Federal Administration AI: Efficiency promised, savings unproven
Process automation and artificial intelligence are meant to simplify federal procedures. The new report relies on a stocktake and prioritised applications; the strategy only turns into a productivity gain once results can be measured.
When people think about artificial intelligence in administration, they usually picture a chatbot. For the Federal Council, the bigger lever lies elsewhere: in recurring processes, documents, data matching and internal workflows. A new report on the implementation of postulate 24.3582 by Andri Silberschmidt describes how process automation and AI could relieve the Federal Administration.
The goal is ambitious but not unusual: procedures should become simpler and faster, quality should improve, and staff resources should be freed where they achieve more. What matters, then, is whether the technology is deployed in the right places.
Understand first, then automate
Automation does not improve a bad rule. Where data is incomplete, responsibilities change hands repeatedly or exceptions are poorly regulated, AI can only spread mistakes faster. The report therefore starts with transparency: existing systems and applications are to be recorded, merged into a portfolio and prioritised according to expected benefit.
The federal government is not a single organisation with one uniform data pool. Departments and offices carry out different tasks, work under particular legal bases and in some cases process especially sensitive information. An application that saves time when searching public documents is not automatically suitable for a decision with legal effect.
Suitable use cases might include sorting standardised inputs, supporting translations, searching large collections of files or preparing data. In every case the same rule applies: humans must retain responsibility wherever a mistake can affect citizens' rights, benefits or obligations.
Productivity needs a yardstick
Public debate jumps quickly from "AI is being deployed" to "AI saves money". That conclusion does not follow automatically. New systems initially cause costs: for procurement, secure infrastructure, data maintenance, training, external expertise and controls. Converting workflows also ties up time.
Real efficiency can therefore only be measured against a clear baseline. How long did a process take before? How high was the error rate? Which work disappears after introduction, and which new control work is added? Without such metrics, a productivity promise is a plausible expectation, but not a proven saving.
The Swiss Federal Audit Office has pointed to risks in governance, data and transparency when AI is used. Administrative action must remain traceable, lawful and verifiable for those affected.
A balancing act between speed and trust
Success depends less on which model the administration uses than on a shared architecture: clear responsibilities, binding security requirements, documented data sources, procurement standards and a body that monitors quality. Where systems affect the public, comprehensible information and ways to file complaints are needed as well.
The report therefore marks a decision on direction, not an efficiency balance sheet. Whether it actually leads to shorter procedures and relief for the federal budget will only become clear through concrete, measurable applications.
Sources
- Confederation: report on process automation and AI: https://www.sust.admin.ch/de/newnsb/NJ04UgpFFUoA
- Postulate 24.3582 Silberschmidt: parlament.ch/de/ratsbetrieb/suche-curia-vista/geschaeft?Aff…
- Swiss Federal Audit Office: use of artificial intelligence: efk.admin.ch/wp-content/uploads/publikationen/berichte/wirt…
- Confederation: overview of digitalisation: bk.admin.ch/bk/de/home/digitale-transformation-ikt-lenkung.…
