Service /1
An AI answer is only useful where the work happens: in the document, in the ticket, in the search screen of your existing system. That is exactly where we put it.
We put language models and automated analysis where your data already sits: into the ERP system, into the document store, into the ticket system. That is the difference between an assistant your people have to open separately and a function that appears in the screen they are working in anyway.
We have been building software since 2003 for companies that depend on grown systems. For us, artificial intelligence is another layer on a foundation that has to carry load. That is why every AI project here begins with the question of whether the data, the permissions and the interfaces underneath are robust enough for continuous operation.
When you bring artificial intelligence into the company, you choose one of two paths. The difference does not show in the demonstration but in the second quarter.
Path A
A tool is bought and placed next to the existing systems. It does not know your master data, it does not know your approval routes, and it writes nothing back. Whoever works with it copies text in and results out.
At first this looks fast, because nothing has to be touched. After a few weeks usage falls, because the detour costs time every day and nobody knows for certain whether the answer reflects the current state. A year later the licence is still in the budget and the level of use is down to a handful of people.
Path B
The model is given defined access to the data it needs for the task and returns its result where it is needed: into the document, into the ticket, into the search screen. The workflow stays the same, it only gets shorter.
That requires work on the interfaces, and that is exactly where our craft lies. We have been connecting systems to each other for twenty years, long before language models existed.
We build the second path only. If a ready-made tool is enough for your task, we will say so and build nothing.
Austria is in a better position on the use of artificial intelligence than the mood in the country suggests. 30 percent of Austrian companies with ten or more employees used artificial intelligence in 2025, against an EU average of 20 percent. In 2021 the Austrian figure was still 9 percent.
Source: Statistics Austria, survey "IKT-Einsatz in Unternehmen", survey year 2025, published 2026. Population: companies with ten or more employees. statistik.at, PDF
The other half of the same survey is more revealing. Of the companies that do not yet use artificial intelligence, 77 percent state that they have never considered the subject. Those who did consider it and decided against it cite a lack of expertise (15 percent), data protection concerns (11 percent) and legal uncertainty (11 percent).
What follows from this is above all a sober reading. In most cases a company falls behind because no decision is taken at all. A first step must therefore be small enough that it is allowed to fail without daily business noticing.
On the subject of artificial intelligence, the market in Salzburg is geared towards consulting and standalone tools. The connection to grown software and to existing permissions usually remains open. That is exactly where we start.
Both shares come from the same publication: Statistics Austria, "IKT-Einsatz in Unternehmen", survey year 2025. The 77 percent refer to the sub-group of companies without AI in use. statistik.at, PDF
These six points are fixed before we connect anything. They are the reason a project can be audited later, including by your legal department.
For the European legal framework we also assign your use case to a risk class and record that classification in writing. Most cases in mid-sized companies fall into the lowest class. That belongs on the record.
Three tasks that occur in almost every company with grown software, whatever the industry.
Quotes, minutes, test reports and contracts sit in several repositories with different structures. Instead of searching by file name, your people ask in full sentences and receive the passage in the document, together with a reference to the file and page. The existing permissions remain unchanged.
Incoming invoices, delivery notes and forms arrive as PDFs or scans and are currently transferred into a screen by hand. A model reads the fields, matches them to your master data and creates the document in the ERP system. Uncertain cases go into a review list, not into the system.
Tickets arrive by email, by form and as phone notes. A model identifies the request, the urgency and the responsible team, attaches similar earlier cases and proposes a reply. It is a person who sends it.
All three share the same construction: the model takes over reading and pre-sorting, the decision stays in the company. Where a dedicated application has to be built first, the path leads through custom software.
AI integration is not a licence purchase. Almost all of the effort lies in the connection to your systems, not in the model itself. Seven factors determine the price, and you influence three of them.
On top of that come running costs: the usage based billing of the model provider or the operation of your own hardware, plus maintenance. These costs grow with usage and not with the number of workplaces.
On the amounts
A price range without your scope would say nothing. After a short assessment you receive a written estimate with line items you can strike out individually. You learn the rates we work with beforehand, in the initial consultation.
Four questions that come up in every initial consultation, plus two that are asked too rarely.
Only if you permit it for the type of data in question. Before building, we define which data may go to an external model provider and which may not. For cases in which nothing may leave, we operate an open model on your own hardware. That costs more and delivers somewhat weaker results, but it is technically possible and we implement it.
For most tasks, no. A current language model does not have to be trained with your data, it is shown the relevant sections at runtime. That makes the entry possible for companies that do not hold large, carefully maintained data sets. What you need is one clearly described work step that costs time today.
Company size is not the deciding factor. What matters is whether there is a recurring work step that occurs often enough to justify the connection. A company with thirty employees that records two hundred documents a day has a clearer case than a corporate group with no recurring task.
We build the connection so that the model remains exchangeable. The link to your systems, the preparation of the data and the logging belong to you and remain in place if the provider is changed. A change is then an adjustment in one place and not a new project.
Every answer is delivered with its sources so that the person at the screen can check it. Where a decision touches money or law, the model proposes and a person approves. Uncertain cases land in a review list that a person works through. We do not build an AI function that decides silently.
A narrowly defined first case is usually in trial operation within a few weeks, provided the interface exists and one contact person in the company is allowed to decide. If the interface is missing, that is where the larger part of the work arises, and the duration depends on the existing system. After the assessment we tell you which of the two cases applies.
Describe a work step that costs too much time today. You receive an assessment from us: whether an AI function pays off for it, which interface would be needed and where you should begin. The conversation takes half an hour and costs nothing.
