Between the idea and a running assistant lie decisions that are hard to correct later. This category covers the technical and organisational frame: embedding into the existing website, connecting to merchandise management, booking or ticketing systems, preparing the knowledge base, handling updates, test procedures before going live and quality assurance in operation. We look at running costs and dependencies, at roles and responsibilities in the project and at how benefit can be evidenced rather than claimed. The goal is projects that start with a clearly defined use case and are then guided by measured results. We additionally describe which test questions to ask before going live — and which answers should stop a launch.
Timeouts, maintenance windows, partial outages: how an AI assistant steps down in stages, shows a visible fallback path and logs every single incident.
Behind the login the AI assistant answers specifically, not generally: bind session and permissions, scope every query, confirm each write action first.
41 percent of German companies now use AI (Bitkom). How an internal AI assistant serves company knowledge, what co-determination requires and how to pilot it.
An AI assistant rarely answers twice the same way. How to accept it anyway: test cases from real enquiries, release criteria, rubrics and regression runs.
What does an AI chat assistant cost, what does it return and when does it pay off? The sober business case with costs, revenue levers and a worked example.
How to embed an AI assistant cleanly into your website or shop: a lean widget instead of heavy bundles, first-party and without hurting your load time.
Why a language model invents facts and how a knowledge base with RAG binds answers to your content: source binding, confidence thresholds, honest handover.