AI knowledge base in practice: the most common use cases in a company

In Austrian small and medium-sized companies, the most valuable knowledge rarely lies in records, but in heads and scattered documents: how did we solve this problem back then? With which settings does the installation run stably? An AI knowledge base collects a company’s documents in one central place and answers such questions in plain language – with a source pointing to the original document. The most common use cases: finding earlier problem solutions, documented installation settings, maintenance knowledge available on the move, onboarding new employees, holiday cover, securing experience before retirement and quick answers on the phone. Finding records such as quotes or delivery notes comes on top as a practical side effect.

In brief

• An AI knowledge base answers questions in everyday language from your own company documents – every answer refers to its source.
• The strongest use cases revolve around know-how: earlier problem solutions, installation settings, maintenance knowledge, onboarding, cover and securing experience before retirement.
• On site or at the customer’s, mobile access counts: instructions and settings can be called up on the phone instead of sitting in a folder in the office.
• Knowledge becomes independent of individual people – noticeably so during holidays, sick leave, onboarding and staff changes.
• For clarity: the ERP knows what was sold – the knowledge base knows how it is done. Finding records is a side effect, not the core.
• You can recognise your own top use cases by recurring questions in the team and by knowledge that only one person has in their head.

What is an AI knowledge base – and how does it differ from folder-based filing?

An AI knowledge base is a central repository for company documents that you can question like a well-informed colleague. Instead of clicking through folder structures or combing through the inbox, you type a question in plain language – for example ‘How did we solve the condensation problem with the ventilation system at the Bergmann property?’ – and get an answer that comes directly from your own documents. A detailed introduction to how it works is given in the article What is an AI knowledge base?.

The difference from classic filing lies in three points. First: the search works by meaning, not just by file names. If you do not know the exact wording or the storage location, you still find the right document – even if the service report says ‘condensation’ and the question says ‘moisture’. Second: text recognition (OCR, i.e. the automatic conversion of scanned pages and photos into searchable text) also makes paper documents findable once they have been scanned – from the old machine logbook to the handwritten maintenance record. Third: every answer names its source. One click opens the original document – you do not have to believe the AI blindly.

The distinction from general AI chatbots on the internet is important: a business knowledge base only answers from your own documents. If something is not in the documents, the system is designed to say so openly. That is exactly what makes it reliable in everyday business. Why this difference is crucial and how ChatGPT with your own data works is a topic of its own.

Just as important is the distinction from the ERP system (the business software for orders, stock and invoices): the ERP manages structured business data – who ordered what and when, which invoice is outstanding, how much is in stock. The knowledge base opens up the unstructured knowledge alongside it: instructions, service reports, settings records, meeting notes, email threads. To put it in one sentence: the ERP knows what was sold – the knowledge base knows how it is done. It does not replace an ERP, but closes the gap the ERP leaves open.

Which use cases occur most often in everyday business?

The following seven use cases come up across all industries – from the trades to manufacturing to law firms. The same pattern applies to each: starting point, question in everyday language, answer backed by a source – and a noticeable change in everyday work.

Use case	Typical question	Who benefits most
Finding problem solutions again	‘How did we solve that back then?’	Technicians, project managers
Installation settings	‘With which values does the machine run stably?’	Production, maintenance
Maintenance knowledge on the move	‘What does error code E42 mean, and how do I fix it?’	Field service, installation
Onboarding and induction	‘What is our procedure for a complaint?’	New employees, trainers
Holiday cover	‘Where are the documents, and what was agreed?’	Office, administration
Retirement	‘What does only the colleague who is leaving soon know?’	Management, successors
Phone enquiries	‘When was the last service for customer Huber?’	Back office, front desk

‘How did we solve that back then?’ – finding problem solutions again

Starting point: a plumbing and heating business with fifteen employees faces a heating problem at a customer’s that seems strangely familiar. Someone in the team has solved something like this before – two or three years ago, at a different property. But who? And how? The project manager at the time is no longer with the company, the service report is somewhere on the server. So the problem is solved anew: with the same wrong turns and the same effort as the first time.

Example question: ‘Have we ever had an underfloor heating system that stays cold only in individual rooms, and what was the cause?’

What the answer delivers: the relevant service reports and project notes from the past, summarised in a few sentences – for example that in two earlier cases a sludged-up manifold was the cause and how it was flushed. Every statement refers to the original document.

What changes in everyday work: the company stops solving the same problems several times. Experience becomes a shared tool instead of the private property of individual heads. And the threshold for briefly documenting solutions drops – because recording them now pays off, since they will be found again.

Calling up documented installation settings and machine parameters

Starting point: in a manufacturing company with thirty employees, an older press brake only runs cleanly when pressure, speed and tooling are correctly combined. These values are not in any manual – they live in the heads of two machine operators and on notes in the tool cabinet. When an order comes in with a material that was last processed a year ago, the guesswork starts all over again.

Example question: ‘With which settings did we last bend 3-millimetre stainless steel on the press brake?’

What the answer delivers: the documented settings from the last comparable order – from the photographed settings note, the set-up record or the project note, with a reference to the source and the date.

What changes in everyday work: set-up times no longer depend on who happens to be on shift. Scrap from wrong first attempts goes down. And the notes in the tool cabinet get a place where they are actually found: a photo of the handwritten settings note is enough, text recognition makes it searchable.

Maintenance knowledge on the move: on the phone in front of the installation

Starting point: the service technician of a refrigeration and air-conditioning business is at a customer’s in front of a unit showing an error message. The operating manual is a PDF on the server in the office, the printed version in a folder in the workshop, the crucial experience – that with this model this fault is almost always the pressure sensor – in a colleague’s head. The call to the colleague in the back office interrupts her work, and his own along with it.

Example question (on the phone, right in front of the unit): ‘What does error code E42 mean on the Model X cooling unit, and how do I fix it?’

What the answer delivers: the relevant section from the stored manual, summarised in clear steps, with a link to the original page in the manual – and, if documented, the note from earlier service reports that with this model the pressure sensor should be checked first.

What changes in everyday work: the technician keeps working, without a call back, without a trip back to the office, without a second appointment. The office is interrupted less often. And manufacturer knowledge and the company’s own experience sit side by side in one place for the first time – available to everyone in the team, not just the longest-serving technician.

Induction and onboarding: new employees help themselves

Starting point: a joinery hires a new work planner. In the first weeks he asks dozens of questions every day: how are orders created? What is the procedure for a complaint? Who orders fittings, and from whom? Every question is justified – and every one interrupts an experienced colleague. After a few weeks, the newcomer no longer dares to ask some questions and prefers to improvise. That is exactly when mistakes happen.

Example question: ‘What is our procedure when a customer complains about a delivered cabinet front?’

What the answer delivers: the documented procedure step by step – who is informed, which form is used, how photos are taken and things are documented – with a reference to the stored work instruction.

What changes in everyday work: new employees answer most of their everyday questions themselves and only ask where real experience is needed. Onboarding becomes easier to plan, experienced staff are interrupted less often, and the answers are consistent – not three versions from three colleagues.

Holiday cover and sick leave: staying able to give information

Starting point: the office manager of an electrical business is on holiday for two weeks. She is the only one who knows where the registration documents for the company lorry are and what was agreed with the tax adviser. Her stand-in sits in front of the same folders – and still finds nothing, because she does not know the logic of the filing. During sick leave it is even trickier: there is no handover, and nobody likes to call.

Example question: ‘What is the procedure when someone calls in sick, and where do I find the form for it?’

What the answer delivers: the documented procedure including the form – without anyone on holiday having to be disturbed and without searching through someone else’s folder structures.

What changes in everyday work: the stand-in remains able to act, and customers and suppliers do not even notice that someone is missing. The holiday actually becomes a holiday – and the company’s ability to give information no longer depends on the rota.

Retirement: knowledge does not retire too

Starting point: in a metalworking company, the workshop manager is retiring after thirty years. He knows every machine and its quirks and has his own records for recurring custom jobs. Traditionally, this is what happens: a few weeks of handover, a thick folder, many good intentions – and a year later the successor still calls the retiree regularly.

Example question (months after he has left): ‘What should be watched out for when setting up the old lathe for brass parts?’

What the answer delivers: the notes from the former workshop manager’s records that were taken over – scanned notebooks, photographed settings notes, short records from the handover period – with a source pointing to the respective document.

What changes in everyday work: the handover gets a goal and a structure. In the last months, the questions that keep ending up with the departing colleague are collected specifically – and the answers documented informally. Whether the knowledge has been secured can be checked before he leaves: you ask the knowledge base the typical questions and see whether useful answers come back. If not, you can ask – the colleague is still there, after all.

Phone enquiries: answering customer questions immediately

Starting point: a customer calls and asks whether their device is still under warranty or when the last service was. Traditionally the answer is: ‘I need to check and will call you back.’ The call back costs time – and sometimes it is forgotten. The customer is left with the impression that nobody really knows what is going on.

Example question (during the call): ‘When was the last service for customer Huber, and what was done?’

What the answer delivers: the information while the customer is still on the line, backed by the service report.

What changes in everyday work: ‘I’ll call you back’ becomes ‘I can tell you right away’ – a noticeable difference in the customer experience. At the same time, anyone in the office can give information, not just the person who looks after the customer. Call-back lists get shorter.

Side effect: finding records, quotes and contracts faster

Besides know-how, the knowledge base also opens up the classic business documents along the way. An old quote for a similar heating renovation serves as a template for costing. The notice period in the maintenance contract, together with conditions agreed later by email, is within reach with one question. And a scanned delivery note signed by hand can be matched to the right invoice before it is paid.

What matters is putting this in the right context: for orders, stock levels and open items, the ERP remains the leading system – the knowledge base does not replace inventory management and does not post anything. Its contribution begins where the ERP ends: with the wording of the contract, with the reasoning in the email thread, with the scanned record. If you measure its benefit by record handling, you underestimate it – its real value lies in the knowledge of how things are done.

A day with a knowledge base: before and after

How the use cases add up over the day is shown by comparing an ordinary working day in an imaginary trades business with twenty employees – not a special case, but everyday life.

Situation	Without a knowledge base	With a knowledge base
07:30 – fitter needs settings for the installation at the customer’s	Call to the workshop, colleague searches for the note, call back	Question on the phone, values with source right in front of the installation
09:15 – new employee asks about the complaints procedure	Experienced colleague interrupts his work and explains – for the third time this week	New employee asks the knowledge base, colleague is left undisturbed
10:40 – customer asks on the phone about the last service	‘I’ll call you back’ – search in the folder, call back in the afternoon	Information during the call, backed by the service report
13:00 – problem with a ventilation system seems familiar	Guesswork again, because nobody knows who solved it back then	Old service reports found: cause and solution are documented and ready
15:20 – office manager on sick leave, stand-in looks for an agreement	Call during sick leave or a decision based on guesswork	Stand-in finds the documented agreement herself
16:45 – quote for a project similar to one two years ago	Recalculated, because the old quote cannot be found	Old quote opened as a template, line items taken over

On its own, none of these situations is dramatic – a quick question here, leafing through a folder there, a piece of information postponed until later. Because the effort is spread across several shoulders and over the whole day, nobody perceives it as a separate item – but it still shapes the working day. It is exactly this invisibility that makes search time so treacherous; a detailed look is given in the article Time lost searching for documents.

How do I find the top 3 use cases for my company?

The most worthwhile use cases are not found on the drawing board, but in your own everyday work. The following self-check helps you narrow them down within a week. Go through the checklist and count how often you answer yes:

• Are there questions that are asked in the team several times a week – among colleagues or by customers?
• Do you regularly hear the sentence ‘Only Franz knows that’ (or Maria)?
• In the last few months, was a problem solved that had already been solved before – without anyone finding the old solution?
• Are important settings or work steps written on notes, in private notebooks or only in people’s heads?
• Do fitters or field staff regularly call the office to ask for documents or values?
• Do new employees ask the same colleagues the same questions for weeks?
• Does the company falter when individual people are on holiday or sick leave, because information is missing?
• Will an experienced employee whose knowledge is not documented anywhere retire in the next few years?
• On the phone, do people often say ‘I need to check and will call you back’?
• Are the same documents stored in several places – server, inbox, paper folders – and nobody knows which version applies?

Evaluation: three or four yeses mean clearly recognisable use cases – the questions where you nodded are your candidates. From five yeses on, the company is very probably losing time every day through searching and asking around; a structured survey is worthwhile.

Three criteria have proven useful for ranking the top 3. First, frequency: a question that comes up daily beats one that comes up monthly. Second, risk: knowledge that only one person has weighs more heavily than knowledge that is tedious but findable – especially if that person is leaving soon. Third, the data situation: a use case for which documents already exist can be implemented immediately; one for which documentation still has to be created goes into the second round.

It also helps to look at the filing itself: where file names such as ‘Instructions_final_v3_NEW’ appear, the version question is unresolved – a clear indication. As a rule of thumb: start with one or two use cases where the pressure is greatest. A small, well-functioning start convinces the team more than a big project that wants everything at once.

Which expectations are unrealistic?

Being honest also means saying what an AI knowledge base does not do. If you start with false expectations, you will be disappointed after four weeks – usually on points that could have been clarified in advance.

It only knows what is documented. The knowledge base makes existing documents findable – it does not create knowledge. The workshop manager’s experience can only be called up once it has been recorded, at least informally: as a note, photo or short record. The good news: the hurdle for this drops considerably, because even unstructured snippets are found. Nobody has to write a manual.

It does not tidy up your filing. Duplicate files and outdated versions remain duplicate and outdated – the AI just finds them faster. If you store an old and a new work instruction side by side, you may get answers from both. A minimum of maintenance is part of it: outdated material is archived or marked.

It does not make decisions. The knowledge base provides the documented basis – whether a goodwill case is granted is still decided by a person. That is intentional: responsibility cannot be delegated, good information can.

Text recognition has limits. Clean scans work well; handwriting is the most difficult case and only shows its results on your own documents; faded carbon copies or very hasty handwriting may be recognised incompletely. For critical old documents, a spot check is recommended.

It replaces neither the ERP nor training. Orders, stock and invoices remain the business software’s job; teaching hands-on skills remains the job of experienced colleagues. The knowledge base takes the burden of answering questions off both – nothing more, and that is enough.

There are no perfect answers from day one. In the first weeks it becomes clear which questions the team really asks and which documents are still missing. If you plan this phase as fine-tuning, you will be further along after a month than someone who gives up at the first ‘I found nothing on this’ – because exactly this honest answer is the mark of quality of a reputable system.

How do these use cases get into your company?

The path from idea to everyday practice is shorter than many people expect – provided you approach it in a structured way. This order has proven effective: first an initial meeting in which the specific use cases are identified – ideally with the results of the self-check in this article; then an analysis of the existing documents; then the setup with the most important documents; after that, we go through the answers on real questions from your everyday work; and finally the introduction for the team. What this looks like in detail is shown on the page Process of the introduction. Which file stores and systems can be connected as sources is covered in the article Connecting documents and systems.

What to look for when choosing a solution: answers must come exclusively from your own documents and always name the source; text recognition should capture scanned documents; access should work on mobile devices; and role management should control who may see which documents – the payroll documents are not everyone’s business. For companies in Austria, servers in the EU and a data processing agreement (DPA, the data protection agreement with the provider) are also important criteria; details are explained in the article AI and the GDPR in Austria. One solution developed for these requirements of Austrian SMEs is junova, the AI knowledge base from Goma-it. If you stumble over a technical term – from OCR to role management – you will find the explanations in the glossary.

The most important advice at the end: do not start with the technology, but with the questions. If you know the ten most frequent questions in your company – and know whose knowledge threatens to disappear with the next retirement – you know exactly which documents belong in the knowledge base first. The benefit then shows at a very concrete point: with the first question that nobody has to answer twice any more.