Skip to content
KisunoAI
Self-host for free Get started
Pricing Docs Blog Get started Self-host for free

AI in everyday work

AI in trades and SMEs across Europe: three workflows that give office time back

Quotes, site reports, invoice formats: in trades and SMEs the office eats the evening, for different reasons in every country. This article walks through three workflows. You see how an AI colleague works with your company’s knowledge, what can be connected, and where the limits are.

Trades and SMEs Reading time: 10 minutes

In brief

  • In a German association survey of 10,630 trade businesses, 74% report that bureaucratic effort rose over the preceding five years, and electronic invoicing sets hard deadlines in several countries.1
  • Kisuno works with your company’s knowledge: quote templates, price lists, regulations and past reports become a knowledge base you can query.
  • In managed operation your data sits in Switzerland, when you self-host on your own server. Customer data does not train any models. The models from OpenAI (ChatGPT) and Anthropic (Claude) run in the AWS data centre in Frankfurt, where the US CLOUD Act applies; for higher data-protection requirements, open frontier models such as GLM, Kimi and DeepSeek are available from European providers on servers in Europe.
  • Software with an open interface can be connected, for example Lexware Office, Pennylane, Fatture in Cloud, Holded or Fortnox. An interface is the point where two programs exchange data. The route for this is called the Model Context Protocol (MCP).

Starting point: the paperwork grows faster than the business

Trades and SMEs are the same kind of company all across Europe: few people, a full calendar, an office that happens in the evening. The pressure comes from a different direction in every country.

In Germany, 74% of the businesses in an association survey with 10,630 participants report that bureaucratic effort rose over the preceding five years.1 On top of that come hard dates. Since 1 January 2025 every domestic company must be able to receive electronic invoices. From 2027 businesses with more than €800,000 in prior-year revenue must issue them too, and from 2028 everyone must.2 In France the deadline is closer: all companies must be able to receive from 1 September 2026, and even the small ones must send from 1 September 2027.3 There, 69% of small and medium-sized companies have invoicing software, but only 20% issue more than half of their invoices in a structured, machine-readable format.4 In Austria the obligation carries a different name. Under section 131b of the Federal Fiscal Code (BAO) a business needs a cash register from €15,000 in annual revenue as soon as cash takings exceed €7,500 a year. Debit and credit card payments count as cash too.10

In Italy, tax obligations cost micro and small enterprises 238 hours a year according to a 2022 calculation by the trades association Confartigianato, 56 more than the OECD average.5 In Spain the subject is late payment: the average payment period in 2025 was 80.5 days, 34% above the legal maximum of 60 days, and 96.5 days in construction in the fourth quarter of 2025.6

In the north it sits elsewhere. The Swedish association Företagarna calculates eight to ten hours of administration per week and owner.7 In Norway it is the next generation: two out of three construction businesses in the NHO Byggenæringen association report an unmet skills need.8 And everywhere the same gap: in 2025, 17.0% of EU companies with 10 to 49 employees used at least one AI technology, against 55.0% of those with 250 or more.9 What small businesses lack is not the technology but an entry point that fits twelve people and three vans.

Use case 1: quote, cover letter and payment reminder from your own company knowledge

An enquiry arrives by email: bathroom renovation, three keywords, two photos. The AI colleague knows the company’s quote templates, price lists and past offers. It suggests line items, writes the cover text in the company’s tone and files the draft for review.

The same mechanism carries the correspondence around it: rescheduling, variations, replying to a complaint, a payment reminder. The payment reminder in particular is more than a courtesy in Spain and France, because every late payment ties up liquidity.6

  1. Enquiry, open invoice or complaint, as brief as it arrives
  2. The company’s templates, price lists, contracts and earlier letters
  3. A draft for the offer, the cover letter and the payment reminder, in the customer’s language
Three drafts from the same company knowledge. Line items and prices are decided by the business.

Working across languages is not an extra, it is everyday life: the client writes in French, the supplier in Italian, the foreman would rather read his mother tongue. The draft is produced from the same documents, in the language it is needed in. For binding texts the contract language remains authoritative.

The difference to a chat in the browser: the draft is based on your documents, and your costing stays in your own instance. An instance is your own copy of the system, kept separate from other customers.

Use case 2: from the job site to a clean report

On the way home the technician dictates three sentences into the phone: what was done, what was used, what stood out, plus two photos. Out of that come the time-and-materials report in the company format, the materials list and a defect description for the site management: worded, numbered, ready to file. A quantities summary is drafted too. The checking stays with the business.

Kisuno runs as a PWA on iPhone, Android devices and the desktop. A PWA is a web application that can be installed on the device like an app. Notes are taken where the work happens. The report waits in the office for approval.

  1. Keywords or a dictation from the job site, with two photos
  2. One input, the company’s own document standard
  3. Report with photo, materials list and defect points, ready to file
Notes from the job site become a document in the company format. Approval happens in the office.

The benefit is not the nicer wording, it is the chain that follows: a report sitting cleanly in the system the same evening becomes an invoice at the end of the month instead of waiting in a drawer.

What deliberately does not happen: no document is sent automatically, none goes into a government reporting system. Notices of obstruction, defect notifications and handover records are legally effective declarations. They wait for approval.

Use case 3: the company’s knowledge becomes queryable

What does this bill of quantities require for insulation? How did we service this installation two years ago? Today, questions like these interrupt the management. A knowledge base is a collection of your own documents that the AI colleague is allowed to search: instructions, manuals, regulations and past reports. From those it answers such questions and names the source. That way the answer can be checked in seconds.

That meets the businesses who cannot spare a training day. A tool that answers in the middle of the work does not ask for a training day. New employees ask in their own language and get answers from the same documents. For safety briefings the official version remains authoritative.

  1. The company’s regulations, installation manuals, contracts and past reports
  2. Permissions per tool, every access logged
  3. An answer with its source, in the employee’s language
The knowledge base answers from the company’s documents and shows where each answer comes from.

One limit belongs here: standard texts are protected by copyright. Whether a DIN, an ÖNORM, an NF or a UNI may go into your knowledge base depends on your licence from the publisher. Your own documents and reports make up the larger part anyway.

Permissions apply down to individual tools: whoever may not see the payroll folder does not see it through the AI either. Every access is recorded in the audit log, the running record of who retrieved what and when.

Connectors: what can be linked to your software

An AI colleague is only as useful as what it is allowed to reach. A connector is the ready-made link between Kisuno and another program. It becomes possible when that program has a documented API, that is, an interface through which programs exchange data. The framework for this is MCP, an open standard for connecting external systems. Permissions apply per tool, and every call is logged. The state of play in August 2026, from the AI connection to Lexware Office and Pennylane through to Fortnox:

Software (country) Interface Connection to Kisuno
Lexware Office (DE) Open REST API documented, with rate limits and event subscriptions Connection via MCP possible: contacts, vouchers, invoices
weclapp and Xentral (DE) Open REST API documented, with an OpenAPI reference at Xentral Connection via MCP possible: orders, articles, stock
Pennylane and Qonto (FR) Open REST API documented, OAuth 2.0 and webhooks Connection via MCP possible: vouchers, open items, incoming payments
Fatture in Cloud and Aruba (IT) Open REST API documented, OAuth 2.0 Connection via MCP possible; sending through the Sistema di Interscambio (SdI), the state clearing point for invoices, remains a step with approval
Holded (ES) Open REST API documented, webhooks Connection via MCP possible: quotes, invoices, projects
Fortnox (SE) Open REST API documented, OAuth2, documented rate limit Connection via MCP possible: customers, projects, invoices
Tripletex and Fiken (NO) Open documentation, with a test environment at Tripletex Connection via MCP possible; for your own company a self-issued token is enough, a key you create in the program yourself
BMD NTCS and RZL (AT) No publicly documented REST API, RZL with a partner programme Access only through the vendor’s partner channel
Bygglet and Cordel (SE, NO) No public API, only ready-made integrations The route runs via the connected accounting software or via exports

One sentence of honesty: in the trades, data exchange traditionally runs not through REST interfaces but through industry standards. These are agreed file formats that a whole sector has settled on. Wholesaler article and price data arrive as DATANORM, the shopping basket via IDS-Connect, bills of quantities as GAEB, costing data in Spain as FIEBDC-3. Invoices follow the European standard EN 16931: XRechnung, Factur-X and the Norwegian EHF implement it directly, while Italy’s FatturaPA is a national XML format of its own that is aligned with the standard.13

These standards move the master data, and AI changes nothing about that. The AI colleague works one level above, on the documents: bill of quantities, offer, report, contract. Is your software missing from the list? Much of the daily work runs without any system connection. How MCP works is described on the integrations page.

Data protection: your costing and customer data stay in the business

Quotes, costings and customer files are trade secrets. Under the GDPR the business remains responsible for personal data even when an AI tool processes it. If a provider processes that data for you, a data processing agreement under Article 28 is required. That is the contract setting out what the provider may do with the data.

When personal data goes to a recipient outside the European Economic Area, Chapter V of the GDPR applies on top. It then takes a legal basis for the transfer and an assessment of the legal situation in the destination country. If storage and processing stay inside the Area, that step falls away. For Switzerland a different route leads to the same result: on 15 January 2024 the European Commission found that revised Swiss data protection law offers an adequate level of protection, so a transfer there needs no additional safeguards.12 This does not make the processing automatically lawful, it makes it easier to justify.

A large share of the AI running in European businesses is off-the-shelf commercial software: in 2025 that applied to 71.0% of the companies using AI in Norway, and to 62.1% in Sweden.9 Such tools can do a lot, but they do not know your business: neither your price list nor your framework contracts.

Kisuno runs managed in the ISO 27001 certified data centres of Green in Lupfig, Switzerland, or self-hosted on your own server, wherever you put it; the certification is held by Green, and one of our own is in progress. Customer data does not train any models. The choice covers the current models from OpenAI (ChatGPT) and Anthropic (Claude); they run in the AWS data centre in Frankfurt. AWS is a US company, so the CLOUD Act applies even with servers in Frankfurt, and we say so openly. For customers with higher data-protection requirements, open frontier models such as GLM, Kimi and DeepSeek are available, operated by European companies on servers in Europe; the CLOUD Act does not apply there. The data processing agreement and the subprocessor list are published openly on the website. This article is not legal advice.

Frequently asked questions from trades and SMEs

My trades software has no interface, for example Bygglet, Cordel or BMD NTCS. Is Kisuno still useful for me?

Yes. Quotes, reports, correspondence and the knowledge base work with documents and need no system connection. Where a connection is missing, the route runs via exports, via DATANORM, IDS-Connect and GAEB, or via the accounting software.

Does this make us ready for the electronic invoicing obligation?

No, not on its own. You meet the obligation with an invoicing system that produces the required format. In France an approved platform comes on top, the plateforme agréée. The AI colleague replaces neither. It can explain the formats and draft the questions for your adviser.

Does the AI write my quote on its own, and does it send payment reminders?

No. It writes drafts from your templates and price lists. Line items, quantities and prices stay with you. Tools that only read are free. Tools that write or send something need a confirmation: payment reminders, terminations and every report to an authority go out only after approval.

Does this replace the tax adviser or the bookkeeping?

No. In Germany, commercial assistance in tax matters is reserved to authorised professionals under section 5 of the Tax Advisory Act, and comparable professional reservations apply in the other target markets, depending on the country. The AI colleague sorts documents and prepares questions. The adviser decides.

Does our business fall under the AI Act?

As a rule not as a high-risk case. Anyone who merely uses an AI system is a deployer under the AI Act, and an assistant that drafts text and searches your own documents counts as minimal risk. Two points still apply: since 2 February 2025 the AI literacy obligation under Article 4 is in force, meaning training and documented rules of use. And an AI that screens applications would be high-risk under Annex III. Those obligations apply under Regulation (EU) 2026/1744 from 2 December 2027, and from 2 August 2028 for systems embedded in products under Annex I.11

What does it cost for a ten-person business?

The managed instance costs CHF 7 (EUR 7.50) per person per month. Through the end of the current month: 50 percent off with a twelve-month term, that is CHF 3.50 (EUR 3.75) per person per month. Testing is free and not limited in time. Every new account starts with a starting credit of CHF 50 (EUR 55) for model usage. No payment details are required; access still runs through the form. Running it yourself costs nothing at all, from a single user upwards. The LibreChat base is MIT licensed; the repository with our own extensions follows shortly.

Where is our data held, and do we need an IT department?

In the managed variant in the ISO 27001 certified data centres of Green in Lupfig, Switzerland, with self-hosting on your own server. The models from OpenAI (ChatGPT) and Anthropic (Claude) run in the AWS data centre in Frankfurt, where the CLOUD Act applies; for higher data-protection requirements, open frontier models such as GLM, Kimi and DeepSeek are available from European providers on servers in Europe. You do not need an IT department. In the managed variant the operator takes care of operation, updates and backups. Otherwise someone who can maintain a Linux server with Docker is enough, which can also be an IT service provider.

See it on a case of your own

Bring a quote, a site report or a bill of quantities, and we will walk through the workflow together.

Book a live demo Get started

Sources

All figures come from these sources, last checked in August 2026. National surveys carry their country.

  1. ZDH, special survey on bureaucratic burden in the trades, first quarter of 2023 (DE, 10,630 businesses): zdh.de
  2. German Federal Ministry of Finance, FAQ on mandatory electronic invoicing (DE): bundesfinanzministerium.de
  3. impots.gouv.fr, facturation électronique, calendar of obligations (FR): impots.gouv.fr
  4. France Num, Baromètre France Num 2025, more than 11,000 companies surveyed (FR): francenum.gouv.fr
  5. Confartigianato Imprese, Ufficio Studi, statement of 27 June 2022: 238 hours of tax bureaucracy per year (IT): confartigianato.it
  6. CEPYME, Observatorio de Morosidad, second half of 2025 (ES): cepyme.es
  7. Företagarna, Klockan klämtar för regelförenkling, 1,279 member companies, 2025 (SE): foretagarna.se
  8. NHO Byggenæringen on the NHO Kompetansebarometer, report of 18 February 2025: 67% of member businesses with an unmet skills need (NO): nhobyggenaringen.no
  9. Eurostat, dataset isoc_eb_ai, survey year 2025, AI use by size class and by the way it was obtained: ec.europa.eu
  10. Federal Fiscal Code (BAO), section 131b, obligation to use a cash register, Federal Legal Information System (AT): ris.bka.gv.at
  11. Regulation (EU) 2026/1744 of 8 July 2026, the Digital Omnibus on AI, OJ L, 2026/1744 of 24 July 2026: eur-lex.europa.eu. The AI Act (EU) 2024/1689 itself, with Article 4 on AI literacy and Annex III: eur-lex.europa.eu
  12. Swiss Federal Data Protection and Information Commissioner on the European Commission adequacy decision of 15 January 2024: edoeb.admin.ch
  13. European Commission, eInvoicing: Directive 2014/55/EU and the European standard EN 16931, ec.europa.eu; on the Italian FatturaPA the country page for Italy, ec.europa.eu

Last reviewed: