Kela AI-Enabled Operational Support and Document Processing Platform (Finland)

The Social Insurance Institution of Finland (Kela) has developed and deployed an AI-enabled platform that supports operational processes across its benefit administration system. Kela is responsible for settling approximately 15.5 billion euros of benefits annually across roughly 40 core social insurance programmes, including health insurance, pensions, unemployment benefits, child benefits, student financial aid, housing allowances, and basic social assistance (AlgorithmWatch, 2020; Kela, 2025). The institution processes millions of benefit applications and related documents each year, with most applications submitted through the OmaKela online self-service platform (Kela, 2025). In many cases, applications require additional documentation such as rental contracts, bank statements, and medical certificates that must be reviewed and classified before benefit specialists can process claims (Solita, 2023). Kela's AI platform centres on an attachment pre-processing system that automatically prepares documents submitted by customers and partner organisations for benefit processing. The system identifies document types, converts file formats, edits images including colour-to-greyscale conversion and rotation to ensure correct orientation, and ensures that documents are in the correct format for benefit specialists to review (Kela AI Register, 2025). All processing of attachments takes place within Kela's own IT infrastructure, with no data transferred externally (Kela AI Register, 2025). In addition to document pre-processing, Kela has deployed several complementary AI applications. The Kela-Kelpo and FPA-Folke chatbots assist customers in finding information about benefits on Kela's website and within OmaKela, using natural language processing to match customer questions to pre-written responses authored by Kela specialists; the chatbot system does not generate content autonomously (Kela AI Register, 2025; ISSA, 2024). A customer feedback analysis tool uses a language model to identify key themes and areas for development from open-ended customer feedback (Kela AI Register, 2025). A call transcription tool converts customer service call recordings to text using AI-assisted speech technology, operating within Kela's internal IT environment (Kela AI Register, 2025). A machine translation service developed by Tilde translates customer and partner documents into Finnish for international cases and debt recovery, with all translations verified by humans before use in decision-making (Kela AI Register, 2025; Slator, 2023). Kela's approach to AI governance is notably cautious and well-documented. The institution explicitly prohibits the use of machine learning, statistical modelling, or scientific modelling in automated benefit decisions, stating that logical rules based on machine learning or statistical or scientific modelling are not used to make automated decisions (Kela, 2025, Automated Decisions page). Automated decisions are restricted to routine tasks where outcomes are completely indisputable, requiring unambiguous legislative application, factual clarity, and no discretionary judgement (Kela, 2025). Kela published an AI register in 2025 listing all AI applications in use, and updated its principles for responsible AI use in autumn 2025 to align with EU AI Act requirements (Kela AI Register, 2025). In 2019, Finland's Chancellor of Justice initiated a formal investigation into Kela's automated decision-making processes following complaints about unemployment benefit communications, examining good governance requirements, official accountability, citizen legal protections, and transparency (AlgorithmWatch, 2020). Kela has also partnered with Solita beginning in early 2023 to explore broader AI opportunities in social services through a strategy-first approach. This partnership has identified candidate application areas and management practices for AI initiatives, with the project operating in an experimentation phase involving small-scale AI deployments (Solita, 2023). The institution operates under Finland's Act on the Openness of Government Activities, the Data Protection Act (2018), and national AI ethics guidelines consistent with the EU AI Act (2024). Kela's benefit administration legislation spans over 200 separate regulations set in place over 30 years across 6 ministries, with different laws containing more than 20 interpretations of income, creating significant complexity for any automation or AI initiative (AlgorithmWatch, 2019). Research during enrichment found no evidence for the v1 characterisation of this case as a predictive operational forecasting system. The case therefore reflects Kela's actual documented AI applications, which focus on operational support through document processing, customer service automation, translation, transcription, and internal efficiency tools rather than predictive workload forecasting.