The challenge
An advertising platform with around 200 active clients handled support and onboarding by hand. Every new signup needed a manager to walk them through setup, and the same questions arrived again and again. Product knowledge lived in scattered documents and in the heads of the team, so even a simple answer took time to find.
The approach
AI agents answer support and onboarding questions in the channels clients already use. Before answering, an agent pulls two kinds of context: the company’s internal knowledge base and the client’s own record in the CRM. The knowledge base makes the answer correct, the CRM data makes it specific. A question about campaign settings gets an answer that matches how that particular client’s account is configured.
Architecture
- AI: OpenAI API, with RAG over PostgreSQL and pgvector
- Orchestration: n8n workflows drive retrieval, CRM lookups, and response generation
The knowledge base is embedded and stored in pgvector, and n8n wires the agents into the support channels the company already runs.
Results
Repetitive first-line questions get answered by the agents, with each client’s setup taken into account. The team’s time goes to the questions that actually need a human.