Innovations in ICT are rapidly extending the scope and impact of social security policies while powering simplified and quality service delivery. In this context, a key trend has been the growing role of data in social security administration (ISSA 2016 and 2019; Ruggia-Frick, 2021). Social security institutions are increasingly exploring new ways to harness the large volumes of data they manage to streamline processes, deliver customized services, reduce fraud and error, and formulate evidence-based policy decisions.

To support member institutions in the effective and secure implementation of data analytics, the ISSA has developed specific guidance as part of the broader ISSA Guidelines on Information and Communication Tecnology (ISSA, 2019b). The guidelines on data analytics (54–59 of the guidelines on ICT) are tailored around the following main categories:

• Descriptive analytics, which involves analysis of historical data to inform future decisions
• Diagnostic analytics, which look towards the processes and causes of an event
• Predictive analytics, which tries to predict future outcomes using forecasting techniques
• Prescriptive analytics, which helps make decisions by analyzing various alternatives
• Analytics of big data, which applies the above techniques to very large datasets
• Machine learning and big data, which goes beyond traditional analytical techniques and examines the nuances of applying machine learning techniques to big data

The type of analytics social security institutions can apply depends on their data maturity. While numerous analytics maturity models have emerged over the last decade, a recent maturity model for data-driven government that was presented at the ISSA ICT Conference 2018 could potentially help social security institutions identify their current capabilities and define their future roadmap.

Table 1. Maturity model for data-driven transformation

	Ad Hoc	Prepared	Demonstrated	Proven	Intelligent
Strategy	Agency strategy does not consider data-driven policy and practice	Strategy for digital transformation with a vision for data-driven policy and practice	Strategy for data-driven policy and practice documented and endorsed	Agency strategy is formulated based on data-driven insights	Government strategy is formulated based on data-driven insights
Data	Data is unavailable or is unsuited to supporting data‑driven transformation	Data sources have been identified and data is being prepared	Quality data is available in some domains	Data is managed as a strategic asset across the enterprise	Data is continually enriched by data-driven processes
Technology	Siloed systems impede longitudinal analysis	Longitudinal analysis is supported by traditional analytics	Point solutions support predictive analytics in some domains	Enterprise platform supports predictive analytics across domains	Enterprise platform includes autonomic refinement of predictive models
Culture	Cultural resistance to change	Executive support for data‑driven policy and practice	Some teams have adopted data‑driven practices	Enterprise‑wide adoption of data‑driven practices	Data‑driven culture extends to government policymakers
Influence	Decisions are based on well‑intentioned gut‑instinct	Decisions are based on unreliable and incomplete data	Some examples of evidence-based decision-making	Business practices are transformed based on data‑driven insights	Government policy is influenced by data‑driven insights
Impact	Government policy and business practices are unchanged	Policy and practice reviews being prompted by data-driven insights	Some examples of dynamic responses to major events and emerging trends	Dynamic responses are frequently informed by data‑driven insights	Policy is continuously reviewed and improved to optimise outcomes
Source: van Leent (2018)

Experiences of applying data analytics from Asia and the Pacific

Social security institutions in Asia and the Pacific have leveraged data across a range of areas such as improving service delivery via automation, health care, detecting and preventing error, evasion and fraud and proactive social policy and programme design, among others. Examples of good practices from the region have been submitted to the ISSA Good Practice Awards competition and presented during the Virtual Social Security Forum for Asia and the Pacific and in webinars and other ISSA activities. This article examines member institutions’ experiences across emerging areas of impact.

Improving service delivery via data-driven automation

Services Australia

Services Australia, the agency responsible for delivering social services and means-tested social security payments in Australia, has leveraged data analytics to reliably assess claims through the use of Straight Through Processing (STP). STP is the concept that a request can be handled without any human intervention. As COVID-19 destabilized livelihoods, the agency extended its existing automated decision-making capabilities to expedite payments to the unprecedented volume of claimants under the JobSeeker programme (Services Australia, 2020). The aim was to provide payments to people in need as fast as possible and to assure the government that automated payments, while socially responsible and administratively efficient, were made in line with four “golden rules”: the right person, on the right programme, with the right rate, commencing on the right date.

While Services Australia had implemented STP for other categories of payments in the past, the crucial difference was the scale and the speed at which it had to be implemented in the case of JobSeeker claims. Therefore, it was important to measure and demonstrate the administrative efficiency of automating the claims approval process while providing assurance on the integrity of the payment outcome. While the agency set up a sound business framework to guide the development of automation products, the data-driven assurance process which conducted checks on a statistically valid sample of automated payment decisions was key to demonstrate reliability and accuracy without human intervention. The agency set a payment accuracy benchmark of >95 per cent and used a range of statistical methods, automated registers and data-driven analysis to measure its achievements against this target. The agency achieved an accuracy target of 99 per cent, well over its benchmark.

Fiji National Provident Fund

The Fiji National Provident Fund (FNPF) leveraged data to improve the efficiency of a range of processes following the COVID-19 pandemic. The FNPF rolled out four phases of income protection assistance to support its members cope with the impacts of the crisis. The FNPF introduced, for the first time, Robotic Process Automation (RPA) to automate the process of registration, processing, approval and payment. FNPF was able to use data insights to identify members that continued to be affected, allowing auto-registration of members using RPA (Fiji National Provident Fund, 2020). The FNPF was able to mine data to estimate its maximum exposure and pre-emptively employ liquidity strategies to manage the cash flow position. Using data analytics and insight, FNPF proactively reviewed rejections and reversed decisions where applicable without members having to file formal complaints.

Controlling fraud and error in health care

Social Security Administering Body for Health, Indonesia

With nearly 80 million hospital uses in 2020, the Social Security Administering Body for the Health Sector (BPJS Kesehatan) in Indonesia needed a credible and robust fraud detection tool. The existing retrospective methods of fraud detection were less efficient, as the post recovery of excess payments required considerable effort and resources. Furthermore, the manual analytical protocols required considerable data processing capabilities, making it cumbersome and expensive.

BPJS Kesehatan has addressed these challenges through a machine learning system for fraud detection. The system’s central advantage is that it’s concurrent, i.e., hospital claims are reviewed before the payments are made. This in turn has several advantages. First, fraud detection is faster and more efficient as the system is able to identify suspicious patterns quickly. Second, staff time is freed up for more strategic tasks around fraud control rather than manual reviews. Finally, with growing historic data, prediction models improve in their sophistication and accuracy, unlike manual methods that simply cannot process data at this scale (Social Security Administering Body for Health, 2019 and 2022).

The machine learning model was developed based on historical transactional behavior using supervised learning techniques. The algorithm is periodically retrained with new data. The machine learning model has been iteratively implemented, starting with 10 hospitals in 2019, extending to 265 hospitals in 2020 and scaling up to 2,511 hospitals in 2021. In 2021, the artificial intelligence (AI) engine screened 5.8 million transactional claims from hospitals to flag 390 000 transactions for additional review. This in turn helped over 700 verification officers of the BPJS Kesehatan to work more effectively.

National Health Insurance Service, Republic of Korea

The National Health Insurance Service (NHIS) in the Republic of Korea has made considerable advancements in using big data for error and fraud detection among health insurance claims. The NHIS houses big data on a range of socio-economic, health behaviour, health-care utilization and long-term care variables. The NHIS applies smart audit algorithms over this data to predict health-care facilities with high probability of fraudulent claims, thereby pre-emptively supporting investigators (National Health Insurance Service, 2022).

The fraud detection model is hybrid and it combines traditional analytics and artificial intelligence. In a first step, all applications received are examined using traditional rule-based algorithms to check whether correct and unusual claims are distinguished. Artificial intelligence is then used to identify correlations between unusual claims, which helps determine the likelihood of an inspection. The artificial intelligence model, first prototyped in 2021, evolved from a previous generation of traditional rule-based models. The model is defined using a range of techniques, including deep learning, random forest, gradient boosting, logistic regression and Support Vector Machine. The model is currently being refined further. Between 2014 and 2021, the fraud detection system detected 567 820 cases of fraud, which is equivalent to 174 million United States dollar (USD) in claim value.

Delivering customized services

Korea Workers’ Compensation & Welfare Service

The Korea Workers’ Compensation & Welfare Service (COMWEL) has been implementing customized rehabilitation plans for injured workers since 2011, helping the successful rehabilitation of workers. While the services have been expanded over recent years, and efforts have been made to design individual plans through internal rehabilitation experts, the process has relied on limited information and the experience of managers in charge, resulting in variable service quality and timeliness.

To enhance its support to injured workers, COMWEL has developed the Intelligent Rehabilitation Recommendation System (IRRS) (Korea Workers’ Compensation & Welfare Service, 2020). The IRRS is an AI-based system developed in 2020 to select the injured workers who with the potential to be active, and design scientifically tailored rehabilitation services for them. The IRRS calculates a vulnerability index based on administrative data on 98 million workers accumulated since 2011, comprising details about workers’ compensation, unemployment insurance, the rehabilitation case management, and it uses rule-based filtering and case-based reasoning methodology. Furthermore, it also suggests a rehabilitation plan based on the AI model. The workers selected for rehabilitation and return to work undergo consultation with the rehabilitation experts of COMWEL before AI-generated plans are finalized. The system was first implemented in early 2020. Although it was difficult to provide rehabilitation services due to COVID-19 in 2020, 13,876 services were recommended to 2,637 injured workers by the IRRS, of which 9,172 services (66 per cent) were actually designed and provided as customized services. The IRRS has helped COMWEL achieve consistent service quality nationally while ensuring timely and appropriate interventions to ultimately improve the return-to-work ratio.  

Results

Table 2 summarizes the results these institutions have been able to achieve using data analytics.

Table 2. Results summary

Institution	Results obtained
Services Australia, Australia	Successfully automated >31,000 claims for means tested social security payments that were assessed, evaluated and paid almost in real time from the time they were submitted, without any staff intervention. Saved staff time, which in turn was redirected to support vulnerable customers and more complex cases.
FNPF, Fiji	Savings of 1.8 million Fiji dollars (FJD) was achieved through automation and efficiencies. Processed and paid 80 per cent of the applications within the committed five working days’ turnaround time. Allowed staff to focus on managing exception cases and resolving queries.
BPJS-K, Indonesia	29,990 potential fraudulent claims detected with a total amount of saving of USD 41.9 million.
NHIS, Republic of Korea	Between 2014 and 2021, the fraud detection system detected 567,820 cases of fraud which is equivalent to a cumulative saving of USD 174.4 million.
COMWEL, Republic of Korea	Time required from the decision date of compensability to the first rehabilitation consultation was reduced by 4.8 days in 2020 compared to 2019, enabling prompt counselling and rehabilitation services.

Critical success factors

The experience of member institutions reveals a number of key factors that are essential to leveraging data analytics successfully for social security administration.

Data-driven innovations can deliver value only when underlying processes are efficiently designed. For instance, if a process is flawed, then automating it via data analytics would only replicate existing errors, albeit in a different form. Therefore, assessing and optimizing processes is a first step before considering data-driven automation, as demonstrated by Services Australia in the case of STP.

Data quality and data governance are central to realizing the potential of data analytics effectively. The quality and governance of data is key to the success of data-driven functions. The ISSA Guidelines on ICT, guidelines 22–67 and guidelines 63–65 in particular, provide guidance on issues of master data governance and data quality management. For instance, the use of data by the BPJS Kesehatan in Indonesia is anchored in a sound data management framework following the recommendations in the DAMA-DMBOK2 (Dama International, 2017). Consistent quality management throughout the data lifecycle is among the core principles of BPJS Kesehatan. The organization has also pursued data standardization, electronic data capture and validation checks to enhance data quality.

Stakeholder engagement across all levels and business areas from the beginning is essential for change management and successful real-time troubleshooting. For instance, Services Australia found that early involvement of policymakers, legal professionals, business owners, and quality assurance teams was key to avoid late-stage overhaul of solutions. 

Exploiting data, like any other ICT investment, requires social security institutions to attract, train, build and retain talented staff. As analytical technologies are rapidly evolving, attracting and retaining the right data professionals can be challenging. Further, data analytics requires people to work within multidisciplinary teams. As the NHIS in the Republic of Korea found, field professionals and ICT professionals can struggle to understand each other. The importance of anchoring data-driven innovation within multidisciplinary teams is also reiterated by global evidence on this topic  (ISSA, 2020; Ruggia-Frick, 2021). Therefore, continuous internal capacity building is essential to capitalize on data analytics by combining domain, methodological and technology expertise.

Continuous learning and adaptation are key particularly in terms of big data analytics. As data volumes grow, machine learning models need to be continuously trained and evaluated for robust performance. Big data analytics are not a one-off exercise, and require resources to be committed continuously, as the BPJS-Kesehatan and the NHIS found in Indonesia and the Republic of Korea respectively. This also implies putting in place a strong performance framework to evaluate model performance using a consistent set of metrics. For instance, the BPJS-Kesehatan has four metrics to periodically evaluate its machine learning model against pre-specified benchmarks: (i) accuracy of prediction; (ii) accurarcy of correct predictions of positive observations (suspected potential fraud) to the total predicted positive observations; (iii) recall, i.e., the ratio of correctly predicted positive observations to all observations in an actual class recall; and (iv) F1 score, which is the weighted average of precision and recall. Periodic evaluation can be useful even in the case of traditional data analytics. For instance, Services Australia carefully tests the performance of STP using metrics and manual checks to avoid financial losses due to incorrect computation of payments.

When applying automated decision-making via big data technologies for eligibility assessment or fraud detection, social security institutions must carefully consider its legal and ethical implications. More generally, the limits of automation for such critical services and the explainability of the algorithms used have to be correctly assessed (ISSA, 2020; Ruggia-Frick, 2021). Indeed, as the United Nations (UN) Special Rapporteur for Extreme Poverty has noted, many automated decision-making systems have been implemented without a firm legal basis and without adequate recourse to claimants to appeal digitally-enabled decisions, which in turn affects their rights in terms of transparency and fairness (UN Special Rapporteur for Extreme Poverty, 2019).

Finally, all practices showcased in this article emphasize the importance of adequate safeguards for data security and privacy. While the enormous amounts of personal and transactional information can help social security institutions unlock value, data breaches can significantly harm individuals and erode trust in social security institutions (Wagner and Ferro, 2020). Guidelines 36–46 of the ISSA Guidelines on ICT provide detailed guidance on establishing a sound data protection and privacy framework that is backed by robust technologies.

Conclusion

By integrating data analytics into business processes, social security institutions can reap tangible results and business outcomes such as improved operational performance, enhanced customer service, reduction in fraud and error, and evidence-based decision-making. Data analytics can enable institutions to seize new opportunities by making significant improvements to existing, or to develop new, products, processes and organizational methods. Consequently, data will be a crucial ingredient in social security institutions’ continuous search for innovative strategies to provide cost-effective quality services over the future decades.

While previous decades have seen institutions explore traditional analytics, social security institutions are progressively applying emerging technologies, such as big data and artificial intelligence. Although the application of these technologies remains nascent, experiences of member institutions demonstrate how they already enable relevant outcomes in key social security areas such as addressing error, evasion and fraud as well as developing proactive approaches and automated solutions to improve social services. As institutions’ data analytics maturity evolves, as it is bound to do with technological advancements, the use of these emerging technologies is going to accelerate further. In turn, institutions have to prepare themselves to embrace these developments by creating adequate data governance and management capabilities to address the risks mentioned above and ensure transparency, fairness and accountability in the application of advanced data-driven approaches.

References