You probably already use AI to analyze campaign results and decide what to do next. But once you have the answer, the work then shifts back to you. You still have to log into the ad platform, find the campaign, and apply the change.
Agentic campaign management can bring that work into one workflow. Give an agent a goal, such as “find out why cost per acquisition (CPA) rose last week.” It can decide which connected data to examine, follow the evidence, and recommend a response. With the right permissions and approval controls, it can also make supported campaign changes from the same chat.
This article explains how to set that up in seven steps with Supermetrics, a marketing intelligence platform that connects marketing data to AI and AI to ad platforms, starting with read-only investigation before you allow the agent to change a live campaign.
Key takeaways
• Give an AI agent a campaign goal and let it choose how to investigate. It can recommend a response with read-only access.
• Start with one read-only account. Add metric definitions and decision rules, test past problems, then introduce limited write access with human approval.
• Select All live changes for your first live test. Supermetrics offers four approval modes, creates new campaigns paused, and records changes in Campaign history.
• As of September 2026, campaign management supports six ad platforms and is in early access, free with an existing Supermetrics subscription. Check that your chosen AI integration supports the action you want.
• Verify each edit in the ad platform, then review campaign results and the agent’s behavior before expanding.
What does “agentic campaign management” mean?
Agentic campaign management means giving an AI agent a campaign goal and letting it decide how to investigate it, rather than prescribing every step.
Ask “What was my CPA last week?” and the AI retrieves an answer. Ask “Why did CPA increase last week, and what should we do about it?” and an agent might check account-level performance, identify the campaigns behind the increase, inspect conversion volume and recent changes, then follow up on what it finds.
Does an AI agent run in the background?
Not necessarily. You can ask an AI agent in Claude to investigate a CPA increase and follow its work in the chat. You can also build a workflow in which a monitoring rule flags a problem and prompts an agent to investigate it. Teams building their own AI marketing workflows can use Build on Supermetrics to connect agents to marketing data through its APIs and Model Context Protocol (MCP) server. MCP is an open standard that lets AI tools connect to external data and actions.
The trigger doesn’t determine whether the workflow is agentic. What matters is whether the agent decides which steps to take as it investigates the goal.
Can you use agentic campaign management in Claude or ChatGPT?
Yes. Supermetrics connects campaign data to Claude, ChatGPT, and Gemini Enterprise so you can investigate performance in your AI chat. Its campaign-management tools let you view, create, and update campaigns, including budget edits and pauses, where the platform and interface support the action.
Making changes requires write access to the relevant ad account. The current setup instructions cover Claude, Gemini, and the Supermetrics MCP server. If you plan to apply changes through ChatGPT, confirm that write actions are available in your integration before starting a live test.
1. Connect the data needed for one campaign task
Start with a familiar problem and keep the agent’s access read-only.
Choose a past campaign problem your team has already investigated, such as a week when CPA rose. Gather the data and your team’s findings so you can later check whether the agent reaches a sensible conclusion.
Connect the accounts and sources needed for that investigation. For a Google Ads CPA increase, that might mean the relevant ad account; add analytics or CRM data only if the question requires it. Leave campaign write access off while you test.
Supermetrics makes data from your connected accounts available through AI Chats, its integrations for Claude, ChatGPT, and Gemini Enterprise. The agent can then investigate the problem in your AI tool without you exporting reports into the chat.
2. Add business context and metric definitions
Tell the agent how your team measures success and what the campaign data cannot tell it on its own.
Specify the outcome you care about, such as CPA or return on ad spend (ROAS), and the constraints that affect your decisions. These might include a minimum conversion volume, product margins, lead quality, or campaigns you cannot change.
Define the metrics precisely. Name the conversion action that counts, the attribution settings to use, and the source or calculation for any blended metric. Add context that could change the interpretation of the results, such as a promotion or a recent tracking change. If lead quality or margins matter, make sure the agent has access to that information; it cannot infer them from ad platform data.
For example:
Investigate why CPA increased in [account] during [period]. Use [conversion action] and [attribution settings]. Our target CPA is [target], but we need at least [minimum conversion volume]. Factor in [promotion or business constraint]. Do not recommend changes to [protected campaigns]. If the data is incomplete or the cause is unclear, say what you need to check next.
Save the context your team will reuse, and update it when targets, definitions, or campaign conditions change.
3. Set decision rules and stop conditions
Tell the agent when it has enough evidence to recommend a change, and when it should hold back.
Set the checks the AI agent must make before recommending action. The right thresholds depend on the campaign, but consider:
- Minimum spend or conversion volume
- A complete comparison period that accounts for conversion delay
- Enough time since the last campaign change
- No known tracking or reporting issues
- Supporting evidence beyond a single KPI
For Google Ads campaigns using Smart Bidding, Google recommends evaluating results over a period with at least 30 conversions, or 50 for Target ROAS, such as a month or longer. This is guidance for evaluating Smart Bidding, not a universal minimum for an agent to investigate or act. Set your decision threshold around the campaign’s conversion delay and the change you are considering. See Google’s Smart Bidding guidance.
Then set limits on what it may recommend. For example:
Do not recommend a budget change until the campaign has generated at least [minimum conversions] and [number of days] have passed since the last edit. Keep any proposed adjustment within [maximum percentage]. If the data is incomplete or the cause of the change is unclear, explain what needs checking before recommending action.
These are instructions for the agent’s judgment, not enforced limits on what it can change. You’ll set account write access and approval controls separately before testing any live changes.
4. Test the agent on known campaign problems
Check the AI agent’s investigation and judgment against past cases before giving it write access.
Select a few problems your team has already investigated. Include cases where a change was justified, where the team waited for more data, and where a tracking issue made the results unreliable. Give the agent the information it would have had at the time, then compare its response with the evidence and your team’s decision.
| What to test | What good looks like |
|---|---|
| Data | Uses the correct account, period, conversion definition, and metrics |
| Investigation | Checks plausible causes and shows the evidence behind its conclusion |
| Judgment | Distinguishes a likely explanation from a proven cause and acknowledges uncertainty |
| Decision | Recommends action, further investigation, or no change as the evidence warrants |
When the agent gets a case wrong, identify whether the problem came from missing data, unclear instructions, or faulty reasoning. Make the relevant adjustment, then test the agent again on another case. Keep a small set of cases to rerun when you change the workflow.
If missing or inconsistent data keeps affecting the results when testing your AI marketing agent, review how to make your marketing data AI-ready before expanding the workflow.
5. Add limited write access and approval gates
Give the agent access to one tested account, and require approval before it changes a live campaign.
In Supermetrics, enable write access for the account you used in your read-only tests. You’ll also need permission to edit that account in the ad platform itself. In the Supermetrics Hub, open Write settings, select the account you want to manage, choose All live changes, and save. You can grant write access only to accounts belonging to data source connections you own.
Supermetrics offers four approval modes: No human approval applies changes immediately; Going live requires approval before enabling or unpausing a campaign; Budget + going live also requires approval for budget changes on live campaigns; All live changes requires approval for every change to a live campaign, including enabling it. For your first test, keep All live changes selected. New campaigns are always created paused.
Tell the agent which actions it may propose and which are off-limits. For example, you might allow it to propose budget edits while ruling out campaign launches and targeting changes. Keep these campaign-agent guardrails distinct from the access and approval settings that control what it can actually apply.
Before the first live test, record the account with write access, the actions in scope, any protected campaigns, and who will review proposed changes.
6. Make one change and verify it in the ad platform
Start with a modest change that you can check immediately and reverse if needed.
Choose a current campaign problem where your decision rules support a change. Ask the agent to prepare the proposal before you approve anything:
Prepare a [percentage] budget change for [campaign]. Show the current and proposed budgets, the evidence for the change, and when we should review the result. Do not apply it until I approve.
Check the account, campaign, figures, and reasoning in the agent’s proposal. If the proposal is correct, approve that change only.
After the agent applies the change, check the new budget directly in the ad platform. Supermetrics also logs changes in Campaign history, where you can review and undo supported actions if needed. Verifying that the edit took effect is a separate check from deciding later whether it improved performance.
7. Monitor the result before expanding the workflow
Check what happened in the ad account and whether the agent followed the process you set.
First, confirm that the approved change took effect. Then give the campaign enough time to produce useful results. Account for conversion delay and normal variation before deciding whether the change helped.
Review the agent’s behavior separately from the campaign outcome:
- Did it use the right account, period, and metrics?
- Did its recommendation follow your decision rules?
- Did it only apply the change you approved?
- Did it report what happened accurately?
One favorable result doesn’t prove the workflow is reliable. Keep human approval in place while you test it across more cases. When you’re ready to expand, change one thing at a time. For example, add another account while keeping the same approval settings and decision rules.
Example: How might an AI agent respond to rising CPA?
An agent can check whether the CPA increase is real, trace it to a broken landing page, and propose pausing the affected campaign for your approval.
Suppose a completed weekly report shows that CPA rose 24% from the previous week and is now above target. Here’s how an agent could investigate before anyone changes a campaign. This example is illustrative.
| Stage | What happens |
|---|---|
| Check the signal | The agent confirms the account, conversion definition, reporting period, and data completeness. It checks whether conversion delay could explain some of the increase. |
| Follow the evidence | It finds that one campaign accounts for most of the change. Cost per click (CPC) is steady, but its conversion rate has fallen sharply. A landing page check shows that the campaign’s destination URL is broken. |
| Recommend a response | The agent proposes pausing the affected campaign while the team fixes the URL. It shows the evidence and the likely effect of leaving the campaign running. |
| Apply the approved change | A marketer approves the pause. The agent applies that change and confirms the campaign’s status in the ad platform. |
| Follow up | The team fixes and tests the page before deciding when to resume the campaign. It then reviews performance after enough conversion data has arrived. |
The agentic part is the investigation: the agent follows the CPA increase from the account to a campaign, checks what changed, and proposes a response to the problem it found. The marketer still approves the live change.
How can you use Supermetrics for agentic campaign management?
Supermetrics provides connected campaign data and tools to view, create, and update campaigns. Available actions vary by ad platform.
As of September 2026, campaign management supports Google Ads, Meta Ads, Microsoft Advertising, TikTok Ads, ChatGPT Ads, and Snapchat Marketing. Check the current campaign-management documentation for platform-specific limitations before planning a live change.
Access is set per ad account in Write settings. Choose from No human approval, Going live, Budget + going live, or All live changes. New campaigns are created paused, and changes are logged in Campaign history.
If you’re building an agent or an internal workflow, the Supermetrics MCP server provides connected marketing data and campaign actions. Your subscription must include MCP server access. Your team builds the logic around those tools, including when the agent runs and what it should investigate.
As of September 2026, campaign management is in early access and free to use with an existing Supermetrics subscription. Managing campaigns in an AI chat requires an active trial or subscription with an AI chat destination.
Start narrow, prove it works, then expand
Give your first agent one campaign problem and read-only access. Check whether it uses the right data, follows the evidence, and knows when to hold back. Once it performs reliably, you can grant limited write access and require approval for live changes.
Use Supermetrics to investigate connected campaign data in Claude, ChatGPT, or Gemini Enterprise. When you’re ready to test live changes, choose an interface with campaign write actions and enable the approval controls for your account.
Frequently asked questions
-
No. You can use Supermetrics in Claude, ChatGPT, or Gemini Enterprise to investigate campaign performance in a chat. The Supermetrics MCP server is available if you want to build your own agent or automated workflow.
-
Supermetrics’ campaign-management overview names ChatGPT, but its current write-access and campaign-editing instructions cover Claude, Gemini, and the MCP server. You can use ChatGPT to analyze connected campaign data. Before relying on it to apply changes, confirm that write actions are available in your integration.
-
As of September 2026, Supermetrics supports campaign management for Google Ads, Meta Ads, Microsoft Advertising, TikTok Ads, ChatGPT Ads, and Snapchat Marketing. Available actions vary by platform, so check whether your intended change is supported before building a workflow around it.
-
Yes. If you choose No human approval, changes apply immediately. The other three modes require approval at different points, and All live changes requires approval for every change to a live campaign, including enabling it. New campaigns are always created paused.
-
Yes, if you build a workflow that passes a monitoring signal to an agent. The monitoring rule detects the issue; the agent investigates the cause. That requires a trigger and orchestration around the agent. Connecting campaign data to an AI chat alone does not make it run continuously.
-
Verify the result in the ad platform and review the edit in Supermetrics’ Campaign history. You can undo supported actions. Before running the workflow again, work out whether the mistake came from the agent’s data, instructions, reasoning, or permissions.
-
Platform automation optimizes parts of campaign delivery within its own ad platform. An AI agent working through Supermetrics can investigate performance across connected accounts, explain what the evidence suggests, and propose a response for you to approve. The two can work together: an agent might recommend a target change, while Smart Bidding continues to optimize bids toward the approved target.
-
Yes. Configure write access separately for each client ad account. Supermetrics only lets you grant write access to accounts belonging to data source connections you own, and you also need edit permissions in the ad platform. Start with one client account and All live changes approval, then add accounts once the workflow performs reliably.
