Germany's Statistisches Bundesamt Trains a Machine-Learning Model to Estimate Regulatory Compliance Costs

Every new German federal regulation must be accompanied by an estimate of the 'Erfüllungsaufwand' — the time and cost that citizens, businesses and administration will bear in complying with it. For low-cost rules (an expected burden under €100,000 a year for businesses), ministries use a 'vereinfachtes Verfahren' (simplified procedure) rather than a full case-by-case calculation. The Statistisches Bundesamt (Destatis) trained a new machine-learning estimation model to make that simplified procedure more accurate, described in a peer-reviewed methodology article by Levagin, Lange, Walprecht, Gerls and Kühnhenrich published in the agency's own journal WISTA – Wirtschaft und Statistik (Vol. 74, No. 3, 2022, pp. 53–67) and revisited in a 2024 WISTA retrospective on machine learning at the agency.

Destatis states the ML-based estimator 'increases the quality of compliance-cost estimates going forward' compared with the prior, simpler calculation approach.

The model is used internally by Destatis staff who support the regulatory-impact-assessment process for federal ministries, rather than being a public-facing tool. The OECD's 2025 'Governing with Artificial Intelligence' report cites the German approach as an example of AI being used in regulatory design and delivery.

Neither the original 2022 article nor the 2024 follow-up discloses a specific accuracy figure or error-rate reduction in the material available for this review, so the claim of improved accuracy rests on the agency's own methodological account rather than an independent, quantified evaluation. Because Destatis is the same institution that ministries rely on to sign off compliance-cost figures, the practice sits inside an existing, long-standing government process rather than being a freestanding pilot — but that also means it has not been externally audited as an AI system in its own right.