Multi-Agent AI Platform for a large marketing agency

Bringing a multi-agent AI platform to a large marketing communications agency

Together with strategic partner DataExpert, MADIS is adapting an AI-powered media strategy platform covering the full workflow from client briefing to strategy and execution.

At a glance

~10

AI agents in the platform

End-to-end

from client brief to strategy and project plan

3

parallel workstreams: AI, DevOps, data

Summary: a large marketing communications agency wants to equip its media analysts and strategists with an AI-powered platform. With partner DataExpert, MADIS is bringing a system of around ten AI agents, customising them, orchestrating the pipelines and integrating data services.

The client

A large marketing communications agency with global and local structure at the same time, makes it a particularly interesting environment for an AI project.

The challenge

Media planning involves many steps: briefing with the client, market research, strategy development and project planning. The client's goal is to turn each of these steps into an automated, AI-enriched process for its analysts and strategists.

The UK team of the group had already developed and operates a system of around ten AI agents. The task was not to start from scratch, but to bring this system to Ireland: reuse the agent code, implement it for the Irish context and connect it to Irish data services.

Our approach

We work in three areas in parallel, so that the platform can be shown early and refined continuously.

AI engineering

Customising the agents and prompt engineering for the Irish market.

AIOps and DevOps

Building the pipelines and orchestrating the agents on Google Cloud.

Data engineering

Integrating Irish data services through ETL pipelines.

The solution

The platform is a multi-agent AI system on Google Cloud Platform that supports the client’s media analysts and strategists across their full workflow, from the first client briefing through market research to strategy creation and execution.

A key milestone was an MVP that the client could present to its own clients. At that stage it still worked with test data and did not yet cover all agents, while data integration and agent customisation continued in parallel.

Technologies: Python · Google Agent SDK · React · GCP (Cloud Run, Vertex AI, BigQuery) · Terraform · Docker

Results

• Reuse instead of rebuild: proven agent code from the UK is reused and adapted, rather than developed from zero.
• Early value: an MVP built for presentation to client’s clients while the full rollout continues.
• Built for the local market: agents and data sources tailored to Ireland.

Facing a similar challenge?

Exploring AI agents for your teams’ everyday work? Talk to our AI engineering team.