AI Phone Assistant for Trades: Reachable Without Office Staff

The phone rings on site. Your hands are in the pipe insulation, the caller waits three rings and then calls the next company. That is daily life in many trade businesses, and it costs jobs. How many exactly, nobody can say credibly. That is what this article is about: measure first, then decide.

Key takeaway: The figures circulating online about missed calls in the trades (30, 35, 62 percent) come exclusively from vendors of phone assistants and cannot be traced to any independent survey. What is solid is the surrounding situation: according to the ZDH economic report, order reach has fallen to 8.3 weeks and the share of businesses reporting a poor order situation has risen from 14 to 18 percent. Anyone losing calls in that situation is losing revenue they need. Measure your own rate for two weeks before investing in any solution.

Why does the phone become a bottleneck in the trades?

Because two developments coincide and reinforce each other.

The first is the shortage of staff. According to the German Confederation of Skilled Crafts (ZDH), around 250,000 qualified tradespeople are missing nationwide, about 70 percent of businesses are hiring, on average 1.4 skilled workers per business. Electrical, plumbing and heating, and metal and welding trades are hit hardest. Office staff to answer the phone simply cannot be recruited in many businesses, and where they exist they are needed first for quotes and invoicing.

The second is the order situation. The ZDH economic report shows an order backlog indicator of minus 13 points, capacity utilisation at 79 percent and order reach of 8.3 weeks. The share of businesses with a poor order situation has risen from 14 to 18 percent.

That is the decisive shift: three years ago a missed call was annoying but bearable because order books were full anyway. At a reach of eight weeks, every unanswered call is a job that goes to a competitor.

How many calls are really lost?

Nobody knows. And that is the most honest answer to this question.

Search online and you find a remarkable range: 30 percent, 35 percent, 40 to 60 percent, 62 percent. All of these figures come from companies selling phone assistants. No independent survey exists that they could be traced back to. We sell such systems too, which is why we say it plainly: do not use any of these numbers as the basis for an investment decision, including ours.

The same applies to the revenue claims. Blanket figures such as "every missed call costs 200 to 1,200 euros" mean something entirely different for a business servicing boilers than for one renovating bathrooms. A boiler service job and a full bathroom renovation differ by a factor of fifty.

What helps instead: your own numbers.

How do I measure my own rate in two weeks?

You need no system and no software for this, just your phone's call list and a sheet of paper.

1. Document for two weeks. Note daily how many calls came in and how many were answered. Most phone systems and every smartphone already keep this list.
2. Go through the missed calls. Call back where you have not already, and record the outcome in three categories: existing customer with a request, new customer enquiry, other (advertising, suppliers, wrong number).
3. Determine the share of new customer enquiries. Only this group is the actual loss, because existing customers usually call again.
4. Multiply by your average job value. Use the figure you know, not one from a blog post. And apply a realistic conversion rate: not every enquiry becomes a job.

The result is a number that applies to your business. It may be considerably lower than the values circulating online, and in businesses with a lot of emergency call-outs, higher.

A side effect of those two weeks: you see when the calls come in. If 80 percent cluster between 7 and 9 am, office staff in the morning may be the better answer than any technology.

What can an AI phone assistant take over in the trades?

Whatever is structured and repeatable. In practice that is four things:

The practical benefit lies less in the spectacular case than in reliability: there is no longer a call that nobody knows about.

What can an AI phone assistant not do?

Just as important, and this is where the industry promises a great deal.

A phone assistant cannot calculate quotes. It can record that someone wants a bathroom renovated, but not say what that costs. Anyone promising otherwise should explain where the prices are supposed to come from.

It cannot give technical advice. Whether a heat pump makes sense in a building from 1962 is a question for the master craftsman, not a machine.

It does not replace office staff. Writing quotes, issuing invoices, chasing payments, ordering materials: all of that still waits. An assistant takes over exactly one task, namely answering and sorting incoming calls.

And it works only as well as its instructions. If it is not recorded that you do not take on electrical work in commercial properties, it will capture such enquiries and you will have callbacks that lead nowhere.

How does the rollout work in a trade business?

In four steps, spread over a few weeks:

1. Collect the requests. Which calls typically come in? From your two weeks of measuring you already have the list.
2. Set the rules. Which appointments may the assistant book itself, which not? What counts as an emergency? At what point is a call put through, at what point only noted?
3. Connect the systems. Calendar and, if in use, trade software or CRM. Without that connection everything ends up in a separate list nobody reads.
4. Trial run. First alongside the existing phone, then as the first point of contact. In the early weeks, spot-check calls.

The most demanding part is step two, and it is pure thinking work inside the business: who gets which appointment, and what is genuinely urgent?

What applies legally?

Four points, and the first is the most frequently overlooked because it comes from criminal law rather than data protection.

Recording only with consent (Section 201 of the German Criminal Code). Anyone who records the non-publicly spoken word of another without their consent commits an offence under Section 201(1) StGB. That applies to businesses too, and it applies even when the recording only serves quality assurance. A phone assistant that records conversations for later processing therefore needs consent at the start of the call, and the caller must be able to refuse it without losing the ability to raise their request.

The clean solution is usually not to record at all: an assistant that captures the request as text and discards the audio avoids the problem entirely. Anyone who does want to record needs an explicit prompt and solid documentation of the answer.

Disclosure duty (Article 50 EU AI Act). Since 2 August 2026 the caller must learn at first contact that they are speaking to a machine. A note on the website or in the legal notice is not enough; the information must come in the conversation itself. In practice that means a greeting clarifying the nature of the counterpart, for example: "Hello, this is the digital assistant of Muster Plumbing. I can book appointments and take down your request. For anything else I will connect you with the team."

Systems already running before the cut-off date have a retrofit deadline until 2 December 2026. The details are in our article on the AI disclosure duty.

Data protection (GDPR Articles 6, 13 and 28). Name, phone number and the request are personal data. For an appointment enquiry the legal basis is regularly Article 6(1)(b), because the processing serves steps prior to entering into a contract. The caller must be informed under Article 13, which can be combined with the Article 50 disclosure in a single announcement. And because the assistant runs at a service provider, a data processing agreement under Article 28 is required, covering the server location as well.

Binding commitments (Sections 145 et seq. and Section 164 German Civil Code). This is the point most likely to cost a trade business money. Whatever the assistant declares on behalf of the business is attributed to the business. If it names a price, that can become a binding offer; if it confirms a slot, the appointment counts as agreed. The configuration therefore needs a hard boundary: the assistant records requests and books slots within the approved range, but it names no prices and makes no commitments on scope. A sentence like "that will be around 800 euros" must not be technically possible in the first place.

Where a business carries additional professional secrecy duties, the question of where the AI model itself may run is covered in our article on LLM hosting under Section 203 StGB.

Measure first, then decide

It makes sense to start with measuring, not with technology. If the rate of missed calls is low, the project does not pay off. If it is high enough, the assistant should be tailored to your own workflows: appointment rules, definition of an emergency, calendar and software connections. The Article 50 disclosure belongs in the configuration rather than being added afterwards, and every call the assistant cannot handle cleanly should go to the team with the full conversation history.

Frequently asked questions about AI phone assistants in the trades

Do customers notice they are speaking to a machine? Yes, and since August 2026 they must also be told. Article 50 of the EU AI Act requires the information at first contact. Practical experience shows this is not a problem as long as the route to a human stays open at all times. What annoys customers is not the machine but the dead end.

Does it work with site noise or poor reception? The assistant answers precisely when you cannot, so the question runs the other way: the caller is calling from their own environment, which is usually quiet. Strong dialects and very fast speech are harder; both can be checked in the trial run before the system goes live.

What happens in a real emergency, such as a burst pipe? That is a configuration matter and one of the most important things to settle before launch. Typically the assistant recognises certain keywords as an emergency and then puts the call straight through or triggers a call chain. Without that rule it would treat the burst pipe like a quote enquiry.

Do I need a new phone system for this? Usually not. The common setup is call forwarding: your existing number stays, calls are forwarded to the assistant after a set number of rings or outside working hours. A modern phone system allows finer control but is not required.

Is it worth it for a two-person business? That depends on your measured rate and your average job value, not on company size. A two-person business renovating bathrooms loses more per missed new-customer call than a larger business with many small jobs. That is why measuring comes first.

What about my customers' data? Processing call data falls under the GDPR as usual: processing in the EU, a processing agreement with the provider, clear deletion periods. If calls are recorded or transcribed, separate information for the caller is required; that is a distinct obligation and not covered by the AI disclosure.

May the phone assistant record calls? Only with the caller's consent. Section 201(1) of the German Criminal Code makes recording the non-publicly spoken word without consent a criminal offence, and that includes internal quality assurance. Anyone wanting to record needs an explicit prompt at the start and must document the answer. It is usually simpler not to record at all: an assistant that captures the request as text and discards the audio avoids the question entirely.

Is an appointment binding if the AI confirmed it? Yes, if the assistant acts on behalf of the business. What it declares is attributed to the business under the principles of agency, so a confirmed slot counts as agreed. For the same reason the assistant must not name prices, because that can become a binding offer under Section 145 of the German Civil Code. The configuration has to enforce that boundary technically, not merely mention it in a conversation guide.

Conclusion

The question is not whether an AI phone assistant works in the trades. It does, for the tasks that are structured. The question is whether the effort pays off for your business, and that cannot be answered with a figure from a blog post, including this one.

Two weeks of call logs cost you nothing but a little discipline and give you the basis every vendor really owes you. If it then turns out that new customer enquiries are regularly lost, the investment is quickly justified. If not, you have saved yourself a project.

If the decision goes in favour of the assistant, the AI phone assistant page describes how phone number, calendar and the handover to the business fit together.

Sources

• ZDH, German Confederation of Skilled Crafts: Economic Report 1/2026 (order backlog indicator, capacity utilisation, order reach)
• ZDH: Securing skilled labour in the trades (around 250,000 missing skilled workers, 70 percent of businesses with vacancies)
• DIHK Skilled Labour Report 2025/2026 (36 percent of around 22,000 companies surveyed cannot fill vacancies)
• Regulation (EU) 2024/1689 (EU AI Act), Article 50 (transparency obligations)
• Section 201 StGB - violation of the confidentiality of the spoken word
• Sections 145 et seq. German Civil Code (binding effect of an offer) and Section 164 (effect of agency)
• Regulation (EU) 2016/679 (GDPR), Articles 6, 13 and 28

This article was published on 8 August 2026 and extended on 11 September 2026 with the legal requirements on recording, data protection and the binding effect of commitments. It does not constitute legal advice.