Three challenges in implementing artificial intelligence: how to avoid “shadow AI,” where real innovation emerges, and how not to lose know-how during generational turnover. Experience from business shows that regulated environments, collaboration, and effective knowledge transfer are key. These insights are applicable to public administration as well.
Innovation emerges at the intersection of business and AI
Success comes when process owners, who know their use cases and can set KPIs, meet AI/IT experts who know what will work technically. At the insurer Kooperativa they set a goal: the policyholder photographs medical reports and the claims adjuster receives a prepared proposal for resolving the case. The system today processes a claim in an average of 12 seconds, while a human standard is roughly 20 cases per day; everything is explainable and auditable (e.g., identifying a “sutured wound with complications” based on mentions of stitches and antibiotics).
In drug development, a top Prague research center (ÚOCHB) needed to explore a space of tens of billions of molecules and model their interaction with proteins. A trained language model predicts properties, binding, and toxicity, increasing the chance of selecting several dozen promising candidates for laboratory tests and ultimately for a patent. Another approach uses OSINT: scanning package inserts across markets reveals new opportunities, for example discovering that there is a successful drug combination in Japan that is missing in Europe.
Demographics and knowledge transfer
Experienced workers are leaving and new ones need to be trained quickly, otherwise knowledge loss threatens. On robotic lines, anomalies are monitored (e.g., torque curves), and when quality drops a voicebot calls a senior to describe the problem and the solution (“two screws are firing, I lowered the vacuum”). From these calls the system automatically compiles a shift log, categorizes problems, and creates playbooks and training materials for juniors.
Such tools are practical outside manufacturing as well: for example, creating a terminological glossary for the automotive industry arose precisely thanks to systematic collection and structuring of knowledge via a voice assistant. The lesson for companies and public administration is the same: a regulated environment, measurable goals, and actively capturing people’s experience are the foundation for AI to deliver real results rather than just be a fashionable accessory.
