The metric the platform optimises is not the metric you sell with
Ranked by cost per conversation, the fear ad was our best performer. Ranked by actual buyers, it was the most expensive lead source in the account — by 12×.
A click-to-WhatsApp funnel for a travel-insurance product. The ad platform optimises toward “conversation started” — which sounds like a lead and is actually a click that opens a chat window. That one-word gap between sounds like and is turned the account’s entire ranking upside down.
The inversion
Ranked by the platform’s metric, the account’s story was clear. The fear-based angle (visa rejection, trip losses) delivered conversations at ₹36 — the best in the account, and the platform’s budget system had rewarded it with the large majority of spend. The reassurance-based angle cost ₹69 per conversation — apparently the worst.
Then I ranked the same ad sets by our own lead log — where every conversation is graded on replies, engagement, and movement toward payment.
The fear angle: 21% of its conversations were hot, averaging under three replies — roughly ₹203 per hot lead. The reassurance angle: 67% hot, averaging seven replies — roughly ₹17 per hot lead. The “worst” ad set produced qualified buyers at about a twelfth of the cost of the “best” one. (The reassurance sample was small — I’d call it directional, not proven. The direction, however, was not subtle.)
The mechanism, and the compounding
Fear is cheap at the top of the funnel because frightened people click to be reassured, not to buy. Reassurance attracts fewer, warmer people who arrive already leaning in.
What makes this dangerous rather than merely interesting: the budget system steers by the cheap metric. It sees fear winning and feeds it, so the account drifts automatically toward its own worst buyers. Over a fortnight I watched cost per hot lead climb from ₹147 to ₹285 while the platform’s cost per conversation sat serenely flat at ₹32–35. The dashboard said nothing had changed. The pipeline said everything had.
The framework
- Keep your own ledger of lead quality downstream of the click, joined to ad source by tag. Without it, the platform’s metric is the only voice in the room.
- Re-rank every ad set by cost per qualified lead before believing cost per result. Expect the rankings to disagree; the disagreement is the information.
- Treat divergence between the two rankings as the primary alarm — flat platform costs while downstream quality decays means the optimiser is winning its game and losing yours.
One caveat earned the hard way: before reading a low-spend ad set as “starved by the algorithm”, check its creation date. One of my “suppressed” ad sets turned out to be a day old. The platform makes enough real mistakes without being charged for imaginary ones.
The platform is not lying. It is optimising, precisely and tirelessly, toward the thing you told it to count. The question is whether that’s the thing you get paid for.