A support chatbot is the wrong first AI project
Almost every business asks me for a chatbot first. It is usually the riskiest place to start. The best first AI project is one your customers never see.
A chatbot sits front of house. It talks to strangers, in your name, on its worst day as well as its best. Getting one to be genuinely safe and useful takes guardrails, testing and content work that most small businesses have not done yet.
Meanwhile, behind the counter, there is repetitive work nobody is emotionally attached to: retyping form submissions into a CRM, drafting the same quote email, summarising enquiries, chasing no-replies. That work is invisible, measurable and cheap to automate.
Start where failure is cheap
- Intake summaries: every enquiry arrives pre-read, scored and filed.
- Quote drafts: the response email is 80% written before you open it.
- Follow-ups: polite nudges go out on schedule without you remembering.
If an internal automation writes a clumsy draft, you fix it before anyone sees. If a chatbot invents an answer about your refund policy, a customer sees it first. Same technology, completely different blast radius.
Automate the work behind the counter before you hire a robot for the front desk.
Once the boring automations have paid for themselves, and taught you how the tools behave, a chatbot trained on your real pages and policies becomes a sensible second step instead of a gamble.