Work
Entry 0072026—06AutomationShipped

WhatsApp Agent for a Children's Sports Club

A full automated responder for parent enquiries: pricing, schedules, trial-class booking and complaints — with no human in the loop. The agent reads and writes to live sheets, and alerts the owner silently on failure rather than showing the customer a technical error.

01

The problem

Parent enquiries arrive on WhatsApp all day: what does it cost, when are the sessions, I want to book a trial, I have a complaint. Answering meant somebody sitting on a phone — and whoever waited too long went to another club.

02

The approach

Not a numbered-menu autoresponder, but an agent that understands the question and acts. I gave it a fixed set of tools to read and write the club's live sheets, and constrained it with a prompt that separates the two pricing stages and forbids inventing links, because the model tends to invent them.

03

What I built

An n8n agent on GPT-4o with six tools: club info, pricing, trial booking, complaints, conversation log, and available slots. Error handling is deliberate: on agent failure it retries three times, and if that fails the owner is alerted silently — the customer never sees a technical error. The current date is injected into every message context, because the model has no notion of it and any "today"/"tomorrow" logic fails silently without it.

04

The result

The system answers, logs and alerts with no human in the loop. Operating figures have not yet been pulled from the conversation log — and until they are, no number is shown here.

Stack
n8n · GPT-4o · Google Sheets · WAHA · Docker
Role
Built end to end — from the server to the conversation script
Disciplines
Automation
Client
Anonymised by agreement
Lesson

The model has no notion of today's date by default — any logic depending on "today" or "tomorrow" fails silently unless the date is injected explicitly into the message context.