01 · The Situation
Prostuty Online is a Bangladesh-based digital education platform offering paid courses across academic and skills-based curricula. The Bangladesh online education market is large, growing, and brutally competitive. CPMs are low — among the lowest globally — but that's not the advantage it sounds like. Low CPMs mean every competitor is also running cheap traffic, which means quality of attention is what separates a working acquisition system from a money-burning one.
Prostuty had been running Meta Ads with a typical playbook: broad audiences, video-led creative, optimization toward landing page views or registrations. The result was a familiar pattern in the region — high vanity-metric performance (thousands of "leads" per month), but actual paid course enrollment lagging far behind the ad spend.
The owner had the right intuition: the wrong conversion event was being optimized. The algorithm was getting very good at attracting people who would never pay.
02 · The Diagnosis
Three issues surfaced during the audit:
The conversion event was upstream of revenue. Optimizing toward "Registration" or "Sign Up" trained Meta to find people who would register and disappear. The model had no signal for whether those registrations converted to paid course purchases — which was the only event that mattered to the business.
Creative was optimized for engagement, not intent. High-watch-rate videos were running because they performed well on the platform's vanity engagement metrics. But engagement among casual scrollers isn't the same as intent among ready-to-buy students.
The funnel wasn't structured around the actual buyer journey. Students don't see an ad and buy a course in one session. They see, they research, they consider, they return. The campaign structure didn't reflect that — there were no remarketing layers, no consideration-stage creative, no pricing-aware messaging for the lower funnel.
03 · What We Did
The first move was the most important: we pushed the optimization event downstream to "Course Purchase." Meta needed time to learn — the first two weeks of the rebuild looked worse on every dashboard metric, because the algorithm was rebuilding its understanding of who actually pays. By week three, the right people started showing up.
We restructured the funnel into four stages with creative tailored to each:
- Awareness — broad cold audiences, problem-aware creative ("Studying for [X] and feeling stuck?")
- Engagement — retargeting video viewers with course-specific value propositions
- Trial — retargeting course page visitors with pricing-aware messaging and instructor credentials
- Purchase — retargeting cart abandoners with course-completion testimonials and limited-time enrollment windows
Server-side conversion tracking via the Meta Conversions API closed the iOS attribution gap and ensured purchase events flowed back to optimization regardless of browser tracking. Offline conversion imports surfaced cases where students paid through alternative channels after seeing ads.
By month three, the system was producing roughly 150 paid course purchases per month at a $1.10 average CPA — and importantly, the quality of those customers showed up in retention and course completion rates, not just initial purchase.
04 · The Results
Steady-state performance:
- ~150 paid course purchases per month — measured at the actual revenue event, not signups
- $1.10 average cost per paid course purchase — in a market where competitors quote cheap registration CPAs that don't translate to revenue
- 4-stage funnel architecture producing predictable cohort behavior
- Course completion rates among ad-acquired students now match organic-acquired students — confirming the system is finding the right buyers, not just any buyers
The shift in framing matters: in low-CPM markets, anyone can show "cheap conversions." The work is making sure the conversions are the ones that pay.
05 · What's Next
Two expansions in scope. First, lifetime-value bidding — feeding back data on students who purchase multiple courses, so the algorithm bids more aggressively for high-LTV cohorts. Second, lookalike modeling based on completed-and-paid customers rather than signups — which means the system gets better at finding the next paying student each month.
The CPA is already strong. The next layer is making the customers worth more.
“Most education ad accounts in our market chase cheap clicks. Zenos built us a system that chases paying students. The CPA is the byproduct, not the goal.”