How to make a Growth Marketing AI agent
Scale paid marketing faster with AI
You're in—expect an email shortly.
A growth marketing agent operates at a different level than a channel-specific agent. It watches the full funnel (acquisition through retention) and reasons about where the highest-leverage intervention is in the system, which may or may not be in a paid campaign.
Building this requires more data than a single-channel agent. You need attribution data joined to product analytics data joined to revenue data. Once that is assembled, the agent can see things that are invisible to any channel-specific tool: that one acquisition channel is producing customers who churn faster, that activation rate dropped across all channels simultaneously (a product problem, not a marketing problem), that LTV varies significantly by campaign even when CAC looks similar.
The data model
The foundation is two datasets joined by user ID:
Acquisition data. Your cross-channel attribution model: every incoming visitor is parsed and identified by traffic source, channel, and campaign. Credit is assigned to each touchpoint according to your model, deduplicated across both paid and organic channels. The key is having campaign-level attribution on all revenue movement metrics: new customers, expansion, churn, contraction, reactivation. A tool like Roadway handles this out of the box, or you build it internally.
Revenue data. Billing events: subscription start, upgrades, downgrades, churn, LTV at various time horizons.
When these two are joined, you can calculate: LTV by acquisition channel and campaign, activation rate by acquisition source, 30/60/90-day retention by channel, trial-to-paid conversion by campaign. These are fundamentally different metrics than CAC and ROAS, and they tell a fundamentally different story about which acquisition channels are actually building the business.
This data join lives in your warehouse. It is a prerequisite for a growth marketing agent that reasons at the system level rather than the campaign level.
Goal, funnel, and guardrail configuration
Goal metric. At the growth level, this is usually something like net new paid customers per month, or MRR growth, or LTV:CAC ratio at a specific time horizon. It is a business metric, not a campaign metric.
Funnel metrics. The full chain: first paid touch, site conversion, sign-up, activation milestone, paid conversion, 30-day retention, 90-day retention. Map each step. The agent watches all of them, because a bottleneck anywhere in this chain limits the goal metric, and the bottleneck could be anywhere.
Guardrail metrics. Growth marketing guardrails include both performance guardrails (CAC ceiling, ROAS floor) and quality guardrails: minimum activation rate for a campaign to stay active, minimum 30-day retention for a cohort before the acquisition channel is flagged. Quality guardrails are what pure performance marketing agents miss. A campaign with a great CAC but low activation rate is buying the wrong customers. The growth agent catches this.
Memory at the growth level
Growth marketing memory includes experiment history, not just campaign history. What growth hypotheses have been tested? What was the result? What was shipped to production versus abandoned? What channels have been tried and why were they deprioritized?
This accumulated knowledge is part of what makes an experienced growth team valuable. Encoding it in your agent means the agent does not recommend retesting things that failed, does not miss context about why certain decisions were made, and can reason about what has been tried and what has not when proposing the next experiment.
Tools and skills
Your agent needs two types of inputs: tools (API integrations that let it read and write data) and skills (markdown files that give it context and decision-making frameworks).
Tools (APIs):
- Google Ads API - read and write.
Standard Accessdeveloper token for mutate operations - Meta Marketing API - read and write. System User with
ads_managementpermission and admin/advertiser role - LinkedIn Ads API - read and write. OAuth2 with
rw_adsscope - Attribution / data warehouse - unified cross-channel attribution with campaign-level revenue movement metrics
- Product analytics (Mixpanel, Amplitude, or equivalent) - activation milestones, engagement patterns, feature usage, churn signals
- Billing / revenue system (Stripe, internal) - subscription events, upgrades, downgrades, churn, LTV calculations
- CRM - customer records, deal stages, sales pipeline data tied to acquisition source
Skills (markdown files):
- Growth model - how the business grows, what the key levers are, where the biggest constraints typically appear, current growth targets and timeline
- Activation definition - what counts as an activated user, which milestone events matter, expected activation rate by channel and segment
- Quality scoring framework - how to evaluate customer quality beyond CAC (activation rate, retention, LTV), minimum quality thresholds by channel
- Experiment history and methodology - what has been tested, results, what was shipped, what was abandoned and why, how to design new experiments
- Funnel diagnosis playbook - how to determine whether a bottleneck is in acquisition, activation, monetization, or retention, and what to do in each case
- Channel evaluation criteria - when to add a new channel, when to cut one, minimum test investment, how to evaluate early results
- Cross-team escalation rules - when an issue is a product problem vs. a marketing problem, who to flag, what information to include
The three levels
Monitoring. The agent watches the full funnel, not just campaign performance, but activation rates, cohort quality, retention by acquisition source, and any early signals that the system is changing. The most important thing the monitoring layer can catch that channel-specific tools miss: quality signals. CAC can look stable while activation rate declines if you are acquiring cheaper but lower-quality users. The cross-funnel view catches this; a campaign-level view does not.
Planning. Given the monitoring output, the agent reasons about where the highest-leverage intervention is. Is the constraint in acquisition volume? Activation rates across all channels? Retention after paying? This localization is the most valuable thing a growth agent does, helping you focus on the actual bottleneck rather than optimizing channels that are not the limiting factor. Proposals are grounded in the experiment history and outcome data in memory, so they are more likely to be genuinely new ideas than things you have already tried.
Action. When the diagnosis is an acquisition problem, the action layer executes campaign changes via platform APIs (Google Ads mutate with Standard Access, Meta Marketing API with ads_management and admin/advertiser role, LinkedIn with rw_ads scope), same as a channel-specific agent. When the diagnosis is an activation or retention problem, the agent surfaces this for the product or CRM team with the data to support it. It cannot fix a product issue by adjusting bids. Knowing which type of intervention is needed is itself the valuable output. The growth agent earns its keep by correctly attributing problems to their actual cause, whether or not that cause is something the agent can directly execute on.
How to set it up in Roadway
- Create a new Coworker
- Filter for the channel
- Choose your goal metric (this is what your agent will optimize for)
- Choose the funnel metrics that lead to your goal metric
- Choose your guardrail metrics and define their limits
- Choose your refresh schedule
- Publish
Work with AI Coworker to plan and execute campaigns. Reach out to us if you need any help - happy building: contact@roadwayai.com
FAQ
What is the difference between a growth marketing agent and a performance marketing agent?
A performance marketing agent optimizes paid channels for efficiency: CAC, ROAS, budget allocation. A growth marketing agent operates at a higher level. It joins attribution data with product analytics and revenue data to evaluate customer quality, not just acquisition cost. It watches activation rates, retention, and LTV by acquisition source. A campaign with a great CAC but low activation rate is buying the wrong customers. The growth agent catches this. The performance agent does not.
What data does a growth marketing AI agent need?
Three datasets joined by user ID. Acquisition data: which channel and campaign each user came from, measured consistently across all channels. Product analytics data: activation milestones, engagement patterns, feature usage, churn signals. Revenue data: subscription events, upgrades, downgrades, LTV at various time horizons. When these are joined, the agent can calculate metrics that no single-channel tool can see: LTV by campaign, retention by acquisition source, activation rate by channel.
How does an AI agent detect customer quality problems?
By comparing acquisition metrics to downstream outcomes. If CAC is stable but activation rate is declining, the agent is acquiring cheaper but lower-quality users. If one channel has strong CAC but poor 30-day retention, that channel is producing customers who do not stick. These quality signals are invisible to a campaign-level view. The growth agent sees them because it connects acquisition data to product and revenue data.
Can an AI agent fix activation or retention problems?
Not directly. If the diagnosis is an acquisition problem (wrong audience, wrong channel, wrong messaging), the agent can fix it by adjusting campaigns. If the diagnosis is an activation or retention problem (product friction, onboarding gaps, feature issues), the agent surfaces this to the product or CRM team with the supporting data. Knowing which type of problem it is, and correctly attributing it to the right cause, is itself the most valuable output.
What are quality guardrails and why do they matter?
Quality guardrails are constraints beyond standard performance metrics. Examples: minimum activation rate for a campaign to stay active, minimum 30-day retention for a cohort before the acquisition channel is flagged, minimum LTV-to-CAC ratio by channel. These prevent the agent from scaling campaigns that look efficient on the surface but produce customers who churn. Without quality guardrails, the agent optimizes for volume and cost, which can actively harm the business if it attracts the wrong customers.