Skip to content
AI solutions

AI that takes real work off your plate

AI is not an end in itself. It gets interesting where the same work is done by hand every day - answering enquiries, retyping documents, sorting text. We find those exact spots with you and build something there that runs inside the processes you already have.

We start with the problem, not the model

The first question is never which language model to use, but which task is costing you time. Usually that turns up within a few hours: support emails answered the same way every time. Invoices and delivery notes someone retypes into another system. Quotes assembled from the same ten building blocks.

Then we do the honest arithmetic. If a task comes up ten times a year, AI is not worth it - and we will say so. If it comes up a hundred times a week and can be described clearly, it is a good candidate.

Chatbots that actually know your material

A chatbot is only as good as the knowledge behind it. So we connect it to your own sources - product documentation, price lists, manuals, internal guidelines. Technically this works through a search across your documents whose results are handed to the model; the model answers from your material, not from something it once read on the internet. You can see how closely that holds on our own chatbot: ask it where we are based and the answer comes from the same maintained source that holds our Neuhausen am Rheinfall address - change that source and the answer changes with it.

That includes citing its source and saying when it does not know something instead of guessing. For cases it cannot answer properly we define a handover to a person, complete with the conversation so far, so your customer does not have to explain everything twice.

Reading documents and automating processes

Invoices, orders, delivery notes, forms, applications: AI can turn them into structured data and write it straight into your ERP, your accounting or your database. In the same way, incoming mail can be classified and routed to the right place, or long minutes boiled down to a summary someone will actually read.

The control point matters here. For anything touching money or contracts we build in a short human approval - the AI prepares, you decide. Everything the model does is logged, so you can retrace later how a result came about.

Data protection and where your data ends up

Before a line of code is written we establish which data the system gets to see at all. Personal and confidential content can often be stripped out or pseudonymised beforehand. For the rest we pick providers and regions so the processing fits revDSG and the GDPR, and we put the required data processing agreements in place.

If your data must not leave the building at all, a model on your own infrastructure in Switzerland or the EU is an option - narrower in capability, but entirely under your control. We discuss openly which route makes sense in your case, drawbacks included.

What's included

  • Review of your processes: where AI genuinely pays off
  • Prototype on your real data before you commit
  • Chatbot grounded in your own documents
  • Automated extraction from invoices, forms and receipts
  • Integration with your existing systems via APIs
  • Human approval steps at the critical points
  • Data protection concept for revDSG and GDPR, hosted in Switzerland or the EU
  • Training for your team and support after launch
Free consultation

How a project runs

  1. 01

    First consultation

    We look together at which tasks keep repeating and whether AI is the right tool for them - free and without obligation.

  2. 02

    Feasibility

    We assess the state of your data and tell you honestly what is realistic and what is not.

  3. 03

    Prototype

    A small, working build on real examples shows you the quality before budget goes into the full implementation.

  4. 04

    Build

    We integrate the solution into your systems, with logging, approvals and a fallback for when a service goes down.

  5. 05

    Testing on real cases

    Your team runs it alongside the old way for a while, we measure accuracy and errors and tune from there.

  6. 06

    Operation

    We monitor cost, quality and response times and adjust when models or prices change.

Frequently asked questions

What does an AI solution cost?

There are two parts: the one-off development and the running per-usage cost charged by the model provider. We estimate both up front against your expected volume, and for the development you get a written fixed-price quote after the first consultation.

What if the AI claims something that is wrong?

It can happen, and nobody should tell you otherwise. We limit it by grounding answers in your own documents, returning sources, and letting the bot say "I do not know". For anything that matters legally or financially, a person reviews the output before it goes out.

Will my data be used to train someone else's model?

The business tiers of the major providers usually rule that out in their terms, and those are the ones we use - not the free consumer versions. What applies in your specific case we check before the start and put in writing. If that is still not enough, we run a model on infrastructure you own - then the content never leaves your systems at all.

Do I need large amounts of data for this?

For the most common uses, no. A chatbot on your knowledge needs well-kept documents, not training data. Data volume only becomes the issue once you want to fine-tune a model to your language and your cases - which we only recommend when the simpler route demonstrably falls short.

How long does it take from idea to production?

Depending on scope, a meaningful prototype is typically a matter of a few weeks. Going live takes longer, because integration, testing and approval processes come on top. In our experience the schedule depends less on the AI than on the state of your data.