According to industry association estimates, German bike workshops are short of around 7,500 skilled workers. You won’t close that gap before next season. What you can close are the hours your existing mechanics spend on the phone. Anyone who calls the workshop in spring either reaches nobody or reaches the one person in the business who should be diagnosing an e-bike motor right now. Both cost money, and the second also costs the team’s good mood.
7,500 missing skilled workers: the workshop bottleneck in figures
A Hamburg bike shop in March 2026: three people work in the workshop, on average they get through 18 bikes a day, and yet customers wait at least four weeks for an appointment. That is how a tagesschau report of 29 March 2026 describes it. According to the report, some shops in the trade now only repair bikes they have sold themselves.
The same report supplies the figure behind the scene: according to industry association estimates, there is a nationwide shortage of around 7,500 skilled workers in bike workshops. Only about 750 are trained each year. So this gap will not be closed by hiring.
At the same time, the workload is growing. According to the market data of the Zweirad-Industrie-Verband (ZIV, German bicycle industry association) of 11 March 2026, there are 90.6 million bicycles and e-bikes in use in Germany, 2.1% more than the year before. Around 17.2 million of them are e-bikes, by the ZIV’s calculation adjusted for scrapped vehicles, and almost ten million have been added since 2020. Every one of these e-bikes has electronics, software and a battery, and hardly any customer repairs those themselves any more.
What this means for you is spelled out by the ZIV in the same tagesschau report, in response to an NDR enquiry: a bike you have to wait weeks to get repaired is more likely to be left standing afterwards. A Shimano study quoted there points the same way: a quarter of the cyclists surveyed in Germany ride less because of repair hurdles. So anyone who waits too long for an appointment with you rides less, and anyone who rides less needs your workshop less often.
A worked example shows how much of this is lost on the phone. The assumptions are set, not measured: 40 calls on a peak-season day, three minutes each. That is two hours, spread over 40 interruptions. At six bikes per person per day, as in the Hamburg workshop, that works out to a good one bike left waiting. The minutes it takes after every call to get back into the job are not even included.
What an AI agent takes over in a bike workshop and what it doesn’t
Technically, an AI agent on the workshop phone is a voice agent: speech recognition, a language model with a clear brief, speech output, and behind them tools that allow it to read and write appointments. The architecture is the same as for other AI agents as digital employees. The difference lies in the scope: few tasks, fixed rules.
It takes over everything that follows a fixed process:
- answering calls, including at 7 pm and on Saturdays
- asking about the type of bike, the problem and the preferred time, and placing the appointment in a free slot in the workshop schedule
- scheduling first inspections and annual checks, which follow a standard routine anyway
- taking callback requests when the team is fully booked
- checking spare part availability, if the inventory management system allows it
- sending a confirmation by text message or email
And the question “Is my bike ready?”. It needs no expertise, just a look at the work order, and yet every time it interrupts someone who has a tool in their hand. If you only want to hand over one task, make it this one.
What it doesn’t do is hear a creak in the bottom bracket or diagnose a motor fault. It doesn’t commit to a quote that hasn’t been stored beforehand as a flat rate. It doesn’t negotiate warranty cases and doesn’t decide whether a complaint is justified.
Example scenario: a customer calls on Tuesday evening because the display of their e-bike shows an error code. The agent asks for the brand, drive system and purchase date, offers the next free diagnostic appointment, books it and notes in the work order “Error code on display, customer reads out code: …”. On Wednesday morning, the mechanic has the case, code included, on the tablet before even seeing the bike.
The hand-off to staff belongs in the design
The most important part of the agent is the point where it stops. It needs a defined boundary beyond which it passes the conversation to a person, together with a summary of what has been said so far. The rules are simple: complaint, warranty, price discussion, audible annoyance or failing to understand twice. Then the phone rings for the team, or the case lands on the callback list with its context.
Example scenario: on Saturday lunchtime, a customer calls. After two years, the battery of their e-bike no longer charges; they consider it a warranty case and get louder. At the second “But that’s covered by the warranty!”, the agent says a colleague will call back and writes on the callback list: “Battery won’t charge, bike bought here in 2024, customer expects warranty, annoyed, call back today by 4 pm.” Whoever calls back doesn’t need to ask anything and can go straight to discussing the case.
So in every quote you get, ask about the hand-off first: when does the agent hand over, to whom, and what goes into the note?
Your team also has to be able to take over these hand-offs. Article 4 of the AI Act requires deployers to support the AI literacy of their staff. A short, documented briefing on what the agent can do and when it hands over is an appropriate building block for this.
Connecting to appointment booking and inventory management
The agent is only as good as the calendar it reads from. A workshop schedule with the real capacity is the prerequisite for the project, not its result. A test question: does your schedule today show that the only colleague with manufacturer training for the drive system is on holiday next Thursday? If not, the agent may well book three motor diagnostics on exactly that day. In that case, you digitalise the calendar first.
Without integration, the agent remains a friendly answering machine that dictates appointments onto a slip of paper. It has to read the calendar that reflects workshop capacity and write the work order the team works from in the morning.
The industry’s inventory management systems come with building blocks for this. Tridata promotes its online appointment booking on the basis that a booking directly becomes a scheduled work order, matched to the workshop’s capacity. The velo.port features page lists, in the workshop section, an appointment planner with capacity monitoring and text message and email notifications, and includes online appointment booking in its feature overview.
An agent can take the same route as online booking, provided the vendor opens it up via an interface. You clarify this with the vendor in advance. For status updates, the agent needs read access to the order status; for the spare parts question, read access to stock. You grant write permissions per system and per action: create appointments yes, delete appointments only after confirmation, change prices never.
Without an open booking route, what remains is a custom-built bridge. For stock queries, Tridata offers, for example, the TriCon-WebService, the shop interface of its inventory management system, whose integration, according to the product page, you can also program yourself. Whether workshop appointments and order status can be reached through it is something to clarify with the vendor. If both are missing, what remains is a calendar system that the workshop maintains in parallel. That works, but it is the weakest option, because two schedules can drift apart.
Data protection and the AI Act: what must be settled before launch
A phone call involves personal data, and the agent processes it at an external provider. So the following needs to be settled before launch.
The agent says in its first sentence that it is an AI. Always. Behind this is Article 50(1) of the AI Act: people must be informed that they are talking to an AI. The obligation falls on the provider and has applied since 2 August 2026. You can find the wording at the European Commission’s AI Act Service Desk.
Because the agent generates synthetic speech, the provider must also mark it in a machine-readable format under paragraph 2. For systems that were already on the market before 2 August 2026, this applies from 2 December 2026. Ask the provider how they implement both.
When choosing a provider, the guidance “Artificial Intelligence and Data Protection” published by the DSK (German Data Protection Conference) on 6 May 2024 helps: define the purposes in advance, put a data processing agreement under Article 28 GDPR in place with the external provider and check that input data is not used for training. EU hosting is not mandatory as long as transfers to third countries are safeguarded under Articles 44 et seq. GDPR, but it is a sensible default. Providers, prices and further GDPR questions are compared in the voicebot overview with costs and a GDPR check.
Section 201 StGB (German Criminal Code) makes the unauthorised recording of privately spoken words a criminal offense, so no recordings without the caller’s consent. For the workshop, the transcript is enough, and even that needs a legal basis, information under Article 13 GDPR and a deletion period.
Pilot in 30 days: how to get started with the AI agent
The most sensible way in is call forwarding outside business hours or when the line is busy: the main number stays the same, and the agent only takes what would otherwise go to voicemail. Forwarding like this can be set up in a few days and switched off again at any time; that is the core of an AI phone assistant.
Week one belongs to the process: which services may the agent book, with which time slots, which questions does it ask, when does it hand over? Your workshop manager answers this, not an agency.
Week two is integration and testing: connect the calendar or inventory management system, and run twenty test calls from your own team covering the typical cases, including the awkward ones. In weeks three and four, the agent runs live during off-peak hours, and you measure three values: how many calls it handles completely, how many it hands over and how many bookings the team has to correct.
A pilot in autumn measures the off-season. It shows whether booking and hand-off work, not how the agent copes with the first warm Saturday in April. Set it up in the quiet months and time the live phase for the start of the inspection season, so you get robust figures over six weeks.
If the hand-off rate is above 40% after two weeks of live operation, two out of five calls still end up with the team, just later. In that case, stop. You set this threshold in advance, even if you work with an AI agency as your partner. The worked example above gives you the comparison against the costs, and the pilot replaces its estimate with your real figure.
Conclusion: relieve the phone, the skills shortage remains
An AI agent won’t solve the skills shortage in bike workshops, and anyone who promises that is selling you something. What it solves is an allocation problem. The scarcest resource in the business is hands that can open up an e-bike drive, and those hands don’t belong on the phone.
One objection still deserves a hearing: for some retailers, the phone is where regular customers are won. An agent that speaks smoothly but impersonally can cost trust there. That is why the pilot starts during off-peak hours, and why the hand-off comes first in the design. Whether you expand it is decided by the three metrics and your team.
If the numbers add up, you will then have something no job advert will deliver by spring: a workshop that can be reached without anyone putting down the spanner.