Skip to main content
DigeeSell logo

EdTech AI tutor platform · United States and Australia

Optimising to the wrong conversion event cost 15x per signup

15x
Difference in cost per result between two events
$2.81
Cost per result, mid-funnel event
$41.49
Cost per result, deep event
240
Trial starts from paid search

The starting point

The platform wanted paid social optimised to its deepest signup event, on the reasonable logic that optimising for the thing you actually want should produce more of it. It did the opposite. The deeper event fired too rarely for the ad platform to learn from, so the algorithm spent the budget guessing.

Meta — measured

Paid social

Under $6,000 of lifetime media across 364,000 impressions, split across three optimisation events in the same account. The mid-funnel event had enough daily volume to clear the platform's learning threshold. The deeper event never did, and the cost per result reflects that almost exactly.

$2.81
Cost per result — mid-funnel trial eventmeasured
$41.49
Cost per result — deep registration eventmeasured
$16.96
Cost per result — application submittedmeasured
under $6,000
Total media across all threemeasured

Google Ads — measured

Paid search

Search carried the majority of the budget and the majority of the conversions. Performance Max and Demand Gen were run alongside it at smaller weights to test incremental reach rather than to carry volume.

240
Trial startsmeasured
~AED 34
Blended cost per trial startmeasured
6,561
Clicks from 84,000 impressionsmeasured
3 campaign types
Search, Performance Max, Demand Genmeasured

What actually drove it

Feed the algorithm an event it can actually learn from

Ad platforms need a minimum volume of the optimisation event per week before their models stop guessing. Choosing an event below that threshold does not produce fewer, better conversions. It produces far more expensive ones.

Optimise to the event you can feed, not the event you want

The deep registration event was the real business goal. The right structure was to optimise to the mid-funnel event that preceded it and let volume flow downstream, rather than asking the platform to find rare users directly.

Test the assumption rather than inherit it

Running all three events concurrently in the same account, on the same product and audiences, is what made the 15x gap visible and attributable to structure rather than to market conditions.

Match campaign type to job

Search carried volume. Performance Max and Demand Gen were deliberately weighted small and judged on incremental reach, not on matching Search's cost per result.

How these numbers were produced, including where they are estimates

  • Figures are taken from Meta Ads Manager and Google Ads exports for the campaign period.
  • This is a cost-per-signup case, not a revenue case. Payment events across the whole period were in single digits on both platforms, so there is no return on ad spend figure here and none should be inferred. The result being measured is trial signups.
  • Costs are United States and Australia media costs. They are not a benchmark for UAE or Indian media, where auction dynamics and cost per thousand impressions differ substantially. The transferable finding is the ratio between the two events, not the absolute figures.
  • Google campaign-level cost is suppressed in the available export, so no per-campaign cost per acquisition is published for paid search.
  • Client name withheld.
All case studies