How We Built the Growth OS: AI Agents for Marketing

Why Did We Build Growth OS?

Every growth marketing agency faces the same tension: clients want sophisticated analysis and strategic recommendations, but 60-70% of analyst time is consumed by repetitive data tasks — pulling reports, formatting spreadsheets, identifying anomalies, and writing performance summaries. These tasks require expertise to interpret but not to execute. They're the perfect candidate for AI automation.

Growth OS started as an internal project to solve this problem. The goal was straightforward: build AI agents that handle the analytical grunt work so our team can focus on the strategic thinking that actually moves the needle for clients.

What Is a Multi-Agent System?

A multi-agent system is an architecture where multiple specialised AI agents collaborate to accomplish complex tasks. Rather than building one monolithic AI that tries to do everything, each agent has a narrow expertise area and well-defined inputs and outputs. They communicate through structured messages and can be orchestrated by a coordinator agent or triggered independently.

The multi-agent approach has several advantages over a single-agent design:

• Specialisation: Each agent has a focused prompt, toolset, and context window optimised for its specific task. A Google Ads analysis agent doesn't need to understand content strategy, and vice versa.
• Reliability: If one agent fails, others continue working. A single-agent system is all-or-nothing.
• Iterability: You can improve or replace individual agents without affecting the rest of the system.
• Cost efficiency: Smaller, focused agents use fewer tokens and can use faster, cheaper models for routine tasks while reserving more capable models for complex analysis.

What Agents Does Growth OS Include?

Google Ads Analysis Agent

The Google Ads agent connects to the Google Ads API via a custom MCP (Model Context Protocol) server. On a scheduled basis — or on demand — it pulls campaign performance data, analyses trends against historical baselines, identifies anomalies, and generates actionable recommendations.

The agent examines metrics at the campaign, ad group, keyword, and search term level. It identifies:

• Budget pacing issues — campaigns that are overspending or underspending against targets
• Keyword performance outliers — high-spend/low-conversion keywords that should be paused or adjusted
• Search term opportunities — queries triggering ads that should be added as keywords or excluded
• Creative fatigue signals — declining CTR trends that indicate ad copy needs refreshing
• Bidding strategy recommendations — when to transition between manual and automated bidding based on conversion volume

Content Performance Agent

The content agent analyses website and blog performance through GA4 data. It identifies which content is driving organic traffic growth, which pages have declining engagement, and where content gaps exist relative to keyword targets.

Weekly, it produces a content performance brief that includes top-performing pages, pages with ranking regression, internal linking opportunities, and content refresh priorities — all based on data rather than intuition.

Reporting Agent

The reporting agent automates the creation of client-facing performance reports. It pulls data from multiple sources (Google Ads, GA4, HubSpot, social platforms), structures it against the client's KPI framework, adds trend analysis and commentary, and outputs a formatted report ready for review.

What previously took an analyst 3-4 hours per client now takes 15 minutes of review time. The agent generates the draft; a human reviews, adds strategic context, and approves.

How Is Growth OS Architected?

Growth OS is built on a surprisingly simple technology stack:

• Agent framework: Claude API (Anthropic) with tool use for structured interactions with external APIs
• MCP servers: Custom Model Context Protocol servers that give agents secure, scoped access to Google Ads, GA4, and other platforms
• Orchestration: A Next.js application that schedules agent runs, manages state, and provides a human review interface
• Storage: Supabase (PostgreSQL) for agent outputs, historical data, and audit logs
• Frontend: React dashboard for reviewing agent outputs, approving recommendations, and tracking implementation

The architecture follows a key principle: agents propose, humans approve. Every recommendation — whether it's pausing a keyword, adjusting a budget, or flagging a content issue — requires human review before action. AI handles the analysis; humans make the decisions.

What Have We Learned?

Building Growth OS taught us several important lessons about applying AI to marketing operations:

Start with the Boring Tasks

The most valuable AI applications automate tasks that are necessary but tedious — not tasks that are creative or strategic. Report compilation, anomaly detection, and data formatting are perfect. Campaign strategy and creative ideation are not (yet).

Prompt Engineering Is Product Design

The quality of agent output depends almost entirely on prompt design. We spent more time refining prompts than writing code. Each agent's prompt includes specific personas, output formats, quality criteria, and examples. Treating prompts as product specifications — with versioning, testing, and iteration — was the key insight.

Human-in-the-Loop Is Non-Negotiable

We initially experimented with fully automated actions (auto-pausing keywords, auto-adjusting budgets). This was a mistake. AI models occasionally misinterpret data or miss context that a human would catch. The approval step adds 15 minutes of human time but prevents potentially costly errors. Every autonomous marketing AI system should have this guardrail.

MCP Changes Everything

The Model Context Protocol (MCP) was transformative for our architecture. Instead of building custom API integrations into each agent, MCP provides a standardised way for agents to discover and use external tools. Adding a new data source — say, LinkedIn Ads — means building one MCP server that all agents can use, rather than integrating it into each agent individually.

What's Next for Growth OS?

We're actively developing additional agents for social media performance analysis, competitive intelligence, and predictive budget modelling. The platform is currently used internally at Ikigai, with plans to offer it as a managed service for clients who want AI-augmented marketing operations.

The vision is straightforward: give every marketing team access to the analytical depth of a large agency, powered by AI agents that work around the clock. The future of marketing operations isn't replacing humans with AI — it's amplifying human strategic thinking with AI-powered analysis.