AI knowledge base in your company: guides for trades, construction, law firms & more

An AI knowledge base in your company is a central, searchable collection of the company’s documented knowledge: instructions, service records, solved problem cases, proven approaches and the experience that otherwise exists only in the heads of individual employees. It answers questions in plain language and always names the source. How great the benefit is depends heavily on the industry: a plumbing and heating business needs different knowledge at hand than a law firm or a manufacturing company. This guide shows which knowledge and which everyday questions are typical for the most important industry groups, where the boundary with the ERP system lies and how you can tell that your company is ready for the introduction.

In brief

• The real value of an AI knowledge base lies in the company’s documented know-how: in the ‘how’ (settings, procedures, solved cases), not in the ‘what’ (records, orders, invoices).
• Records such as delivery notes and invoices belong in the ERP system; the knowledge base adds the experience that no ERP captures.
• Each industry group has its own typical kinds of knowledge and everyday questions – from the service record in the trades to the proven wording in the law firm.
• A company is ready for the introduction when knowledge depends on individual people, the same questions are asked internally again and again, or knowledge is lost when staff change.
• The most common introduction mistakes are loading old file stores unfiltered, confusing record management with knowledge management and a lack of test questions from real everyday work.
• Answers must always come with a source, otherwise the system is unsuitable for business decisions.

Why is experience the most valuable unused asset in your company?

In almost every small and medium-sized company there is knowledge that is written down nowhere and is nevertheless needed every day. The service technician who knows that a certain heat pump always has to be re-paired after a power cut. The law firm employee who knows how a particular court handles deadline extensions. The machine operator who runs the press with different parameters for thin sheet metal because otherwise the material warps. This knowledge has been built up at great cost over years – and with every sick leave, every resignation and every retirement, a little of it disappears from the company.

This is exactly where an AI knowledge base comes in. It is not another store for records, but the place where the company’s ‘how’ is collected and made findable: how is the installation set up? How did we solve this problem last time? How does this procedure work here, and why exactly like that? The difference from the classic folder structure: employees do not have to know where something is or what the file is called. They ask a question in plain language and get an answer with a reference to the source document. What makes such a knowledge base work technically is explained in detail in the basic article What is an AI knowledge base?.

The economic lever is twofold. First, the daily searching and asking around disappears: instead of calling the one experienced colleague who is out on site, the apprentice finds the answer himself. Where these constant interruptions come from is shown in the article Time lost searching for documents. Second – and more important in the long term – the knowledge stays in the company, even when people leave. A company whose know-how is documented and searchable is less vulnerable and easier to hand over; the Austrian Economic Chamber (WKO) also regularly points out the risks of knowledge tied to individuals in connection with business succession.

Where is the boundary with the ERP system?

The most important distinction first, because it causes the most confusion in practice: an AI knowledge base is not a replacement for an ERP system and does not try to be one. An ERP system (enterprise resource planning, i.e. the software for inventory management, orders and commercial processes) manages business transactions: who ordered what, what was delivered, what was invoiced. Delivery notes, order confirmations, invoices and stock levels are record territory and are well looked after there.

The knowledge base answers a different kind of question. Not ‘Which invoice belongs to order 4711?’, but ‘How did we solve the problem that came up with this order – and what do we need to watch out for next time?’ The ERP knows that an installation was delivered; the knowledge base knows which settings it runs with, which faults it typically shows and what helps against them. The two systems complement each other: the ERP holds the company together commercially, the knowledge base holds the company’s skills together.

In practical terms for loading: if you fill your knowledge base with thousands of delivery notes and invoice copies, you dilute it. These records are already stored in a structured way in the ERP or in accounting and can be found faster there. What belongs in the knowledge base is what the ERP cannot capture: the accompanying service record, the instructions with the fitter’s handwritten additions, the email in which the special solution with the customer was recorded. How an AI knowledge base differs from classic document management systems – which often sit between the ERP and the knowledge base – is covered in the comparison Document management and AI.

Which industry groups have which knowledge and which questions?

Five industry groups cover the majority of Austrian small and medium-sized companies, and each has its own pattern: what know-how is created there, in what form is it available, and what questions do employees really ask in everyday work? The following overview shows the typical cases – deliberately worded the way they are actually asked in the workshop or the office.

Industry group	Typical documented knowledge	Typical everyday question
Trades and construction	Service records, installation settings, solved fault cases, assembly instructions with your own additions, photos of rating plates	‘How did we set up the ventilation at the Bahnhofstraße property so that the whistling stopped?’
Law firms and consultancy	Proven wording and templates, earlier assessments of similar cases, internal procedures, documented agreements	‘How did we proceed last time with a missed deadline for an objection?’
Retail and logistics	Solved complaint cases, handling knowledge about suppliers, product peculiarities, internal process descriptions	‘How do you make a complaint to supplier Huber – who is responsible there and what do they need from us?’
Manufacturing	Machine parameters, set-up instructions, documented fault fixes, inspection procedures, experience with materials	‘What settings do we run the press with for thin sheet metal so that nothing warps?’
Public administration and associations	Resolutions with their reasoning, handling knowledge about funding, internal guidelines, established procedures	‘How did the settlement of the regional funding go – what did we submit last year?’

Trades and construction: the knowledge is in heads, photos and scraps of paper

In trades and construction businesses, the gap between existing and documented knowledge is the largest. The fitter knows which settings the heating system in the apartment building needs – but at best it is written on a note in his service case. The site foreman knows why the pipe was laid differently from the plan during the conversion three years ago – but he is only asked when the next damage occurs. Typical everyday questions sound like this: ‘What was the solution when the boiler at the inn kept going into fault F28?’, ‘What insulation thickness did we install in the loft conversion in Feldgasse?’, ‘How do you set the frequency converter on the old lifting station – the manufacturer no longer exists?’

The biggest lever here is the combination of low-threshold capture and text recognition, OCR for short (optical character recognition, i.e. the automatic conversion of photographed or scanned text into searchable text). The photo of the rating plate, the photographed service record and the assembly instructions with handwritten additions become findable without anyone having to type up reports in the evening. What matters is that capture fits into everyday work: a photo and two sentences right after the job are worth more than the perfect report that is never written.

Law firms and consultancy: knowledge of wording and solved cases

Law firms, tax advisory firms and consultancies usually have their documents in digital form, but the firm’s actual know-how is still hard to access: it is spread across files, email inboxes and the experience of the longest-serving staff. The value lies less in finding individual documents again than in access to proven solutions: ‘How did we word this contract clause last time, and why exactly like that?’, ‘What was our line of argument in the similar case two years ago?’, ‘How does our deadline control work when the responsible colleague is on holiday?’

For new employees in particular, this access is worth its weight in gold: instead of asking about every procedure verbally, they find the firm’s established procedures and model solutions themselves. At the same time, confidentiality and access rights are more important in this industry group than in any other – not every employee may see every file. Anyone who needs statutory texts or case law stays with the Austrian federal legal information system, RIS (ris.bka.gv.at), anyway; the knowledge base adds what is written nowhere else: the firm’s own practice.

Retail and logistics: handling knowledge instead of a flood of records

In retail, there is a reflex to fill the knowledge base with records – orders, delivery notes and invoices are, after all, produced in large quantities. That is exactly the wrong way: these transactions belong in the inventory management or ERP system. What is valuable is the knowledge around the transactions: ‘How do you handle a return with this supplier, and which photos do they require?’, ‘Which haulier takes bulky goods to Vorarlberg, and what needs to be considered when booking?’, ‘Why do we always over-order this item – what was the reason?’

Such handling knowledge typically sits with one or two people in the back office. If they are absent, the business stalls. Documented complaint solutions, supplier peculiarities and internal process descriptions make the company more independent of individual knowledge holders and shorten the onboarding of new employees.

Manufacturing: parameters, set-up knowledge and the machine without a manufacturer

Manufacturing companies live from experience that is not in any machine manual: the actual parameters for a particular material, the order of steps when setting up, the trick for clearing the fault on the old machine whose manufacturer no longer exists. Typical questions: ‘How is the milling machine set up after a tool change – step by step?’, ‘What helped last time when the packaging machine kept pulling in the film?’, ‘Which tolerances does the customer really accept for this series?’

System pays off here: if every fixed fault is documented in three sentences – symptom, cause, solution – a fault memory builds up over months that measurably shortens downtime. Inspection reports and certificates complete the picture, but the core is the crew’s process knowledge.

Public administration and associations: resolutions and the question of why

In municipalities, federations and associations, the critical knowledge is often the ‘why behind the what’: a resolution is in the minutes, but the reasoning, the background and the alternatives considered at the time are known only to those who were there. Typical questions: ‘What has the board already decided on this topic, and why?’, ‘Which conditions does the funding notice contain, and how did we do the settlement last time?’, ‘How does our procurement work – above what amount do we need comparative quotes?’

Since terms of office end and volunteers change, the loss of knowledge is built into the structure here. A searchable collection of minutes, guidelines and documented procedures makes every change of officers considerably less painful.

How can you tell that your company is ready for it?

A company is ready for an AI knowledge base when important skills depend on individual people and the same questions are asked internally again and again. Both can be established without external consultancy. Go through the following self-check list honestly:

• There is at least one person whose absence means certain work is left undone because only they know how it is done.
• New employees take months to settle in because most things have to be explained verbally.
• Experienced employees are interrupted several times a day with questions they have already answered many times.
• Solved problems are solved from scratch a second time a year later because nobody remembers the old solution.
• For properties, machines or cases from earlier years, every follow-up question starts with a lengthy search.
• In the next few years, knowledge holders will retire without their knowledge having been recorded so far.
• Instructions and records exist, but scattered: network drive, email inboxes, paper folders, private notes.

If three or more points apply, the pressure is usually great enough for the introduction to be worthwhile. With one or two points, at least a clearly defined starting area is worthwhile. The honest cross-check is also important: a two-person business in which both know everything and little has to be documented gets little out of it – just like a company that produces records in large quantities but has hardly any process knowledge of its own. Records are the ERP’s job, and AI does not change that.

What mistakes do companies make when introducing it?

Most failed introductions do not fail because of the technology, but because of avoidable mistakes in preparation. The following overview shows the five most common mistakes, their typical consequences and the better way.

Mistake	Typical consequence	Better way
Loading the whole network drive unfiltered	The AI quotes outdated instructions and superseded versions, trust is gone after a week	Deliberate initial loading: current, checked documents first, older material added later in a targeted way
Loading records instead of knowledge	The database becomes a worse ERP, the actual knowledge questions remain unanswered	Leave delivery notes and invoices in the ERP; load instructions, records and solved cases
Clarifying access rights only after loading	Confidential cost calculations or files are visible to everyone, the project is stopped	Define users and roles before loading, add sensitive areas last
No test questions from real everyday work	The system passes the presentation but fails at the fitter’s first serious question	Before release, collect twenty real everyday questions from employees and run through them
Introduction without anyone responsible	After the initial momentum nothing new is added, the database goes out of date	Name one person who checks new additions and reminds the team to document

A sixth mistake deserves its own paragraph: the expectation that the AI knows things that are documented nowhere. A document-based AI answers exclusively from the documents available; what was never recorded, it cannot provide. That is a strength, because it prevents invented answers – unlike freely trained chatbots. But it also means: the introduction makes it mercilessly visible where documentation is missing in the company. Companies that understand this as a task rather than as a shortcoming of the system get the greatest benefit.

The last point concerns data protection. The GDPR (General Data Protection Regulation, the EU rules on handling personal data) requires, among other things, a data processing agreement (DPA) with the provider and clarity about where the data is stored; the Austrian Data Protection Authority provides guidance. Servers in the EU and the assurance that company documents will not be used to train AI models should be minimum conditions. What this means in concrete terms for Austrian companies is set out in the guide AI and the GDPR in Austria.

How does knowledge get from heads into the database?

The most common objection to a knowledge base is: ‘Nothing is written down here anyway.’ That is almost never entirely true – and where it is true, it is not an obstacle but the actual reason for the project. Getting knowledge out of heads does not work with a one-off effort, but with small habits tied to specific occasions.

The three-sentence rule after every solved problem case has proven effective: what was the problem? What helped? What should be watched out for next time? These three sentences – typed, dictated or as a photo of a note – are enough, because the AI takes care of finding them again. Nobody has to maintain folder structures or assign keywords; that is exactly what the automatic tagging in the system takes care of.

For the initial loading, this checklist has proven useful:

• Collect the ten most frequent internal questions – they show which knowledge has to go in first.
• Review existing instructions, records and templates and mark the current versions.
• Plan one to two hours per knowledge holder to record the most important topics in their heads as keyword notes.
• Collect photos of rating plates, installation sheets and handwritten notes and capture them via text recognition.
• Deliberately sort out outdated and duplicate documents instead of carrying them along.
• Decide who will record new problem solutions in future and at what moment (for example directly after the service call).

Where the existing knowledge is already in other systems – in the email inbox, in Microsoft 365, on the network drive – it does not have to be transferred by hand: how existing file stores and programs are connected is described in the article Connecting documents and systems.

How does the introduction work, regardless of industry?

The introduction of an AI knowledge base follows the same five steps in every company; only the content differs by industry. A proven process looks like this:

1. Initial meeting: clarify which questions are asked again and again in the company and where the knowledge for them sits today – in documents, systems or heads.
2. Analysis: survey kinds of knowledge, storage locations and access rights; decide what goes into the initial loading and what deliberately stays out (such as records that stay in the ERP).
3. Setup and initial loading: set up the system, create users and roles, load the selected documents and record the first topics from people’s heads.
4. Test questions: use real questions from everyday work to check whether the answers are correct and the sources fit.
5. Introduction for the team: familiarise employees with asking and recording, so that the system is used in everyday work and keeps growing.

The step with the test questions is the most important and the one most often skipped. Only when the office manager, the fitter or the law firm employee ask their own typical questions and get useful answers backed by sources is the system ready for everyday use. What such a process looks like in detail is described on the page Process of the introduction.

To implement it, a small or medium-sized company does not need its own IT department. Systems such as junova, the AI knowledge base from Goma-it are designed so that setup and initial loading take place with support and the company then carries on working without specialist technical knowledge. What this looks like in real companies – with typical questions, initial loading and everyday use – is shown in the practical report AI knowledge base in practice.

What does an AI knowledge base do, and what does it not do?

An AI knowledge base finds and explains documented knowledge, but it replaces neither specialist programs nor professional judgement. Drawing this line honestly saves disappointment and false expectations in the team.

The tool is well suited to everything that exists in text form or can be turned into text through text recognition: instructions, records, solved cases, templates, emails with recorded agreements, photographed notes and rating plates. It is not suitable as a replacement for ERP and accounting software, costing programs or CAD systems (computer-aided design, i.e. programs for technical drawings and plans); their data stays in the specialist applications. Accompanying knowledge about these systems does belong in it, though – such as the internal instructions on how to create a credit note correctly in the ERP, or the plan description for the CAD project.

Just as important: the AI’s answers are a basis for work, not a decision. In a law firm, the lawyer still checks the legal position; in a plumbing and heating business, the master craftsman still decides on the technical solution. The knowledge base shortens the path to the relevant information – responsibility for applying it stays with people, and every answer must be quickly verifiable via its source.

How do you best get started in your industry?

The best start is a small, clearly defined area where the knowledge pain is greatest today. A plumbing and heating business might start with the service records and installation settings of existing customers, a law firm with the templates and documented procedures of recent years, a manufacturing company with the set-up instructions and fault memory of one group of machines. What matters is that the starting area covers real everyday questions – it is not the area with the most files that wins, but the one with the most follow-up questions.

From this starting area you learn two things: which questions your employees really ask and what quality the documented knowledge needs for the answers to be useful. With these insights, the knowledge base can be expanded step by step without the company having to take on a major project. And with every recorded case, the dependence on the knowledge of individual heads falls along the way.

In-depth guides to the individual industries, for example for trades businesses or construction companies, follow in this Guides section and go into the respective kinds of knowledge, example questions and limits in detail.