The Intelligent Rehabilitation Recommendation System (IRRS) is an AI-based decision support tool developed and operated by the Korea Workers' Compensation and Welfare Service (COMWEL), the public agency responsible for administering the Republic of Korea's Industrial Accident Compensation Insurance (IACI) system. The system was deployed in February 2020 and is designed to identify injured workers who would benefit from proactive rehabilitation intervention, predict their return-to-work pathways, assess likely disability severity, and recommend personalised rehabilitation service plans aligned with each worker's treatment stage.
COMWEL has operated a Customized Integrated Services (CIS) programme since 2011, which establishes individualised rehabilitation plans for workers injured in industrial accidents and coordinates systematic rehabilitation services — including medical rehabilitation, vocational rehabilitation, psychosocial support, and return-to-work assistance — according to those plans. Over the decade of CIS operation, COMWEL accumulated a substantial administrative dataset comprising approximately 98 million worker records drawn from workers' compensation claims, unemployment insurance records, rehabilitation case management histories, and employment data. The IRRS was developed to leverage this accumulated big data resource through AI techniques, replacing what had previously been a largely manual and experience-driven process for selecting workers for rehabilitation and designing their service packages.
Technically, the IRRS employs a hybrid analytical approach combining rule-based filtering and case-based reasoning (CBR). In the first stage, the system applies rule-based filters to the population of injured workers to identify those meeting threshold criteria for rehabilitation vulnerability. A vulnerability index is calculated for each worker based on multiple variables extracted from the accumulated administrative data, including injury characteristics, treatment history, employment history, demographic factors, and prior rehabilitation outcomes. Workers scoring above defined thresholds on the vulnerability index are flagged as candidates for proactive rehabilitation outreach. In the second stage, the system applies case-based reasoning methodology, drawing on the database of past rehabilitation cases and their outcomes to identify the most relevant historical precedents for each candidate. Based on the similarity of the current worker's profile to successful past cases, the IRRS generates a recommended rehabilitation plan specifying which services — from among COMWEL's portfolio of medical, vocational, and psychosocial rehabilitation offerings — are most likely to support that worker's return to work.
The IRRS is not an autonomous decision-making system. Rehabilitation plans generated by the AI model are presented as recommendations to COMWEL's human rehabilitation specialists, who review the AI-generated suggestions during consultation sessions with the injured workers. The rehabilitation experts retain authority to accept, modify, or reject the system's recommendations based on their professional judgement and the worker's own preferences and circumstances. This human-on-the-loop oversight model ensures that the AI functions as a decision support tool rather than an automated decision-maker.
In its first year of operation (2020), the IRRS recommended 13,876 rehabilitation services to 2,637 injured workers. Of these recommended services, 66 per cent were reported as customised — meaning they were tailored to the individual worker's specific circumstances and rehabilitation needs rather than being standard packages. This customisation rate represented a significant improvement over the previous manual selection process.
The system's deployment has been associated with incremental improvements in COMWEL's overall return-to-work rates for injured workers: 65.8 per cent in 2019 (pre-IRRS), rising to 66.3 per cent in 2020 and 67.3 per cent in 2021. While these improvements cannot be solely attributed to the IRRS given the multiple factors affecting return-to-work outcomes, COMWEL has cited the system as a contributing factor in the upward trend.
The IRRS was developed through the ICT-based Public Service Promotion Project administered by Korea's Ministry of Science and ICT and the Korea Intelligent Information Society Promotion Agency (NIA). COMWEL applied for and was selected for this government-funded initiative in 2019, which provided the framework and resources for the system's development. Patent applications were filed for the system's methodology.
The system received international recognition when COMWEL was awarded the ISSA Certificate of Merit with Special Mention at the 2021 World Social Security Innovation Awards, administered by the International Social Security Association (ISSA). The ISSA recognised the IRRS as an exemplary application of fourth industrial revolution technology — specifically AI and big data analytics — to workers' compensation rehabilitation administration, describing it as the first such application in the Republic of Korea's industrial accident compensation system.
COMWEL has pursued further development of the system. A 2022 follow-up project aimed to integrate job placement and vocational training recommendations into the existing IRRS framework, expanding its scope from rehabilitation service recommendation to encompass the full return-to-work pathway including employment matching. This expansion reflects COMWEL's broader strategy of applying AI and big data across the workers' compensation lifecycle, from injury assessment through treatment, rehabilitation, and return to work.
The system's data foundation — built on over a decade of administrative records from Korea's mandatory industrial accident compensation insurance system — provides a robust basis for the case-based reasoning approach, as the large volume of historical cases and outcomes enables meaningful pattern matching. However, as with any system relying on historical data, there is an inherent risk that past patterns may not fully capture evolving workplace conditions, emerging injury types, or changing labour market dynamics.