Skip to main content
Workflows turns supported repeated work into a persistent configuration with a trigger, delivery, status, and run record. Preset reports and alert categories use focused setup panels; custom, research, and acting workflows are shaped with the Marketing Agent. Open Workflows and confirm the active workspace.
Workflows library with an Autopilot prompt, example recurring requests, Daily Marketing Pulse, Weekly Marketing Pulse, Custom report, Advantage Plus drift, Spend spike, ROAS drop, CTR collapse, and More alerts cards

Workflows combines an Autopilot request field with preset report and alert cards, so a repeated outcome can start from a description or a known configuration.

​
Choose the correct workflow class

Alert auto-diagnosis and AI weekly brief appear under Coming soon. Do not configure or document them as working automations.

​
Configure a preset report

Choose Daily Marketing Pulse or Weekly Marketing Pulse when the template matches the question. Choose Custom report when metrics, channels, window, or narrative need agent-led definition.
1

Open the report card

Confirm its title, purpose, and whether an existing configured report is already bound to the template.
2

Configure Schedule

Choose the available cadence, day, time, and reporting timezone. The preview and later runs use this recurring date logic.
3

Configure Slack

Choose an available channel. If Slack is disconnected, use Connect Slack or leave Slack out; email works independently.
4

Configure Email

Select workspace members or enter the intended recipients. Recheck the full list before activation.
5

Review real-data previews

The panel shows what lands in Slack and an Email preview. Compare metrics, account scope, date logic, attribution, currency, and wording with Analytics.
6

Save or make it Live

Save the configuration while it is still being reviewed. Turn the live control on only when schedule, destinations, and preview are correct.
The setup supports Slack, email, or both. A report can generate successfully while one delivery leg fails, so do not reduce the result to a single “sent” state.

​
Configure a Custom report

Custom report opens the Marketing Agent with a structured request. Define:
  • Metrics and conversion definitions
  • Data channels and provider accounts
  • Reporting window and comparison
  • Attribution source and currency
  • Cadence and timezone
  • Slack channel and/or email recipients
  • Narrative, tables, or decision criteria
Require a draft and real-data preview before activation. When starting from Analytics with Save as report, verify that the seeded channel, dates, comparison, currency, and visible metrics survived the handoff.

​
Configure an alert

The current library can include:
  • Advantage+ drift
  • Spend spike
  • ROAS drop
  • CTR collapse
  • More alerts…, including supported detectors such as zero delivery, learning-phase stuck, or frequency saturation
All current alert cards require a connected Meta account.
1

Choose the event

Select the alert category that matches the operational risk. More alerts… helps choose another supported detector.
2

Set Slack delivery

Turn the Slack station on and choose a channel, or leave it off. A selected channel is required when Slack delivery is enabled.
3

Set Email delivery

Turn the Email station on and confirm recipients, or leave it off.
4

Review both previews

Confirm the message identifies the event, affected object, metric or state, and enough context for the response owner.
5

Save the live state

Select the setup action only after at least one intended destination is configured. The confirmation states the active delivery legs or that the alert is off.
An alert detects a condition; it does not establish cause. Assign an owner and an investigation path rather than treating the notification itself as a diagnosis.

​
Configure Metric guard

Metric guard is an agent-led acting workflow. Its supported action boundary is intentionally narrow:
  • Meta only
  • Campaign, ad set, or ad level
  • Pause only
  • Never resumes delivery
  • Never changes budgets
  • Requires your approval of the list before any object pauses
When shaping the request, specify:
  1. Object level.
  2. Metric and comparator.
  3. Threshold.
  4. Minimum spend required to qualify.
  5. Trailing number of days.
  6. Evaluation frequency and timezone.
  7. Eligible provider account and object set.
  8. Where to report what changed.
  9. Maximum intended effect and approval owner.
Review the proposed list on every trigger. After approval, verify the exact object state in Campaigns and Meta.
A recurring threshold is not a complete safety rule. Confirm account, level, date window, minimum evidence, eligible list, pause effect, and human approval before activation.

​
Configure Continuous brand research

Continuous brand research is a scheduled read-and-report workflow over the brand’s external landscape. It does not require an ad connection.
  • Read the displayed standard per-run credit cost.
  • Read the additional deep-mode cost when that option is requested.
  • Confirm sources, cadence, timezone, and delivery.
  • Keep it off until the agent’s proposed workflow is reviewed.
  • Adjust the existing workflow rather than creating a duplicate weekly scan, which would repeat usage.
Use a one-time research task when the question should not recur.

​
Operate Your workflows

The register columns include Name, Kind, Trigger, Delivery, and Last run.

​
Interpret status

Delivery can show a Slack channel, email-recipient count, or No targets. Last-run text distinguishes results such as Delivered, a failure/partial state, a normal run, notification, or Stopped at approval.

​
Use row actions

For reports and automations, the row menu can include:
  • View
  • Run now
  • Activate for a draft
  • Resume for a paused workflow
  • Pause for a live workflow
  • Edit in chat
  • Delete
Agent rows expose View runs only when a run destination is available. Run now does not change the schedule. For a report it sends the report pipeline; for a scheduled automation it starts a real run; for an event alert it can be a test fire without a real provider event. Read the resulting run log before treating a manual alert test as proof the live detector fired. Deleting shows: It stops running immediately. Run history is kept. Confirm the exact workflow before Delete.

​
Inspect run detail

For every first run and material edit:
  1. Open View or the workflow row.
  2. Confirm the trigger and current configuration.
  3. Open the run log.
  4. Separate data generation, notification/action result, and each delivery leg.
  5. Read partial and failed items at the lowest available level.
  6. For an acting workflow, verify the provider object’s state.
  7. Resolve Needs you or Stopped at approval in the linked task or Awaiting action.
Configuration is operational only after a terminal run and intended delivery are verified.

​
Completion check

  • The workflow class matches the desired effect.
  • Provider, account, trigger, metric, window, timezone, and destinations are correct.
  • The preview uses recognizable data from the intended frame.
  • Live, Draft, or Paused state is intentional.
  • A first manual or scheduled run reached a terminal result.
  • Slack and email delivery outcomes were inspected separately.
  • A manual alert test is not confused with a real detector event.
  • Guarded actions remained inside the Meta pause-only, human-approved boundary.
  • Provider state was verified after every approved action.

​
Troubleshooting

Read its prerequisite guidance. Reports require a supported ad connection, current alerts and Metric guard require Meta, and delivery depends on the selected Slack or email setup and your role.
Align provider account, date logic, timezone, currency, attribution, and metrics with Analytics. Correct the draft before activation.
Select Connect Slack or leave Slack out and use email when appropriate. Confirm no missing Slack channel blocks save.
Confirm it is Live, its trigger occurred in the configured timezone, and required sources are healthy. Inspect the run log for queueing, an error, Needs you, or approval.
Inspect Slack and email separately, repair channel access or recipients, then use Run now. Do not recreate the workflow to fix one delivery leg.
A manual alert run can be a test fire without provider trigger context. Verify the delivery path, then wait for or deliberately test the real detector condition in an approved environment.
Reject or stop the action, pause the workflow, and compare provider, level, eligible list, threshold, minimum spend, and trailing window. Reactivate only after the proposed list is correct.
Compare triggers, scopes, and destinations, pause the duplicate, and keep one owner. For continuous research, update the existing scan rather than creating a second billable schedule.

​
Continue

Build the Analytics frame

Verify the data view before making it recur.

Review Awaiting action

Decide against the exact proposed object and effect.
Last modified on August 9, 2026