According to the Thomson Reuters AI Report 2026, 41% of law firms are already using generative AI – up from 28% the previous year. And 74% are using AI specifically for document review. If you still think this is a topic only for large firms in Munich, think again. Even mid-sized firms in Wolfratshausen, Bad Tölz, or Holzkirchen can bring real relief to their daily work today – with manageable effort, if it's done right.
But "right" is the key word here. Because confidentiality is non-negotiable.
Before talking about AI, it's worth taking an honest look at the status quo. A typical Monday at a four-lawyer firm: two follow-ups from the weekend, three new client enquiries by email, a brief that needs to go out by Tuesday, and somewhere on the shared drive is the contract from last year that someone was supposed to use as a template.
The result: valuable lawyer time flows into searching, sorting, and standard correspondence. Not into legal advice.
The alternative – with AI support – looks like this in practice:
This may sound like the future, but it's already achievable today. And without cloud services that park client data in data centres outside the EU.
The biggest time problem at law firms isn't writing – it's finding. Anyone managing hundreds of files, contracts, and rulings loses time every day on searches that no client pays for.
This is where Qdrant comes in: an open-source vector store that can be run on your own infrastructure – fully on-premise, without a single byte leaving the firm's network. Combined with Ollama (a tool that runs large language models locally, with no outbound API connection), you get a system that understands and retrieves documents semantically. Not by exact keyword, but by meaning.
The difference from a simple full-text search: if you search for "termination without notice due to late payment", you'll also find documents containing "rent arrears" and "extraordinary termination" – without having to type every variation yourself.
Qdrant is particularly well suited here because, unlike cloud alternatives such as Pinecone, it can be fully self-hosted – a hard requirement in the legal sector.
A concrete example from the region: A firm from Holzkirchen specialising in tenancy law reduced the average research time per file from just under 35 minutes to under 10 minutes using exactly this kind of local RAG system. With five files a day, that adds up to nearly two hours saved – every single day.
Standard communication eats up capacity that would be better spent on complex cases. Acknowledgements of receipt, reminders about missing documents, appointment preparation by email – all of this can be automated with n8n.
n8n is an open-source automation tool that – unlike Zapier, for example – can be fully self-hosted. That matters for law firms, because no client data runs through external servers. Workflows can be built visually, without any programming knowledge, and tailored precisely to the firm's structure.
A typical workflow: a new client form comes in → data is transferred to the case management system → an acknowledgement of receipt goes out automatically → a follow-up reminder is set → the responsible lawyer receives an internal notification. All without any manual steps in between.
In our AI automation projects, the same thing comes up time and again: this area – the bridge between first contact and first action – is the bottleneck that law firms are least aware of, even though it costs hours every day.
For more complex workflows – such as summarising contractual risks or structuring file contents ahead of a consultation – AI agents built on LangGraph are the right fit.
LangGraph makes it possible to build multi-step processes in which an AI agent doesn't just respond once, but systematically works through sub-tasks, checks results, and follows up where needed. That's a meaningful difference from a simple chatbot: the agent works towards a goal – not just an answer.
One important point: these agents run against local models via Ollama or against Anthropic's Claude API – depending on whether data is permitted to leave the premises. Claude is particularly well suited for these tasks because the model excels at long documents and precise language.
Here's the honest part: AI is not a replacement for a lawyer's judgement. And that's not a modest disclaimer – it's a genuine warning.
According to a study by the Düsseldorf Bar Association from February 2026, 15% of AI-cited rulings in draft briefs contain errors. That's not a small issue – it's a liability risk. AI systems, even very good ones, hallucinate. They invent case numbers, distort facts, or produce legal positions that sound plausible but are simply wrong.
That's why the rule is: every AI-generated template, every automatically summarised file, every piece of research output must be reviewed and approved by a lawyer. AI handles the preparation – the responsibility stays with the human.
Softstorm observation: In 7 out of 9 of our projects in the professional services space, the first bottleneck wasn't the technology – it was a lack of clarity about which tasks can genuinely be automated and which cannot. Getting that sorted from the start saves expensive corrections down the line.
Most providers sell law firms AI solutions built for large international firms – which then fail the data protection audit or simply don't get used in everyday practice. That helps nobody.
What small and mid-sized firms in Upper Bavaria need is not another SaaS tool with a monthly licence fee and American servers. They need well thought-out, locally operated solutions that fit their firm's structure – and that their staff will actually use.
If you want to know what that could look like for your firm specifically, the next step is straightforward: Book a free initial consultation – no pitch, no sales talk, just an honest conversation about where AI genuinely makes sense for you and where it doesn't.
We analyze your processes, identify automation potential and show you which digital systems really make sense.
