Contents and how to read this

This ebook is designed as a working document. Chapters 6 to 16 follow the path of a job through the business – from the first enquiry to maintenance. If you are short on time, read the executive summary, the decision matrix in chapter 5 and the roll-out plan in chapter 20.

A strong trade with a
growing administrative load

Sources: ZVEH industry figures 2025 (provisional, skilled-trades reporting); Bitkom study report „Digitalisation of the skilled trades“, a survey of 504 skilled-trade companies in the summer of 2025.

The market is shifting, not disappearing

Installation figures in the electrical trades' core energy-transition business diverged in 2025. Photovoltaics and storage declined, while heat pumps grew significantly. For business planning this means one thing: the market is not shrinking, but it is becoming harder to read.

Source: ZVEH spring business survey 2026, conducted from 10 to 19 February 2026 with 1,641 participating firms.

1. Executive summary

The electrical trades are among the sectors most affected by the energy transition, the digitalisation of buildings and the networking of technical systems. Yet the workload is growing not only on the job site, but above all alongside it.

Photovoltaics, battery storage, charging infrastructure, heat-pump connection, energy management, building automation, network technology and smart metering have widened the traditional range of services. Each of these fields brings its own manufacturer portals, registration procedures, technical connection conditions, documentation requirements and product variants.

As a result, a customer enquiry today rarely produces just an installation date and an invoice. In many cases, photos, existing-system data, manufacturer documents, grid-connection forms, measurements, handover records, day-work reports, operating manuals and as-built documentation have to be brought together and reconciled. This work is necessary, but only partly demanding in technical terms. And it is precisely here that artificial intelligence currently offers the greatest benefit.

What AI does not do

AI replaces neither the master electrician nor the qualified electrician. It takes no professional responsibility for protective measures, cable sizing, discrimination, grid connection, testing or commissioning. Anyone who uses it that way shifts liability to a place that cannot carry it. What it can do is capture, structure, compare, summarise and prepare information for professional review.

What a sensibly introduced system does

The commercial value does not come from a single spectacular application. It comes from many small reliefs along the entire order-to-completion process. That sounds unspectacular, and that is exactly why it holds up.

The starting position is an advantage, not a drawback

Despite its technological importance, the skilled trades are still at the beginning when it comes to using AI. In the Bitkom survey of 504 skilled-trade companies, four percent were already using AI and a further nine percent were planning to. 84 percent said the topic played no part for them at present. At the same time, 89 percent see digitalisation overall as an opportunity for their business.

That gap between recognised opportunity and actual implementation is the real story. It means that anyone who selects suitable processes today, puts data and responsibilities in order and starts with a limited pilot can build a lead – without having to reorganise the entire business.

The six propositions of this ebook

2. The electrical trades under
pressure to change

The electrical trades have long since stopped working only on circuits, sub-distribution boards and lighting systems. With every additional line of business, technical expertise grows – and the administrative load grows with it.

One business, many trades within the trade

Many electrical firms today cover several of the following fields at once. Each brings its own manufacturer portals, forms, technical connection conditions, product variants, certifications and documentation requirements:

• power and building services
• photovoltaics and battery storage
• charging infrastructure and load management
• heat-pump connection
• metering and measurement technology
• KNX and building automation
• security and communication technology
• network technology
• industrial and control technology
• E-CHECK and periodic inspections
• maintenance and fault service
• energy management
• smart-home solutions
• bidirectional charging

The point is not the breadth itself, but that each field brings its own administrative logic. A firm offering wallboxes, heat pumps, PV systems, storage and meter-panel conversions is effectively running five different registration and certification processes in parallel – often with different grid operators and portals.

The rulebooks now move faster than they used to

Anyone who assumes technical standards are a stable backdrop should look at the spring of 2026. Within a few weeks, two central application rules for low-voltage systems were completely revised and reissued.

Sources: VDE FNN and DKE on VDE-AR-N 4105:2026-03; overview of changes to VDE-AR-N 4100:2026-04.

Competition intensifies because the market is tipping

The number of photovoltaic systems installed by electrical contractors fell to around 355,000 in 2025, after roughly 395,000 the year before. Storage systems showed a decline of a similar magnitude, from 260,000 to 235,000 units. A second figure is notable: the share of firms still installing PV systems at all fell from 57.1 to 52.1 percent.

The ZVEH reads this as a warning sign. When firms withdraw from a line of business, expertise and capacity are missing when the market picks up again later. For the individual business, the situation is more sober: the market is not disappearing. It is changing.

Charging infrastructure: the installed base grows, new business stalls

As of 1 June 2026, the Federal Network Agency's register listed a total of 152,915 normal charging points and 53,292 fast charging points in operation, together providing around 8.87 gigawatts of simultaneously available charging power. The public installed base is therefore still growing.

Source: charging-point register of the Federal Network Agency (Bundesnetzagentur), data as of June 2026. The figures also include entries from registration procedures not yet completed.

At the same time, the number of charging points newly installed by electrical contractors fell from 377,000 to 360,000 in 2025; the only growth was in ultra-fast charging stations. The pattern is the same as for PV: the growing installed base creates a need for support and maintenance. New-installation business no longer carries this development on its own.

The skills shortage is changing in character

For years the diagnosis was: too few skilled workers, too many orders. That picture is becoming more nuanced. The number of vacancies in the electrical trades is falling, but remains high. In the ZVEH autumn business survey 2025, 46.5 percent of the firms surveyed reported open positions; in autumn 2024 it was still 52 percent. The number of employees subject to social security contributions fell by 1.2 percent in 2025.

The reason is not that demand has been met. It lies in a combination of demographic change and economic caution. And that is exactly what shifts the question. When additional skilled workers are neither available nor affordable at short notice, the relevant question is no longer „how do we hire more people?“ but:

This is not a rhetorical question. You can answer it for your own business – for example by noting for a week how much time goes into searching, follow-up queries, duplicate data entry and after-the-fact reports. In most firms the result is uncomfortable.

Why training has to be part of the thinking

Around 46,000 apprentices work in the electrical trades. This generation brings digital skills that often go unused in the business. The Bitkom study shows that 54 percent of training firms in the skilled trades let their apprentices help with digitalisation. That is a signal, not a plan – but a workable starting point for a pilot with a low barrier to entry.

All three forces point in the same direction: the value of a business increasingly lies in how well it manages information. That is the real reason AI is a topic in the electrical trades – not because the technology happens to be getting a lot of attention right now.

3. The bottleneck is rarely
on the job site

Many firms could handle more orders if their office organisation kept pace with their technical capability. The order book is rarely the problem. The path from the phone call to the invoice is.

The contradiction every business knows

Digital applications are viewed positively in the trades. In the Bitkom study, 76 percent of firms named time savings as an important benefit. At the same time, firms report being too busy in day-to-day work to engage more deeply with digitalisation.

This contradiction explains why many digitalisation projects fail not on basic willingness, but on lack of time, unclear responsibilities and too broad a project scope. A firm that tries to change telephony, estimating, job-site documentation and knowledge management all at once will fail on all four fronts.

The consequence for adoption

Chapter 20 describes this path in detail as a plan for the first 100 days. Until then, it is worth keeping this question in mind: which process in your business happens every day, costs noticeable time each time and can then be checked reliably? That is exactly where the work begins.

4. What AI really means
in the electrical trades

The term artificial intelligence covers very different technologies. Anyone who fails to distinguish between them ends up buying the wrong thing. For an electrical firm, five categories are relevant.

1. Speech and text processing

These systems understand spoken or written language and turn it into structured information. Today they are the most mature and easiest category to introduce.

• call notes from phone conversations
• summary of a site meeting
• draft of a customer e-mail
• service report from a dictation
• analysis of a schedule of work
• wording of a technical explanation

2. Document analysis

Documents are not merely searched by keyword. The AI recognises content, connections and differences. The benefit depends directly on how well ordered the filing is.

• comparison of two drawing revisions
• detection of missing documents
• summary of an installation manual
• matching data sheets to items
• extraction of line items from schedules of work
• comparison of tender and quotation

3. Image analysis

Photos and scans can be recognised, described and categorised. For job-site documentation this is the most interesting category – and at the same time the one with the greatest risk of confusion.

• recognition of a rating plate
• matching of job-site photos
• capture of visible equipment
• detection of obviously missing labelling
• sorting by room, distribution board or system
• comparison of existing state and drawing

4. Forecasting and pattern recognition

AI can spot anomalies and recurring patterns in existing business data. This category needs several years of data and is therefore rarely a first step.

• frequently recurring faults
• unusual material consumption
• recurring variation orders
• project types with poor margins
• elevated failure risk of certain components
• seasonal fluctuations in orders

5. AI agents and automated workflows

An AI agent does not just analyse information; within defined limits it triggers further steps. This is the category with the highest benefit and the greatest need for careful design.

What matters is that permissions and approval points are defined in advance. An AI agent must make no technical commitments, quote no binding prices and grant no commissioning approval. It prepares. A human decides.

Why the distinction matters in practice

Vendors often sell „AI for your business“. The question you should ask is: which of these five categories are you deploying, and for which specific process? A firm that needs a knowledge assistant but buys a forecasting platform has solved neither one problem nor the other.

5. Where the commercial
value comes from

Not every task is suited to AI. The choice of task decides success or failure far more than the choice of vendor.

Nine features of a good use case

A viable first use case meets as many of the following criteria as possible. The more of them apply, the more likely the case will hold up after the pilot.

Looking at the job, not at the department

The usual departmental boundaries – office, engineering, installation, accounting – are of little help in selecting AI applications, because the friction arises precisely at the handovers. It is more useful to look along the job. The following page therefore organises the fields of application into four phases: before the job, during the job, at completion, and afterwards.

Four phases of order handling

Decision matrix for AI use cases

Rate each possible use case from zero to two points. The value of the matrix lies not in the final number, but in the conversation about why one person scores a row differently from a colleague.

Interpretation

6. Enquiry handling
and phone calls

Customers rarely call with fully prepared information. The real work begins with the follow-up questions – and over the year these cost more time than any single installation.

What enquiries sound like in practice

An AI-supported phone assistant can take such enquiries, classify them and ask for the required details in a structured way. It is reachable outside office hours and then hands the information over to the ERP, CRM or ticketing system. It does not replace advice. It ensures that the advice can start with complete information.

Example: a charging-equipment enquiry

Among other things, the assistant asks:

• residential or commercial property?
• number of charging points required
• intended charging power
• existing photovoltaic system or storage?
• parking space clearly assigned to the building?
• location of the meter panel
• responsible grid operator
• photos of the meter cabinet and mounting location
• billing to employees, tenants or third parties?
• dynamic load management required?
• reverse-feeding charging desired?
• controllability under Section 14a EnWG planned?

The answers are not treated as a finished technical design. They qualify the enquiry and prepare the next step. Whether the wallbox can be connected to this meter panel at all is still decided by the site survey.

Example: fault intake

Transparency towards the caller is mandatory

From 2 August 2026, the transparency obligations of the EU AI Act apply to interactive AI systems. Customers must be able to tell that they are communicating with an AI system, unless this is obvious from the circumstances. This obligation applies regardless of risk class: a simple phone assistant is harmless in regulatory terms and still subject to the labelling requirement.

The second sentence is not a courtesy add-on. A direct route to a person is the practical precondition for customers to accept the system – and the only reason firms usually receive no complaints about AI telephony.

Checklist: AI in enquiry handling

• The caller is informed that AI is being used.
• There is a direct route to a member of staff, available at any time.
• Emergencies and hazard situations are defined as a separate path.
• The system grants no independent technical approvals.
• It quotes no binding prices or dates.
• Call data is processed only for defined purposes.
• Retention and deletion periods are defined.
• Customer and site data is transmitted encrypted.
• The handover to ERP, CRM or ticketing is tested.
• Misclassifications are reviewed regularly.
• Responsible staff check the records created.
• A data-processing agreement with the provider is in place.

What you can realistically expect

A practical note on adoption: do not start with your main number. Start with a clearly delimited channel – such as the answering machine outside business hours or the contact form on the website. There the damage from errors is small and the comparison with the current state is clear-cut.

7. Estimating and
quotations

Estimating in the electrical trades is not material quantity times installer hours. It is twenty things you have to know before you can calculate – and five you only learn on the job site.

What a quotation has to take into account

• existing conditions
• cable routes
• fire protection and sealing
• protective measures
• system earthing arrangement
• meter panel
• distribution boards
• switching and control technology
• site set-up
• scaffolding and lifting equipment
• core drilling and chasing
• scope of documentation
• testing
• registrations with the grid operator
• coordination with other trades
• night, weekend or shutdown work
• material availability and lead times
• disposal and removal
• warranty and maintenance scope
• price escalation over a long contract term

AI cannot size these points correctly on its own. But it can bring together information from various sources and make visible what is missing. In the quotation process that is often the more valuable service.

A draft quotation from a customer enquiry

From the e-mail, call notes, photos and survey record, the AI produces a structured draft in ten sections:

The estimator then decides which line items are kept, changed or removed. The draft is a starting point, not a result.

Analysing schedules of work

For extensive schedules of work, AI can group line items by topic, identify specified manufacturers, flag unusual contract terms, extract quantities and unit-price fields, highlight missing or contradictory details, help distinguish ancillary services from special services, prepare questions for bidder communication, and compare similar line items with earlier projects.

Protection against false confidence

Language models write convincingly even when they lack information. This is not a malfunction, but a property of the technology. A smoothly written paragraph says nothing about whether the underlying detail is substantiated.

Every AI-generated draft quotation should therefore distinguish visibly between three categories:

This labelling matters more than particularly elegant quotation wording. A quote that reads smoothly and contains three unchecked assumptions is more dangerous than one that reads awkwardly and names the assumptions.

Checklist: AI-supported quotations

• Source documents are clearly assigned to a project.
• The current drawing revision is identifiable and dated.
• Assumptions are explicitly labelled.
• Exclusions and interfaces are checked by a qualified person.
• Prices come from the approved estimating system.
• Material alternatives are not adopted unchecked.
• Standards and technical connection conditions are checked in their current version.
• The estimator gives binding approval to the quote.
• Changes to the AI draft remain traceable.
• Variations and deviations are later compared with the estimate.

8. Work preparation
and planning

Between the incoming order and the start on site lies work preparation. Errors in this phase are especially expensive, because they multiply on the job site.

The automatically prepared project overview

From the available documents, an overview is created that the project manager checks and then releases to the installers:

• contacts on both sides
• site address and access
• scope of work
• construction phases
• key dates
• technical particulars
• access requirements
• safety requirements
• required qualifications
• material with long lead times
• required measuring and test equipment
• open decisions
• interfaces to other trades
• planned documentation

Support with material lists

From an approved plan, AI can create a first suggestion for a material list. This is especially robust for recurring standard packages: a wallbox including protection and communication link, a smaller photovoltaic system, a sub-distribution board for one dwelling unit, a lighting refurbishment, a network distributor, a KNX base fit-out or an inspection order of defined scope.

The parts list then has to be reconciled with the actual conditions, the approved manufacturer and the technical requirements. The time saving lies in the first draft, not in the approval.

Qualification-based scheduling

In scheduling, further criteria can be considered alongside availability and distance: qualified electrician or instructed person, live-working authorisation, experience with a particular manufacturer, KNX qualification, PV or storage experience, mobile-platform authorisation, access permission at the customer's premises, language skills, familiarity with the site, and the measuring equipment required.

9. AI on the job site

Installers should be able to document their work without typing long texts on a smartphone after every step. Speech is therefore often the only input method actually used in daily practice.

From dictation to a structured report

The installer checks the text and confirms it. That confirmation is the point at which a suggestion becomes a document someone stands behind.

Photo documentation

AI can organise photos by the following attributes:

• site, floor, room
• distribution board and circuit
• system and construction phase
• condition before the work
• interim state and completion
• rating plate and instrument reading
• identified defect

For matching to work reliably, a simple scheme is needed on site. Scanning a QR code at the distribution board or the room entrance before a series of shots has proven effective. It costs three seconds and saves half an hour of sorting in the office.

Spotting variations early

From site reports, photos and queries, AI can generate clues to possible additional work. Typical triggers:

• existing conditions differ from the documents
• an agreed prior service is missing
• the cable route is unusable
• additional fire-protection measures required
• material changed by the client
• access was not possible
• the shutdown could not be carried out
• the work area was not released by other trades

The project management decides whether this leads to a notice of hindrance, a variation quote, a day-work report or merely an internal note. The value of AI is that the clue arises on the day of the event and not three weeks later when the invoice is reconciled.

10. Documentation, testing
and as-built records

In the electrical trades, documentation is not an office by-product added afterwards. It is part of proper, verifiable work – and therefore part of liability.

What may be required depending on the job

• test records
• measurement lists
• circuit diagrams
• distribution-board layouts
• circuit schedules
• cable lists
• data sheets
• conformity documents
• commissioning records
• grid-connection documents
• system schematics
• operating instructions
• photo documentation
• maintenance notes
• handover records
• as-built drawings

The regulatory framework is moving

Requirements for the initial verification of electrical installations are addressed, among others, in DIN VDE 0100-600, and the operation of electrical installations in DIN VDE 0105-100. For generating plants on the low-voltage grid, the fully revised VDE-AR-N 4105:2026-03 has applied since 1 March 2026, and for customer installations VDE-AR-N 4100:2026-04. Added to this are the technical connection conditions of the respective grid operator, the manufacturer documentation and DGUV Regulation 3.

What AI may do with test records – and what it may not

Completeness check for as-built records

Before handover, an AI system can use a project-specific checklist to verify whether the documents are complete and free of contradictions. This is one of the commercially strongest use cases in the entire process, because the final invoice often hangs on it.

Professional approval remains with the responsible employee. The AI produces a list of what is missing – it does not declare the documents complete.

A word on the order of your filing

AI speeds up searching in ordered filing. In disordered filing it speeds up the spread of wrong documents. If three versions of a distribution-board layout sit side by side and none carries a reliable date, an AI assistant will pick one – and present it convincingly.

This is not an excuse to start only in two years' time. It is a pointer to where the pilot should begin: with a project type whose filing already works. Not with the one that causes the most pain.

11. Service, maintenance and
fault handling

Valuable knowledge builds up in service work over the years. In many firms it is spread across several systems or exists only in the heads of individual service technicians. That is convenient – as long as the person is there.

The site overview before the visit

Before a visit, an AI-supported knowledge assistant can provide a concise overview from approved site records. Example:

Fault suggestions instead of remote diagnosis

Based on earlier cases, the system can suggest possible test steps. These suggestions must not be presented as a confirmed cause. A suitable presentation states in each case: possible cause, basis of the clue, recommended test step, required safety preconditions, reference to manufacturer documentation, and the status of professional confirmation.

The difference is not cosmetic. „The motor protection is probably too small“ leads to an order. „In two earlier cases at this system the cause lay in the inrush current of the extraction unit – X was checked at the time, the result was Y“ leads to a measurement.

Predictive maintenance: a topic for later

For industrial systems, measurement, runtime and condition data can be examined for changes – temperature, current draw, vibration, switching frequency, running time, insulation values, fault codes, power, energy consumption, power quality.

12. Company knowledge and
standards management

The problem for most electrical firms is not a lack of knowledge. It is the inability to find the knowledge they already have at the right moment.

An internal company GPT

An internal company GPT is a protected AI assistant that accesses only approved business knowledge sources: manufacturer documents, work instructions, estimating bases, templates, test procedures, project files, installation notes, maintenance contracts, technical connection conditions, training materials and customer agreements.

Typical questions it can answer:

• Which documents does the grid operator need for this generating plant?
• What standard build do we use for an 11-kW wallbox?
• Which photos must be taken at handover?
• What exclusions do we use for excavation work?
• Which test equipment does the installer need for this job?
• Who in the firm knows this inverter?
• Which documents are missing in project 2026-118?
• What maintenance services were promised to this customer?

Sources must stay visible

A professionally usable answer does not just output text. It shows which source the information comes from, how current it is, which product or project it applies to, whether it is an internal rule or an external requirement, and whether the answer is complete or uncertain. An assistant without source references is unusable in a technical environment – no matter how good the answers sound.

Handling standards

A regulated model is more sensible:

• licensed access to standards via the intended route
• internal work instructions based on the relevant requirements
• references to clauses instead of full texts
• documented update processes
• a clear separation between the text of a standard and internal explanation
• professional review by responsible staff

An AI assistant must not give the impression that it replaces complete and binding information on the standards. It is a tool for finding your way around company knowledge, not the rulebook itself.

Assigning responsibility for knowledge

For every knowledge source, the following should be defined: the person professionally responsible, approval status, scope of validity, version date, next review date, access permission and archiving rule.

13. Materials management
and procurement

Missing or wrongly ordered material costs more than purchasing time. It costs journeys, site interruptions, express surcharges, waiting time, unproductive installer hours, missed dates and rework.

AI can support purchasing when item master data, supplier information and project requirements are available in structured form. In many firms this precondition is already met through Datanorm data and wholesaler catalogues – often better than in other areas.

Four applications

The underestimated side effect

Consumption analysis is often more revealing than expected. If material consumption on five comparable lighting refurbishments is systematically above the estimate, that is either an estimating error, an execution problem or shrinkage. All three explanations matter – and none of them stands out as long as no one places the projects side by side.

14. Project control
and post-calculation

In well-utilised firms, a full order book can mask an inadequate project margin. Many losses only become visible long after the project is finished.

The core of the problem is not that this information is missing. It is that no one places it side by side on a regular basis. That is exactly what AI can do – daily, without anyone having to remember.

Early-warning signals

The AI does not judge whether a project is unprofitable. It provides signals that the project manager interprets. The distinction matters: a system that automatically reports „project off track“ will be ignored after four false alarms.

Learning from post-calculation

The precondition, however, is that post-calculation happens at all and that actual times are assigned to a project. Where hours are simply booked to „site“, no system can learn anything. That is not an AI question but a question of time recording – and it belongs before, not after, the AI roll-out.

15. Customer communication
and sales

Customers do not necessarily expect an immediate technical solution. They expect a reliable response. That distinction wins more orders than any argument about price.

Where AI carries in communication

• acknowledge enquiries automatically
• request missing information in a structured way
• prepare survey appointments
• explain project status clearly
• notify of delays early
• announce maintenance
• provide operating instructions
• evaluate customer feedback
• answer recurring questions

Customer portal

A digital customer portal can provide current orders, appointments, open decisions, documents, test records, invoices, maintenance intervals and system overviews. An AI assistant can answer questions within it – but only on the basis of the information the respective customer is authorised to see.

Quote follow-up

AI can prioritise quotes by order value, contribution margin, probability of closing, the customer's strategic importance, free capacity windows, customer queries, the age of the quote and the follow-up work required. What matters is how the prompt is worded:

Visibility in search engines and AI systems

Customers increasingly search for specialist firms and technical information not only through traditional search engines, but also through AI assistants. Anyone who wants to appear there has to set out clearly: what services they offer, in which region they operate, which customer groups they serve, what qualifications they hold, which manufacturers and systems they support, how a project runs, which documents are needed and which typical questions can be answered.

16. People, training
and knowledge retention

In the trades there is a worry that digitalisation leads to greater control or to the loss of craft skills. That worry is not unfounded and cannot be argued away.

In the Bitkom survey, 56 percent of skilled-trade firms associated digitalisation with greater monitoring in everyday work. An adoption effort aimed solely at time savings, performance measurement and control will meet resistance – and that resistance is rational.

Typical answers from firms:

• duplicate data entry
• searching for documents
• repeated follow-up queries
• unstructured photo documentation
• writing reports after the fact
• manual transfer of measurements
• missing project information
• unclear responsibilities

None of these tasks requires a qualified electrician. All eight cost a qualified electrician's time. Whoever justifies adoption this way gets a different conversation than the one who starts with productivity figures.

Support for new employees

A knowledge assistant helps new employees understand company procedures faster: Where are photos stored? How is a day-work report created? What details does purchasing need? How does a fault escalation run? Who approves material alternatives? Which documents belong to a project handover? What checklist applies to a wallbox installation?

The assistant does not replace professional training or instruction. It relieves experienced colleagues of giving the same answer for the twentieth time.

Support for training

Apprentices can use AI to have technical terms explained, summarise learning material, create practice questions, improve the wording of documentation, reflect on work steps and understand the connections between components. Throughout, it must stay recognisable when the AI provides an educational explanation and when a binding technical source is required.

17. Technical architecture
and system integration

A chat window on its own changes no process. The value of an AI solution depends almost entirely on whether it is embedded in the existing system landscape.

The system landscape of an electrical firm

• ERP or trade software
• estimating
• CRM
• document management
• time recording
• financial accounting
• warehouse management
• phone system
• e-mail and calendar
• project platform
• manufacturer portals
• grid-operator portals
• market master-data register
• mobile installer app

Three stages of build-out

Cloud, EU cloud or local operation

The question of the operating model is often argued on ideological lines. It is a trade-off between protection needs, effort and performance – and it can be answered differently for each use case.

The four questions on data residency

These four questions are also the core of vendor selection in chapter 19. A vendor who cannot answer them in a single sentence has not asked them.

18. Data protection, security
and the EU AI Act

Regulation is manageable if use is kept limited. Most typical applications in the electrical trades fall into the low- or limited-risk range. That does not exempt you from the transparency obligations.

The risk-based approach in practice

The right-hand column is not theoretical: a firm that pre-sorts applications with AI operates a system in the employment domain – and thus a potential high-risk system under Annex III.

The EU AI Act deadlines, as of July 2026

The situation changed in 2026. The so-called Digital Omnibus, on which the Council and Parliament reached a provisional agreement on 7 May 2026, pushes key obligations for high-risk systems significantly back. The transparency obligations are expressly not affected.

As of July 2026. The formal adoption of the Omnibus amendments by the Council and their publication in the Official Journal were still pending at the editorial deadline. Until then, the 2024 version formally applies. Check the current legal position before making your decision.

National supervision in Germany

With the AI Market Surveillance and Innovation Promotion Act, the German federal government has set the national implementation in motion. The Federal Network Agency (Bundesnetzagentur) is becoming the central market-surveillance authority, except where other specialist authorities have jurisdiction; a coordination and competence centre and an AI market-surveillance chamber are being set up there. In practical terms: the Federal Network Agency runs an AI service desk as a point of contact, with guidance and a compliance compass. For firms without their own legal department this is a sensible first port of call.

Twelve questions before adoption

1. What specific purpose is being pursued?
2. What data is processed?
3. Is personal data needed at all?
4. Are customer, employee or applicant data used?
5. Where is the data stored and processed?
6. Does the provider use inputs for training?
7. Which subprocessors are involved?
8. How long are inputs and outputs stored?
9. Who is given access?
10. How can data be deleted or corrected?
11. Is a data-protection impact assessment required?
12. How are wrong results detected and corrected?

In its guidance on artificial intelligence, the German Data Protection Conference recommends, among other things, defining fields of use and purposes in advance, settling responsibilities, raising staff awareness, checking results for accuracy and involving the employee representation.

The works council and performance monitoring

Technical systems intended to monitor the behaviour or performance of employees are subject to co-determination under Section 87(1) no. 6 of the Works Constitution Act (Betriebsverfassungsgesetz). This can become relevant when AI analyses travel times, compares installation performance, rates the content of conversations, attributes error rates to individual employees or automatically assesses productivity.

Information security

For generative AI systems, the German Federal Office for Information Security (BSI) points to risks such as data leakage, faulty outputs, manipulated inputs and dependence on providers. For an electrical firm there is an added factor: it holds information that is of great interest to third parties.

A firm that maintains security technology, access systems or critical-infrastructure installations effectively manages a map of its customers' weak points. This data does not belong in a private AI account – nor, unchecked, in a cloud whose subprocessor chain no one knows.

Ten basic company rules

The point most firms are at

The uncomfortable truth is that AI is already being used in many firms – just without any rules. Employees use private accounts, upload quotes, have e-mails drafted and photos described. This does not happen out of ill will, but because it helps.

19. Selecting suitable
AI solutions

The following ten questions are no substitute for a formal tender. But within a single conversation they separate the vendors who have understood a process from those who show a demonstration.

Ten warning signs

A question that is rarely asked

Who are your reference customers in the electrical trades – and may I speak to one?

This question is so effective because it tests two things at once: whether the vendor knows the trade and whether its existing customers speak well enough of it to pick up the phone. A vendor without a reference in the trades is not automatically bad – but you will be its first learning project, and that should be reflected in the price.

20. A roll-out plan for
the first 100 days

The plan is deliberately kept small. It is meant to work alongside the daily business, not instead of it.

21. Worked examples

22. Calculating the
business case

It is not the number of AI features that decides the benefit, but the number of processes actually affected.

The basic formulas

Worked example

The assumptions are disclosed so that you can replace them with your own values. They do not describe a real business.

Benefits not measured in hours

• invoices sent earlier
• more complete day-work reports
• fewer extra journeys
• less time spent searching
• fewer forgotten services
• faster response to quotes
• higher close rate
• shorter onboarding time

Beware of exaggerated productivity promises

Studies show clear productivity gains for certain language- and knowledge-based tasks. These figures circulate widely in presentations – usually without a note on what they actually refer to.

An honest cost side

What is regularly missing from business-case calculations:

Upkeep of the knowledge sources is the most frequently underestimated item. It is not a project cost but a permanent task – and it decides whether the system is still used after two years.

23. What does not make
a good first step

The following applications are not fundamentally wrong. They are wrong at the start. Anyone who begins with them burns budget, trust and the workforce's willingness to try a second time.

24. A maturity model for
electrical firms

Most firms are at level 0 – including those that believe they have not yet started. That is precisely the problem.

25. Management checklist:
is our business ready?

Answer each question with yes, partly or no. At the end, count only the clear yes answers. „Partly“ is, in this context, a no in a better mood.

26. Ten principles for AI
in the electrical trades

Conclusion

AI will not take over the actual electrical work. Customers will still need qualified firms that plan, build, test, commission and permanently maintain installations. What will change is the organisation around that work.

The competitive electrical firm of the coming years will take in information faster, make project knowledge more readily available and tie documentation more closely to execution. It will handle customer enquiries in a structured way, relieve installers of paperwork and learn systematically from completed projects.

None of this requires full automation. Nor does it require a platform, a data strategy or a transformation programme. What it requires is a clearly delimited process that occurs frequently, costs noticeable time today and can be checked reliably by a qualified person.

Good starting points

Anyone who begins with a use case like this can find out within a few weeks whether AI produces a measurable benefit in their own business. The risk of that decision is small. The risk of postponing it for another two years is not – not because AI would be better by then, but because the firms that start today will already know how it is done.

27. Sources and further
information

All sources were last checked in July 2026. Statistical figures may change through later revisions; the ZVEH industry figures for 2025 are based on skilled-trades reporting and are provisional. Final figures are expected from the 2025 skilled-trades census, likely in April 2027.

Legal and technical notice

This ebook offers an operational and technological perspective. It replaces neither electrical design or testing nor legal, data-protection or standards advice. For every project, the current statutory provisions, accident-prevention regulations, technical connection conditions, manufacturer specifications and the relevant DIN, VDE and VDE-AR-N standards must be checked in their respective valid versions.

The practical examples shown are illustrative cases and do not represent published customer references. Calculated examples are based on disclosed assumptions and are not a guarantee of any result.

Statements on the EU AI Act reflect the position as of July 2026. The legislative process on the Digital Omnibus had not been concluded at the editorial deadline. Check the current legal position before basing any decisions on it.