AI Sales Assistant

The problem

Sales enquiries come in at all hours. When the first reply waits until the next working day, warm leads go cold, and the details of each conversation end up scattered across inboxes.

What we built

A published n8n workflow that receives each enquiry through a webhook and runs it through an AI agent built on the Google Gemini chat model:

1. Receive and validate. The webhook normalises the request and checks it. Invalid requests get a clear error response instead of reaching the agent.
2. Answer with real data. The AI agent replies using a services tool that reads the current service list from a Google Sheet, with short-term memory so follow-up questions keep their context.
3. Structure the result. The agent's answer is parsed into structured fields.
4. Log everything. The full conversation is appended to a log sheet, and a lead record is prepared and upserted, so a returning contact updates their existing row instead of creating a duplicate.
5. Respond. The reply goes back to the website or chat channel that sent the enquiry.

Why it is built this way

• Validation before the AI step keeps junk requests away from the model.
• The agent answers from a sheet the team already edits, so changing an offer needs no developer.
• Every conversation is logged, so the team can review what the AI said.
• The same workflow can sit behind a website form, a chat widget or a WhatsApp number.