All creators together brought about 1,941 extra trials (likely 1,809 to 2,080), 31% of all trials in this period. That total is the most solid number in this report. Splitting it per creator is harder, so below some creators get a clear Renew or Stop and others get "test again" or a group result.
If you book them again, 3 of 6 creators beat your $10 target per extra trial, even at the high end of their range. 2 cost more than your target, even at the low end; together they took $3,600 of this round's budget. Moving that money to the creators marked Renew is the clearest win. @milesandmore: too close to call with this data, so no advice yet; the plan below says what makes the next round decisive.
| Creator | Extra trials this round (likely range) | Next video: extra trials | Cost per extra trial | Versus target | Advice |
|---|---|---|---|---|---|
| @jogjournal3 videos · $900 spend | 736659 to 812 | 284158 to 516at ~196k views, $300 | $1.06$0.58 to $1.89 | Renew | |
| @runwithmaya2 videos · $1,200 spend | 560498 to 621 | 323159 to 696at ~180k views, $600 | $1.86$0.86 to $3.77 | Renew | |
| @coachdre2 videos · $1,800 spend | 338280 to 397 | 20695 to 419at ~189k views, $900 | $4.36$2.15 to $9.45 | Renew | |
| @milesandmore2 videos · $1,500 spend | 150105 to 201 | 9140 to 202at ~130k views, $750 | $8.27$3.72 to $18.84 | Test again | |
| @thepacelab2 videos · $2,400 spend | 14497 to 195 | 5923 to 139at ~246k views, $1,200 | $20.33$8.61 to $51.54 | Stop | |
| @fitfam.lena3 videos · $1,200 spend | 130 to 77 | 50 to 32at ~131k views, $400 | $87.76$12.43 to ∞ | Stop |
The advice is about booking one more video: the creator's effect per 1,000 views, times the views their videos can be expected to get (with that uncertainty in the range), at their usual price. One viral video does not make a creator cheap. Extra trials include everyone the model attributes to a creator's views, also people who never clicked a link or used a code. The green line in each bar is your target.
Write these down. By chance about 1 in 7 lands outside its range. If most of them do, this audit was wrong for your app.
Baseline. Daily trials are modelled as a weekday pattern plus a trend. That is what you would have had without creators.
Creator effect. Each creator's views are spread over the following days (carry-over 0.45: the share of a day's effect that is still there the next day). The model estimates extra trials per 1,000 views per creator.
Uncertainty. The ranges come from a block bootstrap of the daily residuals, which also re-estimates the carry-over. In tests on simulated apps with known effects, the true value fell inside the range about 85% of the time. The price of that caution: depending on the app, a quarter to over half of creators get "test again" or a group result instead of a call. Renew means we are at least 90% sure one more video costs less than your target per extra user; Stop means we are at least 90% sure it costs more. In those tests about 97 in 100 calls were right; for creators who are very alike and close to your target, about 83 in 100.
Built to say "not sure". We tried to fool the model with six tricky cases: one creator warming up viewers for another, press caused by a creator, trials pulled forward, late tracking, saturation. It gave no wrong call; it reported a group or asked a question instead.
Limits. This is observational data, not a randomised test. The model controls for the launches, ads and price changes you listed; anything you did not list can still look like creator effect. Creators who post on the same days are not split: you get their joint effect, marked "Can't separate".