

01 · The Situation
This APAC education provider sells professional certification courses — project management, data analytics, digital marketing — to working adults in Singapore and Malaysia. Meta Ads budget: ~$8,500/month across Facebook and Instagram.
The growth team reported strong top-funnel metrics: ~2,000 "leads" monthly at low CPL. Finance reported ~80 paid course purchases — a conversion rate from lead to purchase that made the channel look broken.
The pattern matched our Bangladesh education platform diagnosis: optimization toward registration, not purchase. APAC CPMs are higher than South Asia — wasting spend on unqualified leads hurts faster.
At audit, Meta attribution trust aligned with our other-industry benchmark: ~38% base attribution loss without CAPI + deduplication. Event match quality scored 4.2/10.
02 · The Diagnosis
Wrong optimization event. Campaigns optimized for CompleteRegistration. The algorithm found people who sign up for free previews — not $299–$899 course purchases.
Browser Pixel only. No Conversions API. iOS privacy removed an estimated 35–42% of purchase events from reporting — consistent with our tracking trust benchmarks for education-adjacent verticals.
No deduplication. Pixel and manual offline uploads double-counted some events; others missed entirely.
Creative funnel mismatch. Top-of-funnel video ads drove to homepage; purchase path required four clicks on mobile. Landing page CR ~1.9% vs 5.0% landing-page benchmark for comparable offer pages.
Audience overlap. SG and MY campaigns competed for the same broad interest stacks — inflating frequency without incremental purchases.
03 · What We Did
CAPI implementation (weeks 1–3). Server-side purchase events with hashed email/phone. event_id deduplication with browser Pixel. Offline import for phone sales from enrollment advisors.
Optimization switch (week 4). New campaign structure optimized exclusively for Purchase — registrations tracked secondary only. Learning phase accepted; CPL rose temporarily, CPA to purchase fell.
Funnel rebuild (weeks 3–6). Dedicated landing pages per certification track. Mobile-first checkout. Social proof above fold — graduate outcomes, not feature lists.
Market separation. SG and MY campaigns — separate creative (currency, testimonials), separate budgets, shared event taxonomy.
Creative testing velocity. 4–6 new hooks per month — offer-led, outcome-led, instructor-led — killed under 1.2× CPA/purchase benchmark after 1,000 impressions.
04 · The Results
At 90 days on purchase optimization:
- Cost per course purchase $19.20 → $11.40 (−41%)
- Purchase ROAS 3.1× → 5.8× — above 5.5× other-industry top-quartile Meta benchmark band
- Monthly purchases 80 → 169 (+112%) on same budget
- CAPI event match quality 4.2 → 7.8
- Registration volume down 38% — intentional; purchase volume doubled
Finance now models Meta on contribution margin per enrollment — not lead volume.
05 · What's Next
Google Ads Search layer for high-intent "certification course Singapore" queries — capturing demand Meta creates at lower incremental CPA.
WhatsApp enrollment advisor integration for MY market — click-to-chat as secondary conversion signal feeding CAPI.
“We had the same registration-vs-purchase problem in a different timezone — cheap leads, expensive truth. We needed paying students, not registration vanity. CAPI made Meta finally see them.”


