AI agents - One workflow, automated in three steps: discover, pilot, run.

Every organization runs on workflows: a request comes in, information is gathered, checked, decided on, documented, and passed on. Much of that work is still done by hand, copied between systems, re-checked, and chased by email. It is exactly the kind of work AI agents can now take on reliably.

We do not start with a strategy paper. We start with one workflow, prove on your real data that an agent can handle it, and only then put it into production. Each step has a result and a fixed price, and you decide after each one whether to continue.

Start small - Four steps, each with a result, a duration and a price

You do not buy a program. You buy one step at a time, and you decide after each one whether to take the next.

  1. Step 11 day

    AI workshop

    A ranked list of use cases with a build-or-buy recommendation for each.

    from CHF 4’500

  2. Step 22 to 3 weeks

    Pre-study

    Process map, data check, architecture and a fixed quote for the pilot.

    CHF 9’500 to 15’000

  3. Step 36 weeks

    Pilot

    A working agent on your real data, measured against the manual baseline.

    from CHF 35’000

  4. Step 4After the pilot

    Production and operation

    Integration into your systems, hosting in Switzerland and monitoring.

    Fixed quote and monthly fee

Step by step - What happens in each step

  1. Step 1

    Discover

    AI workshop and pre-study

    In a one-day AI workshop we map your processes and data together with the people who run them, and rank the possible use cases by value, risk, and data readiness. For each one you get a build-or-buy recommendation: custom software, an existing tool worth rolling out, or something to leave for now.

    In the pre-study we take the best use case and check whether it holds up. We map the process step by step, check the data, design the architecture, and give you a fixed quote for the pilot. That takes two to three weeks.

    What you get

    • Ranked list of use cases
    • Tool analysis: build or buy
    • Process map
    • Data check
    • Architecture
    • Fixed quote for the pilot

    Request a fixed-price pre-study

  2. Step 2

    Pilot

    A proof of concept on your real data in six weeks

    In six weeks we build a working agent for the chosen workflow, on your real data, in short sprints you can review. It is connected to the systems it needs, with a human approval step wherever a decision matters.

    Before we build, we record the manual baseline: throughput, turnaround time, quality, and cost per case. At the end you see the agent’s results next to those figures, and you decide with evidence whether it goes into production.

    What you get

    • Working agent on your real data
    • Human approval steps
    • Comparison with the manual baseline
    • Evaluation of answer quality
    • Fixed quote for production

    Read the AI pilot case study

  3. Step 3

    Run

    Production and operation

    We integrate the agent into your systems, host it in Switzerland, and monitor it. You get a fixed quote for the production build after the pilot, and operation runs for a monthly fee.

    AI models change every few months. As part of operation we keep testing the agent against real cases, so answer quality does not slip when a model changes. If you prefer to run the agent yourself, we hand it over to your team with documentation, runbooks, and training.

    What you get

    • Integration into your systems
    • Hosting in Switzerland
    • Monitoring and support
    • Regular tests against real cases
    • Handover to your team if you want it

    Hosting and maintenance in detail

Best fit - Who this service is for

  • Organizations where skilled people spend much of their day on repetitive, rules-driven work: intake, triage, document processing, reconciliation, reporting, or follow-ups.
  • Leadership teams that want to grow output without growing headcount at the same rate, and want a partner who is accountable for the result, not just the recommendation.
  • Teams that want to know what a first project costs and how long it takes before they commit to it.
  • Organizations in manufacturing, finance, insurance, pharma, and the public sector, where agents must work within rules such as nFADP, FINMA, or GxP, with audit trails and human sign-off where it matters.

Typical architecture - Designed for reliability, grounding, and operational handover

The exact stack depends on the problem, but these are the design principles we usually optimize for.

  • Replace, augment, or create. For every workflow we decide deliberately: agents take over the manual work from start to finish, support the people who do it, or run entirely new workflows that were never economical with human effort alone.
  • One task per agent. Each agent owns a narrow, well-defined task with its own tools and permissions. Where a workflow needs several agents, a coordination layer manages hand-offs, checks, and escalations, the way a strong team divides its work.
  • Controlled and observable. Scoped access, full logging, tests against real cases, and human approval for sensitive decisions, so every action an agent takes is traceable.
  • Running software, not slide decks. We measure ourselves by agents running in production and the output they deliver, not by workshops or strategy papers.
  • The lightest solution that works. Not every problem needs an agent. Many are better solved with a search over your documents, an existing tool, or plain automation, and we say so in the workshop.
  • Built for Swiss accountability. ETH Zürich and EPFL-trained engineers with delivery experience in banking, pharma, and the public sector, deploying agents with Swiss data residency and governance designed in from day one.

FAQs - Questions we hear early in the conversation

These are the kinds of questions that usually matter before a team commits to scope, architecture, and rollout.

What is an AI agent?

Software that handles a task from start to finish instead of only answering questions. It reads the incoming documents, looks up what it needs in your systems, drafts the result, and brings in a person wherever judgment is needed.

Which workflows are a good fit?

Any recurring workflow with clear inputs, steps, and outcomes: case and request handling, document processing, research and reporting, quality checks, and customer and supplier communication. If your people can describe how the work gets done, agents can usually take on a significant share of it.

What does it cost to get started?

The AI workshop takes one day and starts at CHF 4’500. A pre-study takes two to three weeks and costs CHF 9’500 to 15’000. A six-week pilot starts at CHF 35’000. For production you get a fixed quote after the pilot, and operation runs for a monthly fee.

Do we have to start with the workshop?

No. If you already know which workflow you want to automate, start with the pre-study. If you already have a pilot or proof of concept built by someone else, our prototype-to-production review is the right entry point.

Who should join the AI workshop?

A mix of decision-makers, process owners, and team leads usually works best. No AI knowledge is needed: we adapt the depth to the people in the room, and the examples to your workflows, constraints, and available data.

Do all AI use cases need agents?

No. Many problems are better solved with a simpler search or automation, or with a tool that already exists. We help you choose the lightest solution that still delivers the result.

Will agents replace our employees?

Agents take over the repetitive parts of the work. In most organizations that means people shift from doing routine tasks to supervising, deciding, and handling exceptions, and the organization can take on work it previously could not afford to do.

How do you keep agents safe and under control?

Every agent works with scoped permissions, every action is logged, and sensitive steps require human approval. We test agents against real cases before go-live and keep monitoring them once they are in production.

Who runs the agents once they are live?

Your choice. We can host, monitor, and keep improving them for a monthly fee, or hand them over to your internal team with documentation, runbooks, and training.

Do you also train employees and developers?

Yes. Besides the use-case workshop we run a hands-on session on working with ChatGPT and other AI tools for employees, and a developer workshop on reliable AI coding: review and verification workflows, and guardrails for architecture, security, and technical debt.

Related proof - Case studies, articles, and next steps

If this is close to what your team needs, these pages are the best next places to look.

Hosting and maintenance

What the Run step covers in detail: hosting, monitoring, patching, backups, and support.

From prototype to production

You already have a pilot built by someone else? We check it, fix it, and run it.

AI pilot case study

How we built and evaluated AI pilots inside regulated organizations.

Find your first workflow

Start with a short conversation about the workflow where an agent would make the biggest difference.

Which workflow should we start with?

Our office

  • HQ
    Hohlstrasse 206
    8004 Zürich, Switzerland