An AI chatbot for tradespeople is only useful if it knows where its job ends. Customers of an electrician, plumber or joiner usually ask three things: will you come out, how much will it cost and when. The assistant should answer none of them alone. It can explain where you work, how a job runs, what to prepare and how to report an emergency, and sort the rest for a person.
In this guide you split customer questions into those the assistant handles from your knowledge base and those it passes to the tradesperson. Then you decide what happens after hours and check the limits with a test before going live. The examples use a fictional plumbing and heating firm, "Tap and Valve", with two fitters.
Which questions an AI assistant can answer for a tradesperson
A good question for the assistant has an answer you can write once and stand behind for every customer. Your service area, how a job runs, which photos to prepare and the difference between an emergency and planned work all qualify. "How much to replace a mixer tap?" does not, because the answer depends on an installation nobody has seen yet.
Scroll the table sideways to see all columns.
| Customer question | Who answers | Source or next step |
|---|---|---|
| Do you cover my town? | Assistant | List of towns and districts in the knowledge base |
| How does a job work from request to sign-off? | Assistant | Stages: contact, site visit, quote, work |
| What should I prepare before we talk? | Assistant | List of photos and details, such as the age of the system |
| It is leaking under my sink, what now? | Assistant, then a person | Emergency route and the number for urgent jobs |
| How much is a new radiator? | Person | Handoff to the team with the customer's consent |
| Can you come on Thursday? | Person | The tradesperson checks the schedule and replies |
| Is it the valve or the pump? | Person | Diagnosis only after a site visit |
The source column matters most. If you cannot point to the paragraph in your knowledge base the answer should come from, the question is not ready for automation. The assistant does not take phone calls either: calls to your number and voice calls from the website are still handled by people. For calls lost while you are on a job, see the guide to missed calls from customers.
Handwerksblatt (November 2024) reports on the KI-Index Handwerk.NRW survey of more than 1,200 German trades businesses: about a third use or have tried AI, and voice assistants and chatbots are among the most common uses. The survey covers North Rhine-Westphalia only and says nothing about the quality of the answers these tools give.
Prices, dates and diagnosis stay with a person
A knowledge-based assistant informs but does not act. It cannot see your fitters' schedule, so it cannot confirm a date, and it does not create quotes. Trouble starts when the knowledge base contains sentences that sound like decisions, such as "replacing a tap usually takes about an hour". The assistant will pass these on as promises, so state plainly who deals with such matters.
Prices: "We give a price after a call or a site visit, because it depends on the installation. Describe the work and the tradesperson will reply." Dates: "Available slots are checked by the tradesperson. Tell us when you are at home." Diagnosis: "We do not diagnose faults over chat. Describe the symptoms and we will agree the next step when we talk."
These sentences tell the customer what to do next. How to reply to a bare "how much is it?" before you know the details is covered in the guide to price questions in chat. The same rule applies to words such as "booked" or "confirmed": when you may use them is explained in the guide on when a chatbot can say it is done.
Emergency or planned work, and what about photos
The first sort the assistant can do well is emergency versus planned work. Describe both in the knowledge base with simple examples: a flood, no heating in winter or a smell of gas are urgent, while a new bathroom or replacing radiators is planned work. For an urgent case the assistant should give a phone number or the right emergency service, not repair tips. For a smell of gas the only correct step is the gas emergency number, which you add to the knowledge base word for word.
Photos help the tradesperson, but the assistant does not look at them. In Meet AI a message with an attachment sent to the assistant goes to the team instead of being judged by the AI. The knowledge base only needs to say which photos to take: the whole appliance, the rating plate and a close-up of the leak. The customer sends them in the conversation with a person. What to ask before you quote from such photos is covered in the guide to quoting from customer photos.
After-hours enquiries: hybrid mode and callbacks
In Meet AI you choose an operating mode. In Hybrid mode the "Ask the AI assistant" card appears in the widget only when the Availability module shows that nobody on the team is available, so while someone on the team is available, customers write to people. In Self-service mode customers can choose the assistant at any time. A conversation with the AI starts only after the customer picks the card, and the assistant's messages carry an AI badge. Schedules and holidays in hybrid mode are covered in the guide to an AI assistant after business hours.
When the knowledge base has no answer, the assistant says it has no information and offers to pass the question to the team. The handoff happens only with the customer's consent, and when nobody is available the assistant does not promise contact. That is why a callback request helps after hours: if you allow it outside working hours, the customer picks a slot from your schedule and a topic from your list, such as "emergency", "quote" or "site visit". In the morning you see sorted requests instead of late-night questions with no context. The mechanism is explained in the guide to website callbacks.
The European Commission states that people using a chatbot should be made aware they are interacting with a machine, and gives August 2026 as the start of the AI Act transparency rules. The AI Act is an EU regulation, so outside the EU check the rules where you work. The page does not say how a trades firm should word its assistant's greeting. This article is not legal advice.
Test the limits before going live
Before a customer sees the assistant card, ask it questions in a private test in the panel. The test has no public effect, and the result shows citations of the sources the assistant used, or that it found nothing in the knowledge base. Use a dozen questions worded like your customers write, typos included.
- Questions from the table.Ask every question marked "assistant" and compare the answer with the knowledge base sentence by sentence.
- Price and date questions.Ask "how much for a new boiler?" and "can you come tomorrow?". The right result is a referral to the tradesperson, with no amount and no time.
- A diagnosis question.Describe a symptom, such as knocking in the pipes. The assistant should not guess the cause or suggest a DIY repair.
- An emergency and an unknown topic.Ask about a flood and about a service you have not described. The first answer must give the emergency route, the second must admit it has no information.
Repeat the tests after every change to your service list, area or emergency number. The knowledge base does not update itself: a changed document has to be added again or the sources refreshed manually, which is possible once every 7 days. For more check questions, see the guide on testing a chatbot before launch.
Questions from tradespeople
Can the assistant book a customer in for a site visit? No. The Meet AI assistant answers from knowledge and does not record appointments. If you want customers to choose a site visit themselves, you need a separate calendar with manual confirmation, as described in the guide to online booking for trade businesses. The assistant can then only explain how that booking works.
For a one-person business, is Hybrid or Self-service better? If you reply from the mobile app during the day, even briefly, Hybrid leaves those conversations to you and switches the assistant on when nobody is around. Self-service makes sense when you spend most of the day on site and cannot type. In both modes the same knowledge base sets the limits.
