AI for SMEs: why consulting and ongoing support make the difference

Setting up an AI assistant takes minutes these days, and the first results are often impressive. A few months later, many companies take a more sober view. Some employees use the tool every day, others not at all. Nobody knows exactly which data goes into it, and a measurable benefit is hard to name. The technology is rarely the problem. Usually two things are missing: a clear idea of what AI should be used for, and someone who looks after it once it is running.

The task first, then the tool

The most common opening question is: which model is the best? A more useful one is this. Which work should AI take off your team's hands? Good candidates are tasks that take a lot of time, repeat themselves and produce results that can be checked.

In our workshops we collect such tasks together with your departments and assess each one against three criteria:

• Effort: How much work does implementation involve, and which data and systems need to be connected?
• Benefit: How much time does the team actually save, and can it be measured?
• Risk: Which data is involved, and what happens if an answer is wrong?

The result is a prioritised list you can work with, not a slide deck for the drawer. We start with the use case that pays off fastest.

Where AI delivers for SMEs today

The tasks that work best are those where AI does the groundwork and people still make the decisions:

• Finding knowledge: Employees ask in plain language about content from manuals, contracts or minutes and get the answer with its source, instead of clicking through folder structures.
• Preparing correspondence: Quote texts, replies to recurring enquiries and summaries are drafted in your company's tone. People still sign them off.
• Pre-sorting data: AI categorises enquiries, extracts information from documents and flags anomalies in figures, so your team decides instead of sorting.

The data question comes first

As soon as quotes, customer histories or HR records are involved, where the data is stored becomes critical. With public AI services, this content leaves your company, and data processing agreements, storage location and further use all need to be clarified. For non-critical tasks, a public service is still often the most economical option.

For sensitive data, your own AI environment is the better choice. Open language models then run on a GPU server on your premises or on dedicated hardware in our data centre in Frankfurt am Main. Queries stay within your environment, your content does not train anyone else's models, and you control access through roles and permissions. For text assistance, document search and summaries in a mid-sized team, a single GPU server is often enough.

What matters is drawing the line deliberately: which tasks may go to a public service, and which stay in-house? There is also an obligation that many companies have not yet noticed. Since February 2025, the EU Artificial Intelligence Act (AI Act) has required companies that use AI systems to ensure their staff have sufficient AI literacy. A planned introduction with clear rules makes that easier, too.

Why AI needs ongoing support

An AI environment is not a project that ends at go-live. Models evolve quickly, components need updates, and the quality of answers depends heavily on the underlying data. If your filing is messy, the answers will be too. Ongoing support therefore covers:

• updates to models, components and servers
• monitoring of load and response times
• backups of the environment
• regular optimisation through better prompts, cleaner data or better-suited models

We calculate this effort transparently per month, based on environment size and usage. Unlike pay-per-request pricing, you know in advance what to expect.

AI where the work happens

AI is most useful when it is not a separate window next to the actual work. That is why we connect your documents, knowledge bases and business applications, for example our business software dCM. Roles and permissions determine who gets which answers, and the content stays within your environment.

From our own experience

We do not recommend anything we do not run ourselves. Our own models run on our own hardware, with the same interface our customers get. Along the way we have learned which model size still responds smoothly on which hardware, where answers break down when the data is not right, and how much care such an environment really needs. Model, server, integration and operation all come from a single provider. If something goes wrong, you have one point of contact.

Conclusion

AI pays off for SMEs once three questions are answered: which task should it take on, which data may it see, and who looks after it in operation? If you start with a concrete, measurable use case and plan for support from day one, an experiment becomes a tool that holds up in daily work. You can find out more about our services on our AI consulting and AI infrastructure page.

Note: This article provides a general overview. It does not constitute legal advice.

All posts