Marketing AI Agent
Developed an AI marketing agent that auto-generates ad copy, monitors campaign performance, and optimises budgets daily — reducing campaign management time by 70%.
The Challenge
A performance marketing agency managing 15+ client ad accounts was spending 3–4 hours daily per account on ad copy refreshes, budget reallocation, and performance reporting. Scaling the agency meant hiring more people — an expensive and slow solution.
What We Built
We built a multi-agent marketing system using GPT-4o and n8n. Agent 1 monitors campaign metrics via Meta and Google Ads APIs daily, flags underperforming ad sets, and proposes budget reallocation. Agent 2 generates new ad copy variants based on performance data and brand guidelines. Agent 3 compiles a daily performance briefing sent to the account manager via WhatsApp. Approved changes are applied via the Ads APIs automatically.
How It Works
The agency's analysts were talented strategists trapped in a cycle of manual work. Checking metrics across 15 accounts, writing ad copy, reallocating budgets, and building client reports consumed 6–8 hours of their day — leaving no time for the strategic thinking that actually drives results.
The AI agent system runs on a daily cycle. At 6am, Agent 1 pulls the last 24 hours of performance data from Meta and Google Ads APIs for all accounts. It calculates performance deltas, identifies ad sets where CPC has risen more than 15% or ROAS has dropped below target, and generates a prioritised action list.
Agent 2 takes the flagged ad sets and uses brand guidelines plus historical top-performing copy patterns to generate 3–5 new ad copy variants for each. It writes headlines, primary text, and descriptions — formatted correctly for each platform's character limits.
The account manager receives a WhatsApp summary each morning: which accounts need attention, what actions are recommended, and the new copy ready for review. One-tap approval triggers the actual API calls — budget reallocation and new ad set creation happen automatically.
Client reporting, previously a 2-hour weekly task, is now automated: a Google Slides deck with performance charts and commentary is generated every Monday and emailed to clients without analyst involvement.
