When standard events were blocked: rebuilt tracking on custom server-side events tied to real CRM milestones (qualified lead, paid assessment), sent from n8n with click IDs, IP and user agent for strong matching.
- A value watcher that back-fills missing conversion values and skipped events, so Meta learns from complete data.
- A separate click-to-WhatsApp campaign with ad-level attribution captured per conversation.
- Policy-safe ad copy and headlines throughout (no personal health attributes).
Stack
- Meta Ads
- Meta Conversions API
- n8n
- GoHighLevel (Synamate)
- Google Tag Manager
- GA4
- Microsoft Clarity
- TailorTalk
Results
- About 25% lower cost per registration in the first 10 days after consolidation, with both new ad sets beating the old baseline.
- Optimisation kept running after Meta's health-category restriction, on custom events with an event match quality around 7 out of 10.
- Every conversion event tied to a real CRM milestone, not a page view.
The workflow, mapped
The real n8n canvas behind this system (28 nodes), redrawn without credentials or customer data. Scroll sideways to follow a lead through it.
- Every 15 minutes
- Settings and stages
- Fetch stage opportunities
- Collect pipeline snapshot
- Read ledger
- Read WhatsApp click rows
- Fetch Synamate users
- Fetch opt-in form submissions
- Opt-ins needing a Lead
- Fetch opt-in attribution
- Decide what to send
- Anything to report?
- Ledger rows to write
- GA4 events to send
- Synamate tag changes
- Meta events to send
- Slack: post run report
- Write ledger rows
- Send event to GA4
- Synamate: add or remove revenue-missing
- Send event to Meta
- GA4 accepted it?
- Meta accepted it?
- Record GA4 send in ledger
- Synamate: tag event as sent
- Build Meta failure alert
- Record Meta send in ledger
- Slack: Meta failure alert
Want this for your clinic?
Request a quote and we will tell you what we would change first.