AI knowledge base for trades businesses: the practical guide

junova editorial team · 13 July 2026

An AI knowledge base for trades businesses is a central repository that holds quotes, instructions, delivery notes and site photos, which can be found by asking a question in plain language. Instead of rummaging through folders or calling the office, the fitter asks ‘What were the settings on the system in Bahnhofstraße?’ and gets the answer together with the source document. For businesses with piles of paperwork and a full diary, this is an obvious way into using AI, because it solves a real everyday problem.

In brief

Trades businesses lose time because knowledge is scattered across paper, emails, phones and people’s heads.
Text recognition (OCR) makes even scanned delivery notes, old instructions and photographed notes searchable.
Fitters upload photos and documents directly from their smartphone, without installing any special software.
Roles cleanly separate what the office sees (cost calculations, personnel) from what the installation team needs (instructions, job data).
The introduction is not an IT project: analysis, initial loading with the most important documents and test questions are enough to get started.

Why do trades businesses drown in paperwork and scattered files?

Trades businesses create documents in many places at the same time, but there is rarely one place where everything comes together. The quote is written in the office, the delivery note arrives on site, the photo of the finished connection ends up on the fitter’s private phone, and the special arrangement with the customer is in an email from the boss. Each place works on its own; taken together, the result is paper chaos.

A plumbing and heating business with 15 employees quickly has thousands of relevant documents: quotes and order confirmations going back years, maintenance instructions for the installed equipment, inspection reports, time-and-materials reports, photos of pipe runs before plastering. Only a tiny part of this is ever needed, but that is exactly the part that cannot be found at the crucial moment.

The consequences are easy to see in everyday work. The fitter calls the office and waits, the office worker interrupts her work and searches, and if nothing is found, someone drives to the site or the job is done from memory. This is exactly where a searchable central repository comes in.

Which questions does an AI knowledge base answer in everyday trades work?

An AI knowledge base answers the questions that today end up as a phone call to the office or with the most experienced colleague. Typical examples from the everyday work of plumbers, electricians, joiners and similar businesses:

‘What did we quote the Muster family last year, and what was included in the price?’
‘What were the settings of the heat pump at the Gartenweg property after the last service?’
‘Which make and model did we install for this customer?’
‘Are there photos of the pipe run in the bathroom before it was plastered?’
‘What does the manufacturer’s manual say about error code E14 on this boiler?’
‘What agreement was there with the customer on the warranty for the custom-made item?’

The crucial difference from the classic folder search: the answer comes as text with its source, and one click opens the original document. The fitter does not need to know what the file is called, which folder it is in or who filed it.

Important for expectations: the AI bases its answers on the company’s own documents. What was never documented, such as a verbal promise, no system in the world can conjure up. In practice, this often leads businesses to record important agreements in writing more consistently, because the effort now noticeably pays off.

What does text recognition (OCR) do for the paper archive?

OCR stands for optical character recognition and means that software recognises the text on scans and photos and makes it searchable. For trades businesses this is a central building block, because a large part of the company’s knowledge exists on paper: delivery notes with handwritten remarks, old instructions, signed acceptance reports, fax confirmations from earlier years.

With text recognition, the photographed delivery note becomes a searchable document. The question ‘When was the material for the Hauptplatz site delivered?’ then also finds the delivery note that was uploaded as a crooked phone photo. Older PDF files that are only stored as images are made accessible in the same way.

You should stay realistic about the quality of the originals. A clean scan works very well, a crumpled carbon copy with faint writing causes problems for any text recognition, and hasty handwriting remains a weak point. The practical rule: photograph important papers as flat as possible and in good light, and for critical handwritten notes, add the key statement briefly in typed form.

How do documents and photos get from the site into the database?

Uploading works directly from the smartphone in the browser, without any special software having to be set up on the fitter’s phone. The fitter photographs the rating plate, the meter reading or the finished connection and uploads the photo. From that moment on, the information is available to the office and for later jobs.

Automatic tagging takes over most of the filing work. The system reads the uploaded document and fills in attributes such as customer, project, installation, date and topic itself. Only the document type is confirmed by a person in the review area, because it determines who gets to see the sheet later; for an unambiguous delivery note that is one click. Manually sorting into folder structures is no longer necessary – and that is exactly where classic filing systems in the trades regularly fail: they only work as long as someone maintains the folders with discipline.

This also ends the problem of photos on private phones. Site photos that today disappear into private galleries and messenger chats end up in a place that belongs to the business, even if the employee moves on one day.

Who may see what? Roles for office and installation

User and role management determines which employees may see and query which documents. In trades businesses, a simple two-way split has proven effective, which can be refined as needed.

Role	Typically sees	Typically does not see
Office and management	All documents including cost calculations, purchasing terms, personnel documents	–
Installation and site	Instructions, job data, reports, photos, technical documents	Cost calculations, purchase prices, personnel files
Apprentices	Instructions and work instructions	Customer correspondence, commercial documents

The roles should be defined before the initial loading, not afterwards. If you load everything first and then restrict rights, you risk sensitive documents having been visible to everyone in the meantime. Briefly thinking through the question ‘Who needs what for their work?’ is usually enough.

Security also includes looking outwards: servers in the EU, a data processing agreement (DPA) under the GDPR (General Data Protection Regulation, the EU data protection law) and the assurance that your own documents will not be used to train AI models are a must for a trades business’s customer data, not an optional extra.

How does the introduction work without a dedicated IT project?

Introducing an AI knowledge base is feasible for a trades business within a few weeks and requires neither servers in the basement nor an IT department. The process follows the pattern described across industries in the overview article AI knowledge base in your company: initial meeting, analysis, setup with initial loading, test questions, introduction for the team.

For the trades, this means in concrete terms: it starts with selecting the documents of greatest use, usually quotes and job documents from the last few years, instructions for the installations under ongoing maintenance and the reports of active customers. The complete paper archive since the business was founded does not have to be digitised; what gets scanned later is what everyday work actually demands.

The touchstone is test questions from real everyday work. When the office manager asks her five most frequent search cases and the most experienced fitter asks his typical site questions, it immediately becomes clear whether the loading and the answers fit. Providers such as Goma-it from Vorarlberg structure the entire introduction according to exactly this pattern; the individual steps are described on the Process page.

After the start, the system largely sustains itself, because new documents are continuously added when they are uploaded. One task remains: in the review area the document type is confirmed so that the access rights are correct. It therefore makes sense to have one person in the office who acts as the contact and handles these approvals on the side; in a typical business, there is no further ongoing maintenance effort.

Frequently asked questions

What does searching for documents really cost a trades business?

The cost lies in interrupted work: the fitter waits for a call back from the office, the office employee leafs through folders, the boss answers questions in the evening that a filing system could answer. Added to this are mistakes when people work from memory because documents cannot be found. How much this adds up to depends on the business, but it is noticeable almost everywhere.

Does an AI knowledge base also work with handwritten notes?

Handwriting is the most difficult discipline of text recognition; clearly legible block capitals usually work, hasty cursive often does not. In practice it has proven useful to photograph important handwritten notes and additionally add the key information briefly in typed form. That way the original is preserved and the information can reliably be found.

Do fitters need training for this?

The effort is small, because using it matches what fitters do every day anyway: take photos and type or dictate a question. A short introduction of under an hour is usually enough. More important than training is that the first questions people ask themselves deliver useful answers – then the team uses the system of its own accord.

Who maintains the knowledge base in day-to-day operation?

In everyday use, the database largely keeps itself in shape: the system fills in attributes such as customer, project, installation, date and topic itself when documents are uploaded. Only the document type needs to be confirmed, because it determines who may see the document; for unambiguous documents that takes a few clicks in the review area. It makes sense to have one responsible person in the office who handles these confirmations and marks outdated documents. A fixed maintenance day each week is not necessary.

What happens to the old paper archive in the basement?

The paper archive does not have to be fully digitised; in most businesses that would be disproportionate. What has proven effective is to scan specifically what is still needed: documents for current maintenance customers, instructions for installations still in use, quotes for active customers. The rest stays in the basement and is scanned only when needed.

How secure is customer data in an AI knowledge base?

What matters are the server location, the contractual situation and access control. Look for servers in the EU, a data processing agreement (DPA) under the GDPR and the assurance that your documents will not be used to train AI models. Within the company, roles control who may see which documents.

For your company

What this looks like in practice is on the page for your industry — with the questions asked there and the document the answer comes from.