Ernst — Trelleborg's Automated Welfare Caseworker Cuts Social-Assistance Decisions From Days to Minutes

Trelleborg became the first Swedish municipality to fully automate decisions on reapplications for social assistance (försörjningsstöd), digitalising benefit administration from 2015 and introducing a rules-based automated case handler, nicknamed 'Ernst', from 2016. The system pulls data from seven agencies — including the tax authority and student loan board — compares each household's income and expenses month to month, and either approves, adjusts or forwards the case to a human caseworker when circumstances have changed too much for the rules to cover.
Independent evaluators reported sharply faster processing: the OECD's Observatory of Public Sector Innovation found overall case-handling time fell from roughly 10 days to within 24 hours, while automation vendor UiPath's published case study puts the improvement at 94%, with applications that previously took an average of 8 days (up to 20 for new claims) now decided in one minute or less, and reports the municipality served 22% more citizens than the previous year using 2 fewer front-line staff on repetitive processing. By the OECD's account, roughly 70% of Trelleborg's social-benefit applicants now use the online channel the automated system relies on. Vinnova, Sweden's innovation agency, and the Swedish Association of Local Authorities and Regions used Trelleborg's model as the basis for a rollout to 14 further municipalities.
The same system also illustrates the risks of automating welfare decisions without full transparency: independent reporting by AlgorithmWatch's Automating Society project found the decision logic ran to 136,000 lines of rules across 127 XML files, difficult for outsiders or affected citizens to audit, and separately found that roughly 250 citizens' names and national ID numbers had been left exposed in the source code since 2017.

Read the full analysis: https://ai-watch.github.io/AI-watch-T6-X/service/90131.html