How SpendDaddy learns from outcomes
SpendDaddy remembers what happened to each campaign and keyword it changed, and what you dismissed. It uses that to stop repeating proposals that did not help. The rules are fixed and visible; nothing is hidden or trained on your data. You can see the result on the Memory tab of a campaign or placement.
Outcomes
Each applied change is saved with a hypothesis, the result it should produce. After an observation window it gets one of these outcomes:
- Worked: earnings after ad cost rose by more than a small tolerance.
- No measurable effect: the change is within that tolerance.
- Backfired: earnings after ad cost fell by more than the tolerance.
The window is 14 days after a raise and 7 days after a cut. The measure is daily earnings after ad cost, before and after the change.
Two things to keep in mind:
- The first outcome can be provisional. It estimates 30-day revenue from the first 14 days, using your app's earlier ratio between the two. It is labelled that way. Once 30 days of revenue are in, the outcome uses the observed 30-day revenue instead.
- An outcome is observed, not proven. It does not adjust for weekdays or seasonality, and it cannot see edits made outside SpendDaddy. Outcomes that cannot be judged, for example because revenue data is incomplete, are not listed as verdicts.
Dismissals
When you dismiss a recommendation you can give a reason. The reason is optional.
| Reason | What happens |
|---|---|
| No reason | The recommendation stays hidden for 14 days. |
| Wrong intent | The same recommendation for that campaign or keyword stays hidden for 90 days. |
| Already handled | Hidden for 30 days. |
| Too aggressive or Disagree on economics | Hidden for 14 days, and counted toward the rule family below. |
Each hidden recommendation shows up under Held back on the Memory tab, with the date it can return.
If you dismiss three or more recommendations of the same type in an app within 30 days as Too aggressive or Disagree on economics, SpendDaddy ranks that type lower in your list. It is still proposed; it just no longer comes first.
After a raise backfires
If a bid or budget raise backfires, is reverted, or trips its guard, SpendDaddy pauses raises on that keyword or campaign. Raises resume once the keyword or campaign has spent 30 more in your account currency (the default) since the raise, so the next decision rests on fresh data. The Memory tab shows this as Raises paused.
After raises show no effect
After two raises in a row with no measurable effect, SpendDaddy stops proposing raises there. It proposes a look at coverage, search term matching, or the product page instead, because the bid or budget is probably not the constraint. The Memory tab shows Raising stopped.
Step sizes
The standard raise step comes from your targets. SpendDaddy adjusts it per keyword or campaign:
- After a raise that worked, the next step grows by half, up to twice the standard step. It does not grow past the agent change limit in your targets.
- After a raise with no effect, the next step is halved, down to half the standard step.
- After a backfire, the next step is half the standard step.
An adjusted step is explained on the recommendation, for example "Step size is 10% instead of the standard 20%, because the last raise had no measurable effect." The Memory tab shows Step adjusted.
These rules also apply to the autopilot, which skips raises that they would hold back.