The typical client
An SME from Germany, Austria or Switzerland with 10 to 250 employees and a process with high volume and a lot of text: quotes in wholesale, service reports in the trades, receipts and incoming mail at fiduciaries and property managers, customer enquiries in an industrial service department. There is no in-house data team, and there should be no need for one.
Typical use cases
Inbox and customer service: classify emails, route them and draft replies. Quotes: read enquiries and draft quotes from the price list and earlier quotes. Supplier invoices: capture receipts, suggest account coding, flag deviations from the order. Knowledge search: answer questions about manuals, contracts and tickets, always with a source. Order handling: steer steps with deadlines and escalation. Reports and minutes: summarise and derive tasks. In every case a person decides at the agreed point, for example when approving a reply or a payment.
What you have after three weeks
Today's process as a process model with volumes, lead time and effort per case. An assessment of the use cases considered, with benefit, risk and effort. A pilot in live operation with approval by your staff. The before-and-after measurement: time per case, error rate and share of drafts approved without correction. A short AI policy, an entry in the AI inventory and training for the pilot team. And a decision paper on scaling up, adjusting or stopping, with the cost of operation and expansion.
You choose the model
You choose the language model: one you already license, a provider of your choice, or an open model on hardware in Switzerland or the EU. Process, data and audit trail stay on our platform in the EU. The AI reaches your systems only through a gateway with sign-in and per-tool permissions, and every action is logged.
Built with what we use ourselves
The pilot runs on a process engine (BPMN) that handles deadlines, escalations and approvals and records every step in its history. There are ready-made connectors for email, ERP, file storage and tickets, and a knowledge search that always answers with a source and shows only what the person may see. Our own financial management has run on these building blocks since August 2026: key figures from accounting, payments and tax portals are consolidated every hour without a human step.
AI literacy under the EU AI Act
Since 2 February 2025 every company that uses AI must ensure a sufficient level of AI literacy among its staff. This applies directly to SMEs in Germany and Austria, and to Swiss SMEs as soon as their AI system is used in the EU or its output is used there. The sprint provides the pilot team's training and a short policy as evidence.
Three weeks, three decisions
Week 1: process and use case
Kick-off, mapping the process, measuring today's effort on real cases, assessing the candidates. On Friday you decide: continue, or stop at half the price.
Week 2: build the pilot
Connecting your systems, setting the approval points, testing with cases from recent weeks.
Week 3: pilot in operation
The pilot team works with real cases while we measure and adjust. Close-out with the measured result, the rules and the decision paper.
Frequently asked questions
Which language model do you use?
You decide. We connect your existing model, a provider of your choice, or an open model in Switzerland or the EU.
What if no use case pays off?
Then you stop after week 1 at half the price and keep the process map and the assessment.
Does the AI make decisions about people?
No. The AI Act classes candidate screening and credit decisions as high-risk; they are not part of the sprint. In every pilot a person approves at the agreed point.
What does operation cost afterwards?
That is in the decision paper, with a fixed price for expansion and the cost of operation and model. You decide only then.