Knowledge · As of 2026-09-07
How an AI project runs: from idea to production system
The four phases
Phase 1, assessment (1-2 days): where do we stand, which use cases pay off, what is the next step - answered in a structured KI-CHECK. Phase 2, data check (days): a readiness assessment across six dimensions; a red light is not a knock-out but the moment a data roadmap saves the project. Phase 3, implementation (2-12 weeks): plannable with a fixed price, weekly increments, and a guaranteed go-live - a simple website assistant takes 2-4 weeks, enterprise environments with SSO 8-12. Phase 4, operations: models get retired, APIs change, the EU AI Act adds obligations - budget 10-20% of the setup price per year.
The three classic failure modes: building without an assessment, ignoring the data foundation, and skipping team enablement - the latter being mandatory since February 2025 (Article 4 EU AI Act) and largely fundable.
Frequently asked questions
How fast do we see results?
The assessment delivers a solid result after 1-2 days. A first production system takes 2-4 weeks depending on scope - not months.
What do we need to contribute?
One accountable person with decision authority, access to the people who know the processes, and - depending on the use case - access to the systems to connect. No AI expertise required.
Read next
Start with the KI-CHECK →KI-PILOT: implementation packages →