AI and automation that take work off your desk.

That depends on the workflow: how many systems are involved, how many exceptions there are and how reliable the result has to be. Flat-rate prices would therefore not be credible. The main cost drivers are connecting to your systems and preparing your data; the usage fees of an AI model are usually the smaller item. After the no-obligation intro call you receive a written quote with a clear scope. It also states the running costs for operation and maintenance. Where an AI model is involved, its usage-based fees are added, which we estimate in advance for your volumes.

Ideally with a workflow that occurs frequently, follows a recurring pattern and currently involves noticeable manual effort. It also matters that a mistake is caught before it causes harm. Typical candidates are the inbox, incoming invoices or reconciling data between two systems. We deliberately start with one self-contained workflow instead of tackling everything at once, so you see early on whether the approach works. How long it takes to get a first result depends on the scope; we give you a realistic timeframe in the quote.

We decide that together before work begins. There are essentially three routes: models from large US providers via their business interfaces, models from European providers, or open models in an environment you control. With the third route your inputs are not sent to any model provider, but the operating effort increases. For their business interfaces, OpenAI, Google, Microsoft and Anthropic commit not to use inputs to train their models without consent. That does not mean nothing is stored: depending on the provider, inputs are retained for a limited time, for example to detect abuse. Providers commit to processing within the EU for certain models and settings only. We check this for each project individually. Where a provider is subject to US law, access by authorities cannot be ruled out entirely, even with processing in the EU. Before the first step, we record in writing which route suits your data and which contracts are needed, such as a data processing agreement. (As of October 2026)

Language models can produce answers that sound convincing and are still wrong. That is why we plan for errors from the outset: results are checked against your master data and fixed rules, unclear cases go to a person, and before roll-out we use your real cases to test how often the system gets it right. Where a mistake would be expensive, the AI only suggests. Your staff make the decision. Because AI systems can be manipulated through crafted documents or e-mails, we also limit what they can access and what they may trigger on their own.

For many typical office applications the obligations are manageable, but they exist. Anyone using AI professionally must take measures to support the AI literacy of the staff involved; the regulation prescribes neither a specific level nor a certificate. Since 2 August 2026, systems that interact directly with people (a chatbot, for example) must be recognisable as AI, unless that is obvious anyway. High-risk applications, such as AI in the selection of job applicants, are subject to strict requirements; these obligations apply from 2 December 2027. We classify your project before work begins and document how the system works. This does not replace legal advice. In Germany, the point of contact for companies is the AI Service Desk of the Federal Network Agency (Bundesnetzagentur). (As of October 2026)

For individual tasks, yes. But a chat window does not automate a workflow: someone has to copy, paste and transfer the result. Then there is the data question: with private accounts it is often unclear what happens to customer or personnel data that is entered. Germany's data protection authorities therefore recommend company accounts and clear internal rules. A solution integrated into your systems works with your data, by your rules and with the contracts you need for it. Both make sense: rules for chat use, and automation for what recurs every day.

A lot. Since 2025, companies must be able to receive e-invoices for domestic business transactions in Germany. The transition period for issuing them ends on 31 December 2026, or one year later for companies with a prior-year turnover of up to 800,000 euros. An e-invoice is a structured data record, not a picture of an invoice: it can be checked, matched and passed to accounting automatically, without AI. As long as PDFs and paper keep arriving too, two routes run in parallel. That is exactly where merging both into one workflow pays off. (As of October 2026)

In most cases, yes. What matters is whether your software offers an interface: an API, file export and import, database access or at least an e-mail inbox. Systems we often connect are inventory management and ERP, accounting, CRM, shop systems and Microsoft 365 or Google Workspace. We check which routes your system offers during the process review, before we commit to anything.

We automate work steps: retyping, sorting, searching, following up. Whether a solution holds up in daily work depends on the people who know the workflow. That is why we involve them from the process review onwards, test with them on real cases and train them on the finished system. That training is also one of the measures by which you support the AI literacy of your staff, as the EU AI Act requires.

Automations need upkeep: interfaces change, AI models are replaced by their providers, your workflows evolve. On request we take care of operation, monitoring and adjustments. You have a dedicated contact who knows your solution.