Before automating government paperwork, fix the data going into it

Digital minister Karsten Wildberger has promoted Spark Workflow as a way to accelerate approval procedures by up to 50%. In an August interview, open-data specialist Stefan Kaufmann questions the public evidence behind that promise.

The workflow converts Word documents, PDFs, and presentations into Markdown, then sends the contents through language-model prompts to assess completeness, plausibility, and legal classification. Kaufmann says the available documentation lacks a clear overall account of the system, a published quality test, and agreed criteria for acceptable errors.

His alternative starts further upstream: collect information in structured, machine-readable forms with explicit meanings. A field called income, for instance, needs the correct definition for the administrative context. Clear data and explicit rules can make many operations repeatable and inspectable.

The criticism is about the foundations of the proposed modernization. A model may help process today’s document pile, while investment in better data changes what tomorrow’s administration has to process at all.