# Reading Optimizer Decisions (/docs/paid-media/optimizer-decisions)



Every optimizer action is logged and visible. This page shows where to
find decisions and how to interpret them.

## Where decisions show up [#where-decisions-show-up]

| Surface                                           | What it shows                           |
| ------------------------------------------------- | --------------------------------------- |
| Ad detail panel → **Optimizer** tab               | Every action on this ad.                |
| Ad set / campaign → **Optimizer activity** widget | Recent actions in this scope.           |
| Notifications inbox                               | Material decisions (pauses, scale-ups). |
| Audit log (Org-level)                             | All actions across all layers.          |

## Anatomy of a decision [#anatomy-of-a-decision]

```
2026-04-19 06:00 UTC
Action: PAUSE
Subject: Ad "Hook variant 3"
Inputs:
  - 7d spend:        $32.00
  - 7d conversions:  1
  - CPA:             $32.00 (target: $12.00)
  - Frequency:       4.2
  - Age:             12 days
State: FATIGUED
Reason: CPA exceeds target by 167% over a 7d window with high frequency.
Guardrails passed: age >= 3d, spend >= $5, not TOP_PERFORMER, not last in adset.
Replacement: Pinned "Hook variant 7" from creative pool (organic_score 6.4).
```

Every field is meaningful. Read all of them before pushing back.

## Common decisions and how to read them [#common-decisions-and-how-to-read-them]

### Pause + replace [#pause--replace]

The most common action. The optimizer paused an ad and brought a new
one in from the eligible pool. Usually correct. Look at the new ad's
performance over the next 3 days.

### Pause without replace [#pause-without-replace]

The optimizer paused but couldn't find a replacement (eligible pool
empty, last in ad set, etc.). The ad set runs short. Investigate:

* Is your eligible pool too small? Generate more, approve more, lower
  the score threshold via `override = 'include'`.
* Is the ad set's audience too narrow? Re-target.

### Scale up [#scale-up]

The optimizer raised an ad set's budget. Budget adjustments are capped
at 50% of the current value and $100 absolute in a single step. Watch
for spend efficiency over the next 3 days.

### Hold (no action) [#hold-no-action]

The optimizer evaluated but did nothing. Often you'll see "Hold —
within performance tolerance" or "Hold — guardrail (only 2 days old)".
Holds are healthy. Frequent holds = stable system.

### Refresh ad set creative pool [#refresh-ad-set-creative-pool]

Triggered nightly. Disables underperformers, brings in new from pool.
Same as pause+replace at scale.

## Disagreeing with the optimizer [#disagreeing-with-the-optimizer]

Two ways to push back:

1. **Manual override** — pause / activate ads directly. The optimizer
   respects user decisions and won't immediately undo them.
2. **`override` column** — set `'include'` to keep an ad eligible
   regardless of score, `'exclude'` to keep it out.

If you find yourself overriding repeatedly, file feedback with support —
either the optimizer logic needs tuning, or your config (CPA target,
fatigue thresholds) needs adjustment.

## Pause the optimizer [#pause-the-optimizer]

Layer settings → **Pause optimizer**. The optimizer continues collecting
metrics but takes no actions until re-enabled.