KOR-003

CLOVA CareCall – AI-Enabled Welfare Check-In Calls for Elderly Living Alone

Download PDF
Korea, Rep. East Asia & Pacific High income Scaled & Institutionalised Confirmed

NAVER Cloud Corporation; various municipal governments (Seoul, Busan, Daegu, Gwangju, Gyeonggi, and others)

At a Glance

What it does LLMs for content creation, transformation and modality conversion — User communication and interaction
Who runs it NAVER Cloud Corporation; various municipal governments (Seoul, Busan, Daegu, Gwangju, Gyeonggi, and others)
Programme Municipal Elderly Welfare Check-In Programmes (Various Local Governments)
Confidence Confirmed
Deployment Status Scaled & Institutionalised
Key Risks Model-related risks
Key Outcomes 44.
Source Quality 5 sources — News article / media, Report (multilateral / development partner), Working paper / technical note, +1 more

CLOVA CareCall is an artificial intelligence-powered telephone service developed by NAVER Cloud Corporation that conducts routine welfare check-in calls to elderly individuals living alone and other vulnerable single-person households across South Korea. The system uses large language model technology to initiate automated phone conversations that monitor the health, wellbeing, and safety status of recipients, with the primary objective of preventing solitary deaths among isolated elderly populations and enabling a shift from reactive to preventive welfare administration.

The service is built on NAVER's proprietary HyperCLOVA X, a Korean language-specific multilingual large language model that represents South Korea's first hyperscale AI system. HyperCLOVA X was specifically adapted for the CareCall application through fine-tuning on large-scale conversational datasets and the application of unlikelihood training techniques to reduce off-topic responses. The system is designed to simulate natural human conversation and emotional warmth, having been trained on tens of thousands of question-and-answer scenarios. A distinctive technical feature is the 'Remember' function, which enables the AI to retain and utilise information from past dialogues across multiple interactions, allowing for continuous and contextually aware engagement with each individual user over time. The system is capable of handling diverse speech patterns including unclear pronunciation, regional dialects, and colloquial expressions commonly encountered among elderly Korean speakers.

CLOVA CareCall was originally deployed in 2021, predating ChatGPT by approximately one year, making it one of the earliest large-scale deployments of LLM technology in a public health context globally. The service was initially developed during the COVID-19 pandemic for fever monitoring purposes and was subsequently repurposed for broader elderly welfare check-ins and emotional support services. Beta testing commenced in the Haeundae district of Busan, where after one month of operation, 90 out of 100 enrolled seniors requested to continue using the service. The system was offered free of charge by NAVER during the trial phase.

The service operates through a two-tiered workflow. First, the AI system initiates regular automated phone calls — typically twice weekly — to registered elderly users living alone. During these calls, the system checks on health status indicators including sleep quality, meal consumption, physical discomfort, and exercise levels through natural conversational dialogue. Second, social workers and welfare officials receive the call recordings along with AI-generated analyses that flag abnormalities or distress signals, enabling immediate confirmation and prompt intervention where necessary. This human oversight model ensures that welfare officers can identify signs of health deterioration, emotional distress, or emergency situations flagged by the AI system and take appropriate action.

As of 2024, CLOVA CareCall has been implemented across approximately 150 institutions throughout South Korea, including municipalities in Seoul, Gyeonggi, Busan, Daegu, Gwangju, North Jeolla, Gangwon, and South Chungcheong provinces. The service serves an estimated 50,000 older adults. A joint study conducted by NAVER Cloud and Yonsei University's ESG and Business Ethics Research Center, titled the 'Naver CareCall Social Value Measurement' report and led by Professors Ho-young Lee and Young-seok Bang, analysed publicly available municipal data comparing service regions with non-service areas. The study found that CLOVA CareCall was correlated with a 44.2 percent decrease in solitary deaths in localities where the programme operates, a 9.2 percent reduction in emergency room visits, and a 1.5 percent increase in general hospital visits — the latter attributed to earlier detection of health issues prompting users to seek preventive care before conditions worsen. The estimated annual social value of the service is approximately KRW 34 billion (approximately USD 22.7 million), with projections suggesting that extending the service to 20 percent of South Korea's elderly population could generate preventive medical benefits valued at KRW 417.2 billion annually.

User engagement and satisfaction rates have been high. Sources report that 96 percent of all users respond to CLOVA CareCall's calls and share daily greetings, with internal surveys showing an average user satisfaction rate of approximately 90 percent nationwide. There are documented cases where the system has directly contributed to saving lives in emergencies; for example, in Suncheon City, a welfare officer identified signs of health abnormalities in a user's speech patterns during an AI-generated call analysis and successfully facilitated emergency medical intervention for a cirrhosis patient.

The system was developed by NAVER Cloud Corporation in collaboration with the Seoul National University AI Policy Initiative (SAPI). The research underpinning the system has been published at leading academic venues, with papers presented at NAACL 2022 and EMNLP 2022, and the system was awarded Best Paper at ACM CHI 2023 for research examining the benefits and challenges of deploying conversational AI leveraging large language models for public health intervention.

Data privacy measures include explicit consent requirements for the collection of personal information, with separate consent procedures for sensitive health-related information, and the use of advanced encryption techniques for voice data storage. However, the probabilistic nature of the underlying LLM technology presents ongoing challenges for response control and consistency. Developers have noted inherent difficulties in maintaining absolute control over LLM outputs, and local government requests for domain-specific modifications — such as incorporating dementia screening questions — require generating entirely new training datasets, a resource-intensive process that cannot guarantee consistent inclusion of targeted queries. Academic observers have cautioned that leaving senior care entirely in the hands of AI would be inappropriate, emphasising that the technology should supplement rather than replace human interaction and oversight.

Classifications follow the DCI AI Hub Taxonomy. Hover over field labels for definitions.

Social Protection Functions

Implementation/delivery chain
Case management primaryOutreach/communications/sensitisation
SP Pillar (Primary) The social protection branch: social assistance, social insurance, or labour market programmes. Social assistance
Programme Name Municipal Elderly Welfare Check-In Programmes (Various Local Governments)
Programme Type The type of social protection programme, classified under social assistance, social insurance, or labour market programmes. View in glossary Other
System Level Where in the social protection system the AI is applied: policy level, programme design, or implementation/delivery chain. View in glossary Implementation/delivery chain
Programme Description Local government welfare programmes targeting elderly individuals living alone and vulnerable single-person households. CLOVA CareCall is deployed by municipal authorities across South Korea as a tool for automated welfare monitoring, supplementing existing social worker-led outreach and community care systems.
Implementation Type How the AI output is produced: Classical ML, Deep learning, Foundation model, or Hybrid. Affects validation, compute requirements, and governance profile. View in glossary Foundation model
Lifecycle Stage Current stage in the AI lifecycle, from problem identification through to monitoring, maintenance and decommissioning. View in glossary Monitoring, Maintenance and Decommissioning
Model Provenance Origin of the AI model: developed in-house, adapted from open-source, commercial/proprietary, or accessed via third-party API. View in glossary Commercial/proprietary
Compute Environment Where the AI system runs: on-premise, government cloud, commercial cloud, or edge/device. View in glossary Not documented
Sovereignty Quadrant Classification of data and compute sovereignty: I (Sovereign), II (Federated/Hybrid), III (Cloud with safeguards), or IV (Shared Innovation Zone). View in glossary Not assessed
Data Residency Where the data used by the AI system is stored: domestic, regional, or international. View in glossary Not documented
Cross-Border Transfer Whether data crosses national borders, and if so, whether documented safeguards are in place. View in glossary Not documented
Is Agentic Whether the system autonomously plans and executes multi-step workflows, selecting tools and chaining actions with limited human intervention. View in glossary Partial
Agentic Pipeline Description of the chained workflow steps in the agentic pipeline. Automated outbound phone calls initiated by the system on a scheduled basis; AI conducts full conversational interaction autonomously; flagged results are routed to human welfare officers for review and action.
Agentic Autonomy Degree of autonomy: fully autonomous, semi-autonomous (human checkpoints), or supervised (human approval at each step). Semi-autonomous
Override Points Where in the pipeline human review or override is triggered. Welfare officers review AI-generated call analyses and flagged abnormalities before any intervention action is taken. Emergency alerts are confirmed by human social workers prior to escalation.
Decision Criticality The rights impact of the decision the AI supports. High criticality requires HITL oversight; moderate requires HOTL; low may operate HOOTL. View in glossary Moderate
Human Oversight Type Level of human involvement: Human-in-the-Loop (active review), Human-on-the-Loop (monitoring), or Human-out-of-the-Loop (periodic audit). View in glossary HOTL
Development Process Whether the AI system was developed fully in-house, through a mix of in-house and third-party, or fully by an external provider. View in glossary Fully third-party developed
Highest Risk Category The most significant structural risk source identified: data, model, operational, governance, or market/sovereignty risks. View in glossary Model-related risks
Risk Assessment Status Whether a formal risk assessment, informal assessment, or independent audit has been conducted for this system. Informal assessment

Risk Dimensions

Data-related risks
Market, sovereignty and industry structure risks
Operational and system integration risks

Impact Dimensions

Privacy and data security
Systemic and societal
  • Data minimisation controls
  • Grievance mechanism
  • Human oversight protocol
CategorySensitivityCross-System LinkageAvailabilityKey Constraints
Beneficiary registries and MISPersonalLinks data across multiple systemsCurrently available and usedMunicipal welfare registries of elderly individuals living alone used to identify and enrol service recipients.
Unstructured and text-based contentSensitiveSingle source (no linkage)Currently available and usedVoice conversation data including health status disclosures. Explicit consent required for personal information collection with separate consent for sensitive health-related information. Advanced encryption for voice data storage.

Park, C. (2022) 'In South Korea, AI phone calls check up on lonely seniors in a country ageing faster than Japan', South China Morning Post, 14 February. Available at: https://www.scmp.com/week-asia/people/article/3166142/south-korea-ai-phone-calls-check-lonely-seniors-country-ageing (Accessed: 25 March 2026).

View source News article / media

European Commission (n.d.) 'AI for people: CLOVA Carecall Service by NAVER', Futurium – European AI Alliance. Available at: https://futurium.ec.europa.eu/en/european-ai-alliance/best-practices/ai-people-clova-carecall-service-naver (Accessed: 25 March 2026).

View source Report (multilateral / development partner)

NAVER CLOVA (2023) 'CLOVA CareCall Research Paper: Understanding the Benefits and Challenges of Deploying Conversational AI Leveraging Large Language Models for Public Health Intervention', CLOVA Tech Blog. Available at: https://clova.ai/en/tech-blog/en-clova-carecall-research-paper-understanding-the-benefits-and-challenges-of-deploying-conversational-ai-leveraging-large-language-models-for-public-health-intervention (Accessed: 25 March 2026).

View source Working paper / technical note

NAVER Corporation and Seoul National University AI Policy Initiative (2022) 'Knowledge Interactive, CLOVA CareCall: NAVER-SAPI AI Report 2022 Case Study'. Seoul: NAVER Corporation. Available at: https://www.navercorp.com/static/20230830170049_2.pdf (Accessed: 25 March 2026).

View source Report (government / official)

The Pickool (2025) 'Naver AI CareCall Reduces Solitary Deaths by 44.2%', The Pickool, [online]. Available at: https://www.thepickool.com/naver-ai-carecall-reduces-solitary-deaths-by-44-2/ (Accessed: 25 March 2026).

View source News article / media
Deployment Status How far the system has progressed into real-world operational use, from concept/exploration through to scaled and institutionalised. View in glossary Scaled & Institutionalised
Year Initiated The year the AI system was first initiated or development began. 2021
Scale / Coverage The scale and geographic or population coverage of the deployment. Approximately 150 institutions across South Korea serving an estimated 50,000 elderly individuals living alone. Deployed in municipalities including Seoul, Gyeonggi, Busan, Daegu, Gwangju, North Jeolla, Gangwon, and South Chungcheong.
Technical Partners External technology vendors, academic partners, or development partners involved. NAVER Cloud Corporation (developer); Seoul National University AI Policy Initiative (SAPI); Yonsei University ESG and Business Ethics Research Center (impact evaluation)
Outcomes / Results 44.2% reduction in solitary deaths in service areas compared to non-service areas. 9.2% reduction in emergency room visits. 1.5% increase in general hospital visits attributed to earlier health issue detection. 96% user response rate. Approximately 90% user satisfaction rate. Estimated annual social value of KRW 34 billion (USD 22.7 million). Documented emergency life-saving interventions including Suncheon City cirrhosis case.
Challenges LLM response control challenges — probabilistic nature of language models prevents absolute control over conversational outputs. Domain-specific customisation requires generating entirely new training datasets, a resource-intensive process. Speech recognition difficulties with unclear pronunciation and regional dialects among elderly users. Risk of over-reliance on AI for welfare functions that require human empathy and judgment. Local government requests for modifications (e.g., dementia screening) cannot guarantee consistent question inclusion.

How to Cite

DCI AI Hub (2026). 'CLOVA CareCall – AI-Enabled Welfare Check-In Calls for Elderly Living Alone', AI Hub AI Tracker, case KOR-003. Digital Convergence Initiative. Available at: https://socialprotectionai.org/use-case/KOR-003 [Accessed: 1 April 2026].

Change History

Updated 1 Apr 2026, 08:11
by system (system)
Created 30 Mar 2026, 08:40
by v2-import (import)