AI + Paid Media
The Agent Started the Overnight Audit. Morning Arrived Before Anyone Checked Whether It Finished.

Updated: November 20, 2026
A background response can remain queued or in progress after the initiating request returns. Treating acceptance as a finished audit can publish stale recommendations or leave a budget review silently unfinished. Persist the response or session identifier, expected completion window and business deadline. Mark the workflow complete only after a recognized terminal state and a successful retrieval of the final output.
Accepted work is not completed work
A background response can remain queued or in progress after the initiating request returns. Treating acceptance as a finished audit can publish stale recommendations or leave a budget review silently unfinished.
Persist the response or session identifier, expected completion window and business deadline. Mark the workflow complete only after a recognized terminal state and a successful retrieval of the final output.
Subscribe to every terminal outcome
A handler that listens only for completion loses failed, incomplete and canceled work. The missing event then looks like an ordinary delay even when no usable result will arrive.
Handle completed, failed, incomplete and canceled states explicitly. Route each outcome to a deterministic next step, including retry eligibility, owner notification and whether downstream actions must stay blocked.
Operating control table
| Control | Evidence | Failure prevented |
|---|---|---|
| Owner | Named reviewer and decision | Unclaimed change |
| Configuration | Saved source state | Cannot reproduce |
| Validation | End-to-end fixture | Partial success accepted |
| Threshold | Success and rollback rule | Result reinterpreted |
| Evidence | Public or reporting read-back | Configured mistaken for delivered |
Verify the webhook before trusting its payload
An unsigned or incorrectly verified callback can be forged, replayed or associated with the wrong environment. Paid-media automation should never release a change because an untrusted request claimed that a job finished.
Verify the signing secret, enforce an acceptable timestamp window and persist event identifiers for replay protection. Retrieve the authoritative response by ID before using its result.
Return quickly and continue in a worker
Webhook endpoints that perform a full report build before returning can time out and trigger repeated delivery. The duplicate callback can create multiple reports or approval requests.
Acknowledge the event with a successful response as soon as it is safely accepted, then move longer processing to a durable worker. Make the worker idempotent by event and response ID.
Reconcile the result with the source account
A completed model response reflects the inputs it received, not necessarily the live account at review time. Campaigns, budgets and statuses can change while the job runs.
Read the relevant account objects again before presenting or applying the result. Display the source timestamp, current timestamp and any material diff that changes the recommendation.
Give stalled jobs an owner and deadline
Polling forever is not recovery. A queued audit that misses a reporting meeting needs a defined escalation even if the platform has not returned an error.
Set a maximum age for each workflow, stop polling after a terminal state and assign an owner for overdue jobs. Preserve the evidence needed to resume, cancel or rerun without duplicating prior work.
Connect the control to the wider system
Use this process alongside the strict tool-schema and mutation read-back framework. The platform setting, creative promise, measurement rule and downstream experience should describe the same decision.
Pre-launch checklist
- Write the decision and owner in plain language.
- Capture the source configuration and evidence.
- Test one realistic customer or reporting path end to end.
- Verify mobile presentation, links and accessibility.
- Record the success threshold and rollback trigger.
- Schedule the next review before material spend accumulates.
Run a cross-functional release review
Ask the channel owner, creative lead, analyst and conversion owner to inspect the same case. Each reviewer should state what was configured, what the customer sees, which event proves the outcome and which limitation changes the interpretation.
Repeat the test on mobile and with one adverse condition. Resolve differences in the campaign brief and measurement plan before launch.
Set the review cadence before launch
Assign a monitoring owner, review frequency and evidence threshold before the change is released. Preserve the pre-change baseline and avoid silently editing the definition after results arrive. This turns the launch into a measurable operating decision instead of a one-time interpretation of platform guidance.
At the first review, reconcile the source platform, public experience and one downstream business system. Close the change only when all three support the same conclusion, and record any limitation that future analysts must preserve.
Frequently asked questions
Should this be automated?
Automate repeatable validation and monitoring, but keep material policy, claim and budget decisions inside named approval limits.
What evidence should be preserved?
Keep the source configuration, decision owner, expected outcome, observed result and any exception that changes interpretation.
When should the team roll back?
Rollback when the control is inactive, the customer experience breaks, totals do not reconcile or the agreed threshold is breached.
