A media buyer has checked a rule that would reduce a campaign budget when its performance falls below the team's agreed criteria. The condition looks sensible, but live operation adds another question: would the team approve that change at the moment the rule triggers, with the data available then?
Quick take: Start an ad automation rule rollout with notifications, compare each proposed action with a recorded human decision, then enable the action only for a defined set of supported objects. Keep unresolved measurement or judgment gaps in observation. Before activation, assign responsibility for disabling the rule and restoring affected settings.
This is a recommended acceptance process for rules that have already passed initial validation. Use the broader ad automation rule validation guide for the checks that precede rollout.

Observe condition matches before enabling changes
The first stage should reveal whether the rule identifies situations worth acting on without giving that rule authority to change delivery. Configure Notify as the observation action and retain the intended budget, bid or status change in your review record. Check that the observation configuration contains no action that changes campaign delivery.
XMP's Smart Assistant capability overview documents Notify across channels and object levels, while other actions depend on the channel and level. That supports a notification-first approach. It does not establish a built-in simulation mode, approval queue or automatic comparison against human judgment; maintain the acceptance record separately where necessary.
Keep the rule's scope, metric definitions, comparison window and evaluation schedule stable during observation. Record a new version whenever you change them. Otherwise, a review can accidentally combine evidence from several different rules.
For each notification, capture the object ID, evaluation time and timezone, metric values available at that time, reporting window, and intended action. Preserve enough evidence to reconstruct the decision later. A refreshed report may contain revenue or conversions that were unavailable when the notification arrived.
Also inspect selected objects that did not produce a notification. Ask whether the rule correctly left them alone or missed a situation that should have matched. Repeated alerts from one unchanged campaign provide less coverage than observations across the operating situations the rule will actually face. A quiet observation period alone cannot establish readiness.
Compare the proposed action with a human decision
A campaign automation acceptance review should answer whether the proposed change fits the business situation. Have the reviewer record what they would do using the same information available at the evaluation time, then compare that decision with the intended rule action.
Separate three findings: agreement, disagreement, and insufficient evidence. For a disagreement, identify whether the problem lies in the data, the condition, the action size, the timing or a business exception. For insufficient evidence, record what must become available before a decision is possible.
Use one row per reviewed event in a shared record. The following fields make that record usable by someone who did not receive the original notification.
| Field | What to record |
|---|---|
| Rule and scope | Rule version, channel, account, object ID and object level. |
| Decision-time evidence | Timestamp, timezone, metric source, window, values and known reporting delays. |
| Proposed change | Current setting and intended resulting setting, including applicable limits. |
| Human judgment | Act, hold or investigate, with the business reason and reviewer. |
| Comparison | Agreement, disagreement or insufficient evidence; explain the difference. |
| Disposition | Keep observing, revise and recheck, or accept for a stated limited scope. |
Illustrative example: a budget reduction alert
Suppose a rule would reduce the budget of an app campaign when spend and ROAS meet the team's predefined conditions. The notification arrives while revenue reporting is incomplete. The buyer records “hold” because the available return does not yet support a budget decision.
That disagreement identifies a timing problem even if the numerical condition matched correctly. The next step is to revisit the reporting window or eligibility conditions, then observe the revised rule. If the required maturity check cannot be expressed reliably in the available configuration, retain notification and manual review for that situation.
Conversely, if comparable mature data supports the reduction and the reviewer agrees with its size, the event supports acceptance for that case. It does not prove that the same action is appropriate across every account. This example describes a review method and contains no observed campaign results.
Set acceptance criteria before reviewing outcomes: relevant situations covered, unexplained disagreements resolved, limits checked, and reviewers able to handle the expected workload. Choose observation duration and sample sufficiency around the rule's frequency and consequences. There is no universal number of alerts or days that makes a rule ready.
Enable one supported action within an explicit boundary
Limit initial execution to objects whose configuration and economics the team understands. Define the selected accounts and object IDs, the permitted action, its limits, the activation period and the review owner. A new channel, object level or campaign type should receive its own support check before joining that scope.
The February 2026 automation overview provides a useful capability baseline. Confirm the exact configuration in the account before activation, especially when moving from notification to an action that changes spend or delivery.
| Configuration | Documented boundary | Rollout check |
|---|---|---|
| Pause or start | Most channels at campaign, ad set and ad levels. | Verify the selected channel and level support the action. |
| Budget or bid change | Campaign or ad set level; platform-dependent. | Check the editable field, adjustment method and resulting value. |
| Target ROAS | Network-specific support; Meta, Google and TikTok use bid-change logic. | Confirm what field changes for the exact campaign configuration. |
| Selection and conditions | Precise or fuzzy selection; AND/OR logic; no NOT logic in the overview. | Inspect actual selected objects and how exceptions are excluded. |
| Repetition controls | Cooldowns, bid and budget limits, and daily budget reset. | Verify configured boundaries and repeated-action behavior. |
| Evaluation frequency | Standard minimum interval: 15 minutes; enterprise configurations: 5 minutes. | Confirm available cadence; it does not establish data freshness or instant execution. |
Newer editing capabilities need careful interpretation. The October entry in the 2026 XMP release notes documents Moloco campaign status, budget and target editing, plus status editing at AdGroup and Creative levels. Editing support alone does not establish that a Smart Assistant rule can execute every one of those changes.
Before switching to execution, compare the saved rule with the accepted version. Carry forward the evidence from scope testing and the cooldown and repeated-execution worksheet. Check overlapping rules and manual workflows using the rule conflict review, rather than assuming this rule is the only process changing the selected objects.
After activation, reconcile the intended change with the actual setting on the ad platform. Record attempted, confirmed, failed and uncertain outcomes separately. A condition match or notification does not prove that a budget change succeeded. If the result is uncertain, inspect the current state before retrying an adjustment that could compound.
Define when to disable the rule and who restores settings
The team needs a clear response when execution differs from acceptance. Write down stop conditions before activation: an unintended object changes, the resulting setting differs from the approved action, input data becomes unreliable, another process creates a conflict, or execution status cannot be reconciled.
Assign a person who can disable the rule, a backup, and an owner who can inspect and correct affected campaign settings. Record the relevant pre-activation values and configuration evidence. Keep the location of that record in the handoff so recovery does not depend on the original operator being available.

When a stop condition occurs, disable further delivery-changing rule execution and check whether any changes are still pending or have already completed. Compare affected objects with the saved settings and any legitimate intervening edits. The recovery owner should decide which values to restore, apply supported corrections, and verify the resulting state.
If inputs become unreliable, confirm that delivery-changing execution is disabled before returning to notification-only review. Investigate the data issue while automatic changes remain disabled, and repeat the acceptance review before re-enabling them.
The documented daily budget reset restores an original budget the next day in its supported context. It is not evidence of a general one-click rollback for bids, statuses, deletions or all campaign settings. Restoring a setting also cannot reverse spend already incurred.
Return a corrected rule to notification observation when the fix changes its conditions, scope or action behavior. Preserve the incident and revised acceptance record so the reason for the change remains clear.
Choose the next rollout step from the evidence
At the review, make the decision in this order:
- Validate measurement. Confirm that the rule used the intended source, reporting window and timezone. Resolve incomplete or unexplained inputs before expanding execution.
- Check maturity and coverage. Confirm that business outcomes have had time to develop and that the review includes relevant matched and unmatched situations. Keep sparse evidence provisional.
- Connect cost with business value. Review whether the action remains justified by retention, payer quality or revenue under consistent definitions. Lower spend alone does not demonstrate a better acquisition outcome.
- Choose one next action. Continue observation if evidence is incomplete; revise if judgment differs; retain the limited scope if execution is correct but coverage is narrow; expand only to another explicitly reviewed scope. If an active rule's inputs become unreliable, disable delivery-changing execution before returning to observation.
Record the decision, owner and next review point beside the rule version. Notification review establishes whether the proposed decisions are acceptable; limited execution establishes whether those decisions are carried out as intended. Together, they give the team a concrete basis for the next ad automation rule rollout.