What AI in mid-sized companies actually means
Anyone searching for "AI for SMEs" finds plenty of advice aimed at very small businesses: a chatbot on the website, copy for social media. This guide is written for mid-sized companies, meaning companies with 50 to 500 employees in Austria and Germany. The questions there are different. They concern processes that span several departments, an ERP system, permissions and data that has grown over years.
First, a distinction that shortens many conversations. There are two ways to use AI. The first is a tool for individual employees: an assistant that drafts, summarises or translates text. The second is AI as part of a process: it reads incoming documents, proposes an assignment or answers questions based on your own data. The first is a licensing question. The second is a software project.
Where AI helps mid-sized companies
AI is strong with unstructured content: text, emails, PDFs, meeting notes. Wherever a person today reads, classifies and retypes information, it can do a lot of the groundwork.
Typical use cases:
- Capturing incoming documents: reading orders, delivery notes or email enquiries, extracting the relevant fields and proposing the next processing step.
- Making knowledge findable: answering questions about manuals, contracts or internal guidelines, with a reference to the passage in the original document.
- Preparing texts: quote texts, summaries of service cases or minutes as a draft that a person checks and approves.
- Cleaning up data: finding duplicates in master data, harmonising inconsistent spellings, flagging missing information.
- Software development: AI tools write code and tests under supervision and so shorten the timeline of software projects.
All of these have one thing in common: the AI makes a proposal, and a person or a fixed rule decides. This division of labour is not a concession to caution. It is the reason such solutions hold up in everyday work.
Where AI does not help
The opposite case matters just as much. AI is not a good choice when a result always has to come out exactly the same. A price calculation, a payroll run or the check whether an order exceeds a department's capacity are rules. Rules belong in conventional software that always returns the same result for the same input and can be tested.
Nor does AI help when the process itself is unclear. If three departments handle the same task in three different ways, AI speeds up the confusion instead of resolving it. And it does not replace the decision about who is responsible for what.
A rule of thumb: where meaning is involved, AI can help. Where rules are involved, the logic belongs in software you can trace and test.
Prerequisites: processes, master data, permissions
Whether an AI project in a mid-sized company succeeds is rarely decided by the model. It is decided by three prerequisites that at first have little to do with AI.
Clear processes
Before AI takes over a step, it has to be clear how that step works today, who is responsible for it and what happens with exceptions. That sounds obvious. In practice, this very knowledge often sits in Excel spreadsheets and in the heads of individual employees.
Clean master data
An AI that is supposed to match orders to items needs an item master you can rely on. Duplicate customers, outdated item numbers or free-text fields with their own logic lead to wrong proposals that sound very convincing. Maintaining master data is therefore often the first and least visible part of an AI project.
Settled permissions
As soon as an AI answers questions using company data, the question arises of who may see what. An assistant that answers questions about all documents may well reveal salary data or contract terms. Permissions therefore have to be settled before the AI, not afterwards. This includes the question of where the data is processed: in a provider's cloud, on your own premises or in a data centre in Austria.
What AI projects realistically cost
When it comes to costs, the distinction from the beginning helps. An AI assistant for individual employees is an ongoing licence per person. The costs are easy to plan. The benefit depends on whether employees use it sensibly and whether rules exist for handling confidential data.
As soon as AI becomes part of a process, it is a software project. Then the usual orders of magnitude for custom software apply. As a guide from our projects, all amounts net: a small business application for one process in one department usually costs between 10,000 and 30,000 euros. A system for several departments with roles, permissions and interfaces costs between 30,000 and 90,000 euros. Company-wide systems start at around 90,000 euros. Operation and maintenance add from 300 euros a month.
The share that goes to the AI itself is often smaller than expected. The larger part is the connection to the ERP and existing systems, the permissions concept, data migration and testing. If these points are missing from a quote for an AI project, it is worth asking why.
First steps
How to approach the start in an orderly way:
- Choose one process: not "AI in the company" but a concrete task with noticeable effort, for example capturing incoming orders.
- Record the current state: who does what, with which data, how often, and what happens with exceptions?
- Check the prerequisites: is the master data reliable? Are permissions settled? Where may the data be processed?
- Start small and check: build a first stage with a clear scope, test it in real day-to-day work and only then expand.
- Assign responsibility: for every step it has to be clear whether the AI proposes and a person decides, or whether a fixed rule applies.
With us, a project like this starts with a paid analysis lasting one to two weeks. It costs 3,000 to 6,000 euros net and delivers requirements based on real examples, a first prototype and a binding offer. The fee is fully credited when you place the order. If there is no usable result, you get it back. Implementation then follows at a fixed price per stage.
In mid-sized companies, AI is rarely a technology problem. It is usually a question of processes, data and responsibility.
Conclusion
AI can take a lot of groundwork off mid-sized companies, especially with documents, texts and knowledge that is hard to find. It does not replace clear processes, clean master data or a permissions concept. Those who put these prerequisites in place first have a foundation on which AI features work reliably.
If you would like to find out whether AI is worthwhile for a particular process in your company, we are happy to talk it through. The initial call takes 30 minutes and is free of charge. Afterwards you will know whether a project makes sense and roughly what size it would be.