7 agentic performance marketing workflows to run in Claude or ChatGPT (2026)

With Supermetrics, you can use Claude or ChatGPT to create cross-channel reports, investigate performance anomalies, reallocate budgets, review pacing, analyze creative, build and launch campaigns, and connect ad spend to revenue. Below are seven agentic performance marketing workflows, with prompts you can adapt to your own accounts.

Your CPA has jumped, the weekly report is due, and three ad platforms need budget changes. You know what needs investigating. But getting the data together and acting on it takes up time you don't have.

Once you connect Claude or ChatGPT to your marketing data through Supermetrics, you can work through those tasks in one conversation. Ask why a campaign is underperforming, let the agent investigate, and approve the next action in the same chat.

Agentic performance marketing workflows cut the time spent assembling data, so more of your week goes on deciding what to change.

Key takeaways

• Supermetrics connects Claude and ChatGPT to ad platforms, GA4 and CRM data, so one conversation can report, investigate and prepare campaign edits.
• The seven workflows are cross-channel reporting, anomaly investigation, budget reallocation, pacing review, creative analysis, campaign build and launch, and spend-to-revenue analysis.
• Campaign edits need write access in Supermetrics Hub. With "All live changes" selected, every live edit waits for your approval, and Supermetrics creates new campaigns paused.
• Agent explanations are hypotheses: check conversion lag, CRM match coverage and platform budget settings before acting.
• Verdict: Start with the weekly cross-channel report, the lowest-risk workflow, then add write access once the numbers match your source platforms.

What is an agentic performance marketing workflow?

An agentic performance marketing workflow uses AI to work toward a marketing goal by choosing which data to retrieve, what to investigate next, and which actions to propose or carry out through connected tools.

For example, ask an agent to investigate why cost per acquisition (CPA) exceeded $80. It can compare campaigns, drill into the largest changes, and recommend a response. If a budget change makes sense, you can ask it to prepare the edit for your approval. For the wider picture, see what agentic marketing is and the best AI agents for marketing.

How do you set up Claude or ChatGPT for these workflows?

Connect Supermetrics to Claude or ChatGPT to give your AI access to the accounts you want to analyze. For workflows that include campaign changes, you'll also enable campaign write access.

You don't need to build an agent. Claude and ChatGPT already choose tools and work through tasks based on what they find. Supermetrics connects them to your marketing data and campaign tools.

You'll need a Supermetrics trial or subscription with Supermetrics for AI Chats, the Supermetrics product that connects your marketing data to AI assistants such as Claude and ChatGPT. Then:

1. Open Supermetrics for Claude or Supermetrics for ChatGPT and connect your data sources.
2. Confirm the client, account IDs, currency, and time zone. Check last week's spend against the source platform.
3. For campaign actions, grant campaign write access in Supermetrics Hub to the intended accounts. Select "All live changes" to require approval for live campaign edits.

Supermetrics creates new campaigns paused, so you can review them before launch.

Try the prompts below on your own accounts, adapting them to your goals. Once a workflow works well for your team, save the prompts alongside your goals and metric definitions so you can reuse them next time.

Which agentic performance marketing workflows should you run in 2026?

The seven agentic performance marketing workflows worth running in Claude or ChatGPT in 2026 are cross-channel reporting, anomaly investigation, budget reallocation, pacing review, creative analysis, campaign build and launch, and connecting ad spend to revenue.

Each workflow uses its own example budgets and thresholds, so replace them with your own. Run each workflow's numbered prompts in the same conversation, starting with this instruction:

Use only [workspace/account IDs], reporting in [currency] and [time zone]. Confirm dates and available fields. Our conversion is [purchase/demo/qualified lead definition]. Keep different conversion events separate and calculate ratios from totals. Flag missing data, zero denominators, and recent results affected by conversion lag. Don't change campaigns without my approval.

1. How do you create a cross-channel performance report?

Ask the agent to compare channel performance, investigate the biggest changes, and write a stakeholder summary. This replaces repeated exports with a conversation grounded in your connected data.

Sources: Your paid channels and GA4. Use CRM or ecommerce data if your report needs a shared revenue or acquisition measure.

Try these prompts:

Using Supermetrics, compare last week's Google Ads, Meta Ads, and LinkedIn Ads performance with the previous week. Show spend, conversions, CPA, and return on ad spend (ROAS) where available. Label each source's conversion event and attribution settings, with percentage changes.
Investigate the two channels with the largest CPA changes. Break results down by campaign and show changes in click costs and conversion rate. Identify the biggest contributors and what to check next.
Write a 150-word summary for our head of marketing: what changed, the evidence, and one recommended next step. Don't combine overlapping platform conversions into a unique acquisition count.
Publish a dashboard of this analysis to Supermetrics Studio, using refreshable Supermetrics data and clear source labels. Give me a shareable link.

Want your team to use the report after the conversation ends? Publish it to Supermetrics Studio. Your AI-built dashboard draws on the same Supermetrics connections, so colleagues or clients can check current performance without asking you to rebuild the report.

2. How can an agent investigate campaign performance anomalies?

Give the agent a threshold and ask it to investigate breaches. It can identify where performance changed and suggest explanations, but the data may not establish a single cause.

Sources: Your ad platforms, plus GA4 and your CRM or ecommerce platform for cross-checks.

Try these prompts:

Using Supermetrics, compare yesterday's Google Ads, Meta Ads, and LinkedIn Ads campaigns with the same weekday across the previous four weeks. Calculate baseline CPA and conversion rate from combined totals. Flag CPA increases above 20% or conversion-rate drops above 30% for campaigns with at least $200 spend yesterday. Show spend, conversions, and baseline sample sizes.
For each flag, drill into the available ad group or ad set and ad data. Look for changes in delivery, click costs, and conversion rate. Cross-check GA4 activity and CRM or Shopify results where comparable data is available. Check conversion lag before treating a decline as a performance problem.
Rank findings by priority. Show the observed change, likely explanation, supporting evidence, and next check. Treat explanations as hypotheses, not proven causes.

For example, the agent might find the following. These numbers are illustrative:

Observation: Meta prospecting CPA rose from $64 to $91, while click-through rate (CTR) fell from 1.4% to 0.8% and frequency increased.

Hypothesis: Creative fatigue may be contributing.

Check next: Compare audience delivery, recent edits, and conversion lag before testing replacement creative.

To see how an agent handles this kind of investigation end to end, read our guide to AI campaign management in Claude.

3. How do you let an agent reallocate budget across channels?

Ask the agent to recommend budget changes within explicit constraints, using comparable acquisition metrics. Review the proposal before applying it to the selected accounts.

Sources: Your paid channels and GA4 or CRM data that reflects your chosen conversion goal.

Try these prompts:

Using Supermetrics, plan next month's $60,000 budget: Google Ads $20,000, Meta Ads $18,000, LinkedIn Ads $12,000, TikTok Ads $10,000. Our goal is CRM-qualified demos at $80 each. Compare the last 28 days by channel and campaign. Check CRM match coverage and qualification delays. Ask for a prepared dataset if comparable outcomes aren't available.
Propose allocations totaling $60,000. Change no channel by more than 15% and keep LinkedIn above $8,000. Leave planned allocations for campaigns launched in the last 14 days unchanged, asking for those amounts if needed. Explain each move and flag conflicting constraints.
Translate the proposal into campaign-level budgets and dates, showing current and proposed settings. Explain how platform spending rules affect the monthly plan. Apply only the edits and timing I approve. If future scheduling is unsupported, give me the changes to apply next month instead of changing live budgets now.

Illustrative plan: the total stays at $60,000, and no channel changes by more than 15%.

Before approving, check lead quality and whether campaigns can absorb more spend. A channel's historical CPA doesn't predict the return on extra budget. Use incrementality testing to assess whether additional spend brings additional conversions.

4. How do you run a weekly pacing and optimization review?

A budget pacing review projects month-end spend against your plan. It helps you spot overdelivery or underdelivery before deciding whether budgets need adjusting. For the underlying method, see our guide to ad spend tracking and pacing.

Sources: Your ad platforms and the monthly budget plan, including any planned launches or spending bursts.

Try these prompts:

Using Supermetrics, check this monthly plan: Google Ads $25,000, Meta Ads $20,000, LinkedIn Ads $10,000, Microsoft Advertising $5,000. Show spend through yesterday, expected spend to date, remaining budget, and projected month-end spend. Forecast using this month's average daily spend. Use an even daily plan unless I provide a spending schedule.
Flag channels forecast more than 10% over or under plan. Ask about upcoming launches or promotions before recommending changes.
Find campaigns above $120 CPA over the last 14 complete days with at least 30 conversions, plus high-spend campaigns with few conversions. Identify campaigns below $70 CPA with evidence of budget constraints. Propose pauses or budget edits, showing exact settings and projected spend. Apply only approved changes without exceeding the monthly plan.

Before acting, account for planned promotions and campaign purpose. A brand-defense campaign may need different performance criteria from prospecting.

5. How can an agent analyze creative performance and draft new ad variants?

Ask the agent to compare ad performance and draft variants for your next ad creative test. Patterns in winning ads are starting points for a test, not proof that a hook caused better results.

Sources: Ad-level performance, available ad copy or creative descriptions, and your brand guidelines. Supply missing creative context yourself.

Try these prompts:

Using Supermetrics, pull the last 30 complete days of Meta Ads performance by ad: spend, impressions, link clicks, conversions, link CTR, and CPA. Rank by CPA where ads have at least $500 spend and 10 conversions. Show lower-volume ads separately.
Compare top and bottom performers within similar objectives and audiences. Using available copy and the assets I supply, identify patterns in hooks, formats, offers, and CTAs. Ask for missing creative rather than guessing.
Propose three creative hypotheses and write four copy variants to test them. Follow this project's brand guidelines and approved claims.
After I select variants and supply assets, show the proposed ads, URLs, tracking, and target ad set. Create approved ads paused, leaving existing ads unchanged. Recommend a test setup and review point.

To assess the creative, upload images or paste copy into the chat, labeled with ad IDs. For video, provide a transcript and key frames or a supported file. Check claims and brand voice before publishing.

6. How do you build and launch a new campaign with an agent?

Give the agent a campaign brief and ask it to use relevant past performance to draft a new campaign. Supermetrics creates new campaigns paused so you can inspect the setup before enabling them.

Sources: Historical performance from your chosen platform, plus GA4 or CRM data where needed to identify valuable conversions.

Try these prompts:

Using Supermetrics, plan a LinkedIn Ads campaign for US mid-market SaaS marketing leaders: $5,000 over 30 days, starting [date], targeting demo requests. Review up to three comparable campaigns from the last year. Use CRM-qualified demos where matching data exists, or ask for an agreed proxy. Draft targeting, exclusions, dates, budget settings, and four copy variants. List missing assets or decisions.
Show the proposed campaign, including URLs, tracking, and supported targeting settings. Verify that platform budget settings fit the $5,000 plan. After approval, create it paused.
List pre-launch checks for targeting, exclusions, tracking, assets, and landing pages. Once I confirm those checks and approve launch, enable this campaign and report its status. Recommend a review date allowing for conversion lag.

Before launch, verify the campaign in LinkedIn Campaign Manager, including budget limits. A prompt alone doesn't enforce a spending cap.

See what AI agents can change in your ad accounts for supported platforms and actions as of October 2026.

7. How do you connect ad spend to revenue and pipeline?

Connect ad spend to CRM or ecommerce records to examine attributed revenue and acquisition costs. The result depends on your matching method, customer definitions, and reporting period.

Sources: Your ad platforms and HubSpot, Salesforce, or Shopify, with usable campaign identifiers or tracking fields.

Agree with sales or finance how to assign credit and define revenue. You'll need campaign tracking fields and links between leads, customers, and deals. If your connections don't expose those records, use a prepared CRM export or warehouse dataset.

Try these prompts:

Using Supermetrics, pull Q3 [year] ad spend by campaign and CRM records for leads acquired that quarter. Use captured source and campaign fields to propose an attribution method. Confirm the revenue definition and reporting cutoff date with me before matching.
Match that lead cohort to its resulting deals as of the agreed cutoff. Count each eligible new customer and each deal's revenue once. Keep spend separate to avoid multiplying it across leads. Show open pipeline, closed-won revenue, new customers, and unmatched records. Reconcile totals and report match coverage.
Calculate paid-media customer acquisition cost (CAC) as Q3 spend divided by matched new customers, excluding existing customers. Calculate attributed ROAS as matched closed-won revenue divided by spend. Flag incomplete matching and deals still in progress, which can inflate CAC and depress ROAS.
Summarize findings for our CMO, including attribution rules, match coverage, and measurement limits.

CRM-based marketing attribution assigns credit consistently. It doesn't prove which sales advertising caused, and paid-media CAC excludes other acquisition costs.

How do agentic workflows compare with automated rules and manual reporting?

Platform rules act on one metric in one channel, and manual reporting means exports and spreadsheets. An agent working through Supermetrics compares channels, investigates the cause, and prepares the edit in one conversation, but still needs your approval to act.

Automated rules in Google Ads or Meta Ads are reliable for simple triggers, such as pausing an ad when CPA passes a threshold. They can't check whether the same problem shows up in another channel, or whether conversion lag explains it. Manual reporting can answer those questions, but each answer means another export. Agentic workflows sit between the two: they investigate like an analyst and act only through the permissions you grant.

What are the limitations of agentic performance marketing workflows?

Agents explain changes but can't prove causes, recent results are distorted by conversion lag, prompts don't enforce spend caps, and campaign actions only work on platforms Supermetrics supports.

Each limitation has a control. Treat every explanation as a hypothesis to check. Exclude the most recent days when conversion lag is likely. Set spending limits in the ad platform and require approval for live changes in Supermetrics Hub, because a prompt instruction alone doesn't stop an edit. Check which ad platforms and actions are supported before planning a workflow that changes campaigns.

Which agentic performance marketing workflow should your team run first?

Start with the cross-channel report: it's read-only, runs weekly, and shows within one conversation whether your connected numbers match the platforms.

Once the report matches your source platforms, add the anomaly investigation, then grant write access for budget and pacing changes. Connect Supermetrics for Claude or Supermetrics for ChatGPT and try it on your accounts. Our step-by-step guide to managing ad campaigns with AI agents can help you take the next step.

Frequently asked questions