1. Map
We identify, with your team, one real workflow that causes friction today. We map the authoritative sources, critical decisions, user roles, existing tools, and the place where missing context creates measurable cost. We define success criteria, stop conditions, and what we will not change.
- One workflow, not the whole enterprise.
- Authoritative sources with permissions and owners.
- Clear success criteria, measurable in weeks.
- Stop conditions, defined before we start.
2. Connect
We build the retrieval and memory layer on top of your existing systems. We define permissions, citations, freshness rules, and model routing. We install the AI gateway shell so every call is metered, budgeted, and observable.
- Source readers against Confluence, Notion, SharePoint, GitHub, PDF, Office.
- Memory layer with citations, freshness, RBAC.
- Gateway with LiteLLM, OpenRouter, or Portkey.
- Budgets, alerts, runaway-loop protection.
3. Pilot
A small user group gets a controlled agent workspace and evaluates real tasks on quality, continuity, cost, and risk. We evaluate with real tasks, not demo data.
- Real tasks from the real workflow.
- Real users, real acceptance, real cost.
- Weekly evaluation, not quarterly.
- Stop condition is checked, not ignored.
4. Operate
We harden the successful workflow. Behaviour and cost are monitored. Weak context is iteratively improved. Expansion happens only where the evidence from phase 3 supports it.
- Hardening against drift, model updates, load spikes.
- Cost and quality dashboards per workflow.
- Quarterly re-evaluation, not only at the start.
- Expansion by evidence, not by sales pressure.
Frequently asked questions
How long does a pilot take?
2-4 weeks for a bounded pilot. Longer pilots are usually a sign the scope is too large.
What happens if the pilot does not work?
Stop condition triggers, you keep the source code, and we recommend the next iteration. No sales pressure.
Is data kept in Switzerland?
By default yes, with Swiss-resident inference. Where that is not possible, we document the residency explicitly.
Can we take over after the pilot?
Yes. We ship on a portable runtime, document the deployment, and your team takes over after the pilot.
Related services
Next step
Start with a bounded pilot.
Tell us about your workflow. We will respond within one business day with a Map session.
Request a discovery session