Let AI Agents help optimize your performance marketing campaigns
How AI Agents Can Run the Five-Stage Performance Marketing Funnel
AI agents can do most of the weekly work in a five-stage paid funnel. The marketer approves new ads, new offers and budget changes past the limits.
Paid media funnel has five stages. The funnel splits one ad budget across awareness, consideration, conversion, retargeting and retention. The split is 30%, 25%, 25%, 10% and 10%. Each stage has its own audience, creative, message, call to action (CTA) and key performance indicators (KPIs).
The funnel is simple to read. To run it, a marketer must build audiences, brief creative, read reports, move budget and plan tests. The marketer does most of these tasks again every week.
This post answers one question: which tasks in this funnel can AI agents do, and which tasks must the marketer keep? Part 1 shows how a human performance marketer runs the funnel. Part 2 shows the tasks that an agent can take over. The last section asks whether AI agents can run the full funnel as a loop.
The five stages at a glance
The fifth stage connects the end of the funnel to the start. Retention data shows which customers spend the most. The marketer uses that customer profile to find new people in Stage 1.
The five-stage funnel, stage by stage
Select a stage. Then select Human marketer or With agents to see who does the work in that stage.
Percent = share of the total ad budget. The dashed line is the return path: the profile of the best customers goes back to Stage 1.
Stage 1: Awareness
Cold audience. 30% of budget.
- Audience
- Broad and interest-based
- Creative
- Short-form video, educational content, problem-aware ads
- Message
- Explain the problem to the viewer
- CTA
- Learn More, Watch More
- Goal
- Build awareness and qualified traffic
- KPIs
- CPM, reach, video view rate, CTR
Human marketer
- Set up broad and interest-based ad sets on Meta, TikTok or YouTube.
- Brief short-form videos that state the problem in the customer’s words.
- Monitor CPM, reach, video view rate and CTR.
- Stop the videos that people skip. Give more budget to the videos that people watch.
- Save the video viewers as an audience for Stage 2.
With agents
- The research agent searches reviews and sales calls for the words that customers use.
- The creative agent turns those words into scripts and hooks.
- The analyst agent reports the videos with a low view rate.
- The media agent saves the video viewers as the Stage 2 audience.
Marketer approvesScripts and final videos, before launch
Stage 2: Consideration
Problem-aware audience. 25% of budget.
- Audience
- Video viewers, website visitors, social engagers
- Creative
- Product demos, benefits, comparisons, case studies
- Message
- Show how the product solves the problem
- CTA
- Explore, Learn More
- Goal
- Build consideration and intent
- KPIs
- CTR, CPC, landing page views, engagement rate
Human marketer
- Build warm audiences from video viewers, website visitors and social engagers.
- Brief demo, comparison and case study ads.
- Compare landing page views with clicks. A large difference often shows a slow or broken page.
- Send engaged visitors to Stage 3.
With agents
- The media agent builds warm audiences from video viewers and site visitors.
- The creative agent drafts demo, comparison and case study ads from approved proof.
- The analyst agent compares landing page views with clicks and reports a large difference.
Marketer approvesNew claims in demo, comparison and case study ads
Stage 3: Conversion
Solution-aware or ready to buy. 25% of budget.
- Audience
- High-intent website visitors, product viewers, leads
- Creative
- Testimonials, offers, product proof, benefit-led ads
- Message
- Give a reason to convert now
- CTA
- Buy Now, Book Demo, Sign Up
- Goal
- Generate conversions
- KPIs
- Conversion rate, CPA or CPL, ROAS
Human marketer
- Target product viewers, pricing page visitors and leads.
- Run testimonials, offers and product proof.
- Make sure that each conversion event fires correctly in the ad platform and in the customer relationship management (CRM) system.
- Measure each ad set by CPA or CPL and by ROAS.
With agents
- The analyst agent checks every day that the conversion events fire.
- The media agent moves budget to the ad sets with the best CPA or ROAS, inside the limits.
Marketer approvesNew offers, prices and testimonials
Stage 4: Retargeting
Very high intent. 10% of budget.
- Audience
- Cart abandoners, form abandoners, repeat website visitors
- Creative
- Testimonials, objections, urgency, FAQs, offer reminders
- Message
- Remove the last reason that stops the conversion
- CTA
- Complete Purchase, Book Now
- Goal
- Recover lost conversions
- KPIs
- CPA, conversion rate, ROAS, frequency
Human marketer
- Build audiences of cart abandoners, form abandoners and repeat visitors.
- Write ads that answer the most common objections from sales calls and support tickets.
- Set a frequency cap. The cap limits how often one person sees the same ad.
With agents
- The research agent collects the most common objections from support tickets and sales calls.
- The creative agent drafts one ad for each objection.
- The media agent keeps frequency below the cap.
Marketer approvesUrgency claims and offer deadlines
Stage 5: Retention
Existing relationship. 10% of budget.
- Audience
- Existing customers, previous leads, high-value customers
- Creative
- Upsells, cross-sells, new launches, loyalty offers
- Message
- Give customers the next reason to buy
- CTA
- Upgrade, Buy Again, Explore
- Goal
- Increase customer value
- KPIs
- Repeat purchase rate, LTV, CAC to LTV ratio, revenue
Human marketer
- Sync the customer lists from the CRM or the store.
- Run upsell, cross-sell and launch ads.
- Exclude recent buyers from Stages 1 to 3. This stops the funnel from paying again for customers that it already has.
- Give the list of best customers to the ad platform as a source list. The platform uses it to build a lookalike audience for Stage 1: new people who are similar to that list.
With agents
- The lifecycle agent sorts customers by value and syncs the lists to the ad platforms.
- The lifecycle agent excludes recent buyers from Stages 1 to 3.
- The lifecycle agent sends the best customers to Stage 1 as the source list for a lookalike audience.
Marketer approvesMessages to existing customers
Terms in the funnel:
- CPM: the cost per 1,000 impressions.
- CTR: the click-through rate.
- CPC: the cost per click.
- CPA: the cost per acquisition. CPL: the cost per lead.
- ROAS: the return on ad spend.
- LTV: the customer lifetime value. CAC: the customer acquisition cost.
Each stage sends its audience to the next stage:
- People who watch the awareness videos go to consideration.
- People who visit the site go to conversion.
- People who leave a cart or a form go to retargeting.
- Buyers go to retention.
- Retention sends the profile of the best customers back to awareness.
Part 1: How a human performance marketer runs the funnel
In the human version, one marketer does all of the tasks in the funnel above. To see the tasks for one stage, select the stage and then select Human marketer. Most of the tasks repeat each week, in this cycle:
- Pull the numbers from each ad platform, the analytics tool and the CRM into one report.
- Compare each stage with its KPI targets.
- Find the weakest stage.
- Plan one test for that stage: a new creative, a new audience or a new offer.
- Move budget between ad sets, inside the budget split for each stage.
- Update the audiences and the exclusions.
- Report the results to the founder or the client.
Steps 1, 2, 5 and 6 follow fixed rules. Steps 3 and 4 need judgment about the market and the customer. In step 7, the marketer explains the results to the people who pay for the ads. Part 2 uses this difference to decide which tasks go to agents.
Part 2: Where agent-led GTM takes over
Each task in Part 1 has four parts: collect data, compare it with a target, make a change, and report. AI agents can do these four parts every day.
An AI agent is software that uses a large language model (LLM) to read data, use tools and take steps toward a goal. Agent-led go-to-market (GTM) is a new go-to-market motion. In this motion, agents do the repeated GTM work. The marketer sets the goals, the limits and the approvals.
The agent team
| Agent | Job in the funnel | Data and tools |
|---|---|---|
| Research agent | Finds the words that customers use for their problems and objections | Reviews, sales call transcripts, support tickets, competitor ad libraries |
| Creative agent | Drafts scripts, hooks, captions and ad variations for each stage | The brand guide, past ads that performed well, the research agent’s notes |
| Media agent | Builds audiences, sets exclusions and moves budget inside the limits | Ad platform APIs for Meta, Google and TikTok |
| Analyst agent | Pulls data, compares each KPI with its target and reports problems | The ad platforms, the analytics tool, the CRM |
| Lifecycle agent | Sorts customers by value and syncs customer lists to the ad platforms | The CRM or the store |
| Orchestrator agent | Runs the weekly cycle, gives tasks to the other agents and asks the marketer for approval | A task board, a schedule, a shared memory |
An application programming interface (API) is the access point that lets software read and change an ad account. To see the agent tasks for one stage, select the stage in the funnel above and then select With agents.
The weekly cycle with agents
- The analyst agent pulls the numbers every day.
- The analyst agent compares each stage with its KPI targets and reports the weakest stage.
- The orchestrator agent recommends one test for that stage. The marketer approves the test or selects a different test.
- The creative agent drafts the test assets. The marketer approves them.
- The media agent moves budget and launches the test, inside the limits.
- The lifecycle agent updates the audiences and the exclusions.
- The orchestrator agent writes the report and lists the open decisions for the marketer.
The marketer still makes the decisions in steps 3 and 4. The agents do the work before and after those decisions, and they do it every day.
What the marketer keeps
An agent with full access to an ad account and no limits can spend the full budget on the wrong audience. The marketer sets limits to prevent this.
| Area | What the marketer does |
|---|---|
| Targets | Sets the budget split and the KPI target for each stage, for example a target CPA or a minimum ROAS. |
| Spend limits | Sets a daily cap for each stage and a maximum budget change per day. The media agent must stay inside both limits. |
| Brand and claims | Approves each new ad, offer, price and testimonial before launch. Each testimonial must come from a real customer. |
| Strategy | Decides on a new market, a new offer or a new channel. An agent can recommend these changes. An agent must not make them. |
| Meaning of results | Decides what a result means and what to do next. Attribution data often gives an incomplete picture. |
With these limits in place, the weekly review is a short list of open decisions for the marketer.
Can AI agents run the funnel as a loop?
Yes. A control loop measures a result, compares it with a goal and makes a change to reduce the difference. A thermostat is a simple control loop. The five-stage funnel already has the four parts of a control loop:
| Part of a control loop | Where it is in the funnel |
|---|---|
| Goal | A KPI target and a budget share for each stage |
| Sensor | Daily results from the ad platforms, the analytics tool and the CRM |
| Actuator | Platform APIs that let an agent change audiences, budget and ads |
| Feedback path | Retention data that goes back into Stage 1 targeting |
The marketer approves one gate in each agent cycle
Select a step, or use Next, to walk through one cycle of the loop.
After step 5, the loop starts again at step 1.Next cycle: daily checks, weekly tests
Step 1: Observe
- Pulls spend, reach, clicks, conversions and revenue for each stage from the ad platforms, the analytics tool and the CRM.
- Checks that each conversion event fires.
OutputOne daily report for all five stages
Step 2: Compare
- Compares each KPI with its target.
- Reports the weakest stage.
- Reports problems that need fast action, for example a broken conversion event or frequency above the cap.
OutputA list of the stages that miss their targets
Step 3: Decide
- Selects the weakest stage.
- Recommends one test for that stage: a new creative, a new audience or a new offer.
- Asks the creative agent to draft the test assets.
OutputOne test plan with draft assets, sent to the approval gate
Approval gate
- Approves or rejects new ads, new offers, prices and testimonials.
- Approves or rejects budget changes past the limits.
- Decides on any new channel or new market.
- Changes inside the limits do not stop at the gate.
OutputAn approved test, or a rejected test with a reason
Step 4: Act
- The media agent moves budget inside the limits and launches the approved test.
- The creative agent hands the approved assets to the media agent.
- The lifecycle agent updates the audiences and the exclusions.
OutputLive changes in the ad accounts, each one logged
Step 5: Report
- Writes the results of the test and of each budget change.
- Lists the open decisions for the marketer.
- Starts the next cycle at step 1.
OutputA weekly report and a short list of decisions
In each cycle, the agents do five steps: observe, compare, decide, act and report. The marketer approves at one gate, between decide and act. The gate stops the agents before they launch a new ad or offer, or move budget past the limits.
The funnel as a configuration file
A short configuration file can describe the full funnel. Each line is a rule that the agents follow. The marketer sets the targets and the limits.
loop: five-stage-paid
cadence:
observe: daily
test: weekly
report: weekly
budget_split:
awareness: 0.30
consideration: 0.25
conversion: 0.25
retargeting: 0.10
retention: 0.10
stages:
awareness:
audience: [broad, interest]
kpis: [cpm, reach, video_view_rate, ctr]
sends_to: consideration # video viewers
consideration:
audience: [video_viewers, site_visitors, social_engagers]
kpis: [ctr, cpc, landing_page_views, engagement_rate]
sends_to: conversion
conversion:
audience: [high_intent_visitors, product_viewers, leads]
kpis: [conversion_rate, cpa, roas]
sends_to: [retargeting, retention]
retargeting:
audience: [cart_abandoners, form_abandoners, repeat_visitors]
kpis: [cpa, conversion_rate, roas, frequency]
retention:
audience: [customers, previous_leads, high_value_customers]
kpis: [repeat_purchase_rate, ltv, cac_to_ltv, revenue]
sends_to: awareness # lookalike source list
limits:
kpi_targets: set_by_marketer
max_daily_budget_change: set_by_marketer
frequency_cap: set_by_marketer
approval_gate:
- new_creative
- new_offer_or_price
- testimonial
- budget_change_past_limit
- new_channel_or_marketThree conditions for the loop to work
- Clean conversion data. The agents optimize toward the data that they get. When a conversion event breaks, the loop optimizes toward the wrong result. The analyst agent must check tracking every day.
- Changes at the correct speed. Conversions often arrive days or weeks after the click. Ad platforms also need a learning period after each large change. The loop can check results every day, but it must change campaigns only once a week.
- A steady supply of creative. The loop needs new ads every week. Agents can draft many ads in a short time. The marketer’s approval then becomes the slowest step. A clear brand guide helps the marketer approve ads faster.
What the ad platforms already automate
Meta Advantage+ and Google Performance Max already automate audience targeting, budget allocation, bidding and some creative work. Each tool works inside its own platform. It uses the data in that platform and the conversion data that you send to it. It does not read your sales calls or support tickets. It does not move budget to a different platform.
The agent loop works at the level of the full funnel. It moves budget between stages and between channels. It also sends retention data back to Stage 1. Inside the loop, the platform tools continue to optimize each campaign.
The answer
Yes, with limits. Agents can do the daily and weekly work in all five stages. The marketer must keep the targets, the spend limits, the brand approvals and the decisions about what the results mean.
With this setup, one performance marketer directs a team of agents. The marketer approves decisions and does not do each step by hand.
Plan and run your own loop
Start with a plan for each stage. Then decide which work goes to agents.
Loop Campaign Builder
Plan your own loop campaign stage by stage. Download the plan when you finish.
Open the builderGTM Agent Kit
Eleven GTM agents draft the brief, the research, the messaging, the plan, the ad copy and the launch checklist. They stop for your approval at each gate. They never launch a campaign or change an ad budget, so the spend decisions stay with you.
$39 Free for the first 50 founding marketers.
Common questions
What are the five stages of a performance marketing funnel?
The five stages are awareness, consideration, conversion, retargeting and retention. In this funnel, the budget split is 30%, 25%, 25%, 10% and 10%. Retention sends the profile of the best customers back to awareness, so the funnel works as a loop.
What is agent-led GTM?
Agent-led go-to-market (GTM) is a new go-to-market motion. AI agents do the repeated GTM work. The marketer sets the goals, the limits and the approvals.
Which performance marketing tasks can AI agents do?
Agents can pull and compare data every day, find the words that customers use, draft ads, build audiences, set exclusions, move budget inside set limits and write the weekly report.
What must the human marketer keep?
The marketer keeps the targets, the spend limits, the approval of each new ad, offer, price and testimonial, the strategy, and the decisions about what results mean.
Can AI agents run a performance marketing funnel as a loop?
Yes, with limits. The funnel has a goal, a sensor, an actuator and a feedback path, which are the four parts of a control loop. The loop needs clean conversion data, weekly changes and a steady supply of approved creative.
How is an agent loop different from Advantage+ or Performance Max?
Advantage+ and Performance Max optimize campaigns inside one ad platform. An agent loop works across the full funnel. It moves budget between stages and channels, reads sales calls and support tickets, and sends retention data back to Stage 1.
