Should a small service business add an AI chat assistant? A practical checklist

AI assistants have become easy to add and easy to get wrong. A good one feels like a helpful front-desk person who knows your business. A bad one invents prices, frustrates visitors, and quietly loses leads. Here’s how to tell whether one makes sense for you.

Signs an assistant will help

• You get steady website traffic. An assistant helps visitors who are already there. It doesn’t create traffic.
• Inquiries arrive when you can’t answer. Evenings, weekends, and busy job days are when fast responses are hardest.
• The same questions repeat. Service area, what’s included, how quoting works, how to prepare, what happens next.
• Leads arrive incomplete. If every inquiry starts with a round of “what size is the home?”, an assistant can collect that up front.

Signs it won’t, yet

• Your website gets very few visitors. Fix visibility and the website first.
• Your key answers change daily and can’t be written down (for example, live pricing you haven’t systematized).
• Nobody will follow up on the leads it collects. The assistant starts the conversation; your team still has to finish it.

What a good assistant should do

1. Answer common questions using only information you’ve approved.
2. Help visitors pick the right service or next step.
3. Ask a small number of qualifying questions, conversationally.
4. Say clearly when it doesn’t know, and offer a person.
5. Let the visitor review their details and choose to send them.
6. Deliver the lead where your team will actually see it: email, CRM, or shared inbox.

What it should never do

• Quote prices, discounts, or timelines you haven’t approved.
• Tell someone they’re booked when nothing has been confirmed.
• Make guarantees or claim results.
• Give legal, tax, financial, or medical advice.
• Collect sensitive information it doesn’t need.

We cover how to enforce these in how to keep an AI assistant honest.

Questions to ask any provider

• Where do the assistant’s answers come from, and who approves them?
• What happens when it doesn’t know something?
• How does it hand leads to my team, and can I test that end to end?
• Does the visitor see and approve what’s sent?
• What are the running costs, and what stops a surprise bill if someone abuses it?
• Where are conversations stored, for how long, and who can see them?
• Can I review conversations to improve answers?
• What happens if the AI provider has an outage?

What it costs, in general terms

There are two parts: building it (setup, knowledge writing, integration, testing) and running it (mostly the AI provider’s usage fees, which scale with conversations and the model used). A well-built assistant includes limits so costs stay predictable. Ask for an estimate based on your actual traffic rather than a generic figure.