Trial cost down 37%: the post-mortem

Q: What was actually broken when this account arrived?

Nothing dramatic, which is the dangerous kind. The ads were tasteful, the targeting sane, the budget stable. The problem was stillness: the same four creatives had run for five months, and auction prices for tired ads only move one direction. Cost per trial had climbed 22% across those months without a single alarming week. Fatigue doesn't knock; it leans.

Q: What did the rotation discipline look like in practice?

A fixed Tuesday ritual, eight weeks running: read the bench, retire the weakest live variant, promote the strongest challenger, ship one new hook family. Never more than one variable moving per slot, or the reads turn to soup. The cadence matters more than any single ad; the account's trial cost fell in six of the eight weeks, and neither of the two flat weeks justified breaking the ritual.

Q: Tell the story of the hook the client hated.

The winning family showed a phone face-down on a nightstand, untouched all night, with the sleep score assembling in the morning. The founders' instinct was to show the app's beautiful graphs; the hook showed the absence of the phone instead, because the product's honest promise isignore your screen, we'll still know. It tested 41% cheaper per trial than the graph-forward control. The client approved it on data and said, on the record, that they still hated it. We've framed the email.

Q: How did you keep measurement honest on iOS?

By deciding, before scaling, what the SKAdNetwork conversion values would mean: trial starts in the first postback window, coarse values mapped to onboarding depth. With opt-in around 35%, deterministic iOS data is the minority report, so we treated the MMP's modeled numbers as directionally true and cross-checked monthly against actual subscription revenue, which has no opinion about attribution. The two never disagreed by more than we could tolerate, and when Android and iOS trends agreed, we trusted the direction fully.

Q: What died that deserved better?

A gorgeous UGC-style testimonial family: real-feeling, warm, expensive to produce, and it recruited people who installed, admired the onboarding, and never started a trial. Cost per install: excellent. Cost per trial: worst on the bench. It taught the account's central lesson, that install metrics flatter creatives which sell the idea of sleep rather than the habit of tracking it. Its headstone is in theGraveyard next to concepts far uglier.