In recent years, social protection systems worldwide have been experiencing some interesting trends on two fronts: privatisation and digitisation. Nearly every stage of a social assistance delivery chain, from back-end management information systems to last-mile delivery mechanisms, has witnessed varying levels of renovation along these two lines. Our earlier work has explored the involvement of private players in last-mile delivery of welfare. The previous piece in this series has delineated one particular case study of privatisation in South Africa. However, to explore privatisation in welfare systems without discussing the very motivation for that private involvement (the digitisation mandate) would be incomplete. This blog post offers a more complete narrative.
The permeation of digital technologies into social protection has manifested in myriad ways. While countries like India and Kenya are only now switching to digital transfers of benefits, the First World has been toggling with automated decision-making regarding eligibility for social welfare programs using machine learning and artificial intelligence. The difficult decisions about which citizens are eligible for welfare are no longer made by human caseworkers who can exercise discretion – but by algorithms and technology. In 2014, the United Nations (UN) Special Rapporteur on extreme poverty and human rights presented a report to the UN General Assembly that flagged various concerns about the digitisation of welfare systems, warning that we were ‘stumbling zombie-like into a digital welfare dystopia’[1]. Proponents of digital welfare systems claim the upgraded systems are cost-saving, can eradicate human prejudice, and are generally more efficient for the end-user[2]. However, there have been many instances when these technocratic innovations have created additional inconveniences and costs for the end-beneficiary.
The problems associated with digitisation do not exist in a vacuum. As the UN Special Rapporteur points out, welfare authorities are increasingly relying on private actors – who may have conflicting motives for their involvement in social assistance systems. This post deconstructs three crucial examples of states seeking to implement digital welfare systems and contracting out design and implementation to private entities. Learnings from such instances can inform the construction of stronger social protection systems.
Ontario’s Social Assistance Management System (SAMS) by IBM
Social assistance in Ontario consists of three programs[3] which provide eligible beneficiaries with financial aid, reimbursements for certain expenses, and allowances for specific uses (such as shelter or basic-needs). In 2009, the Ontario Ministry of Community and Social Services identified the earlier information management system[4] as being high-risk, ineffectively designed, and due for modernisation. To that end, they were granted funding of CAD $202.3 million to modify and implement an off-the-shelf information system[5] by 2013[6]. The primary function of the new system, SAMS, was to help caseworkers in managing the case files of welfare recipients. This entailed determining applicant eligibility, calculating and distributing social benefits (worth CAD $6.6 billion annually), and automatically generating letters to update beneficiaries.
SAMS was developed by Curam – a company acquired by IBM in 2011. Curam was responsible for customising the off-the-shelf system specifically for Ontario’s social assistance needs, and IBM was to migrate client data from the previous system onto SAMS[7]. IBM missed vital deadlines for data conversion, causing the launch data of SAMS to be postponed thrice. IBM representatives claimed that their inability to meet deadlines was due to the Ministry defining its requirements later than expected. Errors in the software and in data conversion from the old system also contributed to delays. SAMS was finally launched in November 2014. The delay cost the exchequer CAD $37 million additionally[8]. Upon launch, glitches in SAMS caused overpayments to recipients worth CAD $20 million[9], which went unnoticed for some time. These overpayments later had to be recovered from recipients, who in many instances had already spent the funds. Caseworkers found themselves inefficiently spending more time (30 minutes more than in the earlier system) on new applications[10]. Documents which were automatically generated to be sent to recipients were often erroneous, and the burden of reporting such errors left to them[11]. Such problems were significant inconveniences to beneficiaries. The Auditor General of Ontario found that the Ministry did not properly oversee Curam and IBM consultants, and instead relied on them to design and develop most of SAMS, without much regulatory oversight.
Indiana’s Privatised Public Assistance by IBM and ACS
The system of social protection in Indiana comprises various programmes[12] administered by the Family Social Service Administration (FSSA). In 2006, the state government pursued the modernisation of its welfare system, through a USD $1.3 billion contract with IBM. This exercise was to bring benefits ranging from reduced eligibility errors to time reductions per case for caseworkers[13]. The administration also claimed that the old data system solicited caseworkers to fraudulently manipulate it into overpayment of benefits, and hence needed to be changed.
The new system deliberately shifted away from permitting personal relationships between caseworkers and recipients, and instead introduced a call centre where each worker had a set of tasks to complete. Two-thirds of Indiana’s social service agency’s staff were now employed by IBM and its partners[14]. A phone and internet-based system was introduced, through which recipients would apply for benefits, seek clarifications etc.
Beneficiaries found these new systems difficult to navigate. Even when they were able to, long phone waiting times were an additional hurdle. Applicants were required to re-register themselves for benefits, no matter how long they had already been receiving entitlements, by submitting dated and difficult to locate documents. Applicants who did submit documents (to be processed at a centralised office), would sometimes learn that they had been misplaced. Call centre workers had perverse incentives to reduce the average time per phone call and would often direct applicants to cancel an existing application and restart afresh to minimise the turnover time of an open case. However, the most pertinent issue was of wrongly denied benefits. Many applicants reported receiving notices saying they had ‘failed to cooperate’ with the eligibility determination process – minor errors in paperwork, delays in transporting documents, etc. would be blamed on the applicant. It is notable that between 2006-08 eligibility error rates tripled from 5.9% to 19.4%. The inconveniences faced by beneficiaries at every step along the way severely eroded the public’s confidence in the state welfare system.
Ultimately, the contract was cancelled in 2009, with the Governor of Indiana admitting the experiment a failure. In 2010, the state filed a USD $437 million lawsuit against IBM for breach of the contract, claiming that the automation experiment harmed needy citizens. IBM countersued for about USD $100 million, claiming that the Indiana government continued to use their servers, hardware, and automated processes even after termination of the contract. The court awarded USD $52 million to IBM, compensating them for loss of the contract, noting that neither party deserved to win the case. The judge reprimanded the state for ‘misguided government policy’, and IBM for ‘overzealous corporate ambition’.
North Tyneside’s Risk-Based Verification with TransUnion
North Tyneside, a metropolitan borough in North East England provides its low-income residents with housing benefits and council tax support[15]. Low-income residents are eligible to apply for public assistance to pay their rent and council tax, under certain circumstances. In 2015 the North Tyneside Council sought approval to implement a risk-based verification system to classify applicants for these benefits as being at low, medium or high risk (for fraud). The broad motivation was to reduce the burden of evidence on claimants and eliminate fraud by scrutinising higher risk applicants more closely. TransUnion was contracted in 2015 to automatically process data from new claims.
The council abandoned the contract in 2019. The new system threw up multiple errors – payments to many applicants were wrongly delayed when TransUnion’s predictive analytics identified low-risk claims as high-risk[16]. TransUnion’s system was unable, in most cases, to provide reasoning for why a case was classified as high-risk[17]. It is not difficult to imagine that such erroneous profiling may have effects on the morale and dignity of claimants.
Conclusion
In the cases described here, governments invited private participation in order to digitise and modernise their welfare systems. These projects often focus on cost reduction and improving efficiency – not ignoble goals on their own, but ones that seem to be prioritised at the expense of the beneficiary experience. Governments often learn that private involvement doesn’t work and are forced to abandon projects.
The cases discussed above come from developed countries with strong institutions, largely capable of constructing strong regulatory mechanisms. A natural follow-up is what the Indian context, lacking in the same regulatory capabilities, can take away with regards to these trends.
Digitisation in India is already being seen as an antidote to fragmentation and various other problems in Indian welfare schemes[18]. For instance, the states of Bihar and Jharkhand launched mobile applications to provide emergency cash transfers worth Rs. 1000 to residents of the state stranded elsewhere during the lockdown[19]. The chairman of the Unique Identification Authority of India (UIDAI) has even expressed interest in exploring the capabilities of artificial intelligence and machine learning for fraud detection and improving welfare enrolment processes[20].
In November 2019, the Government of Odisha hosted a workshop on integrated social protection delivery platforms, and soon announced that the department would develop a social registry[21]. This registry refers to a single database, to be created by 2021, to standardise beneficiary lists as they currently differ across schemes. Ultimately the exercise is intended to weed out ineligible beneficiaries and reduce errors of inclusion[22]. Further, the state government also has plans to use automated algorithms to detect cases of fraud when farmers are buying seeds for paddy or selling oilseeds[23]. Even at this incipient stage, there are striking similarities between the cases explored above and Odisha’s plan. It will be up to the Odisha government to implement scheme design in an inclusive manner, respecting the impact of going digital on the beneficiary experience.
Decision-makers have repeatedly emphasised that the use of technology can provide cost savings by limiting inclusion errors and eliminating undeserving beneficiaries. We have seen this in the past when the main motives for Aadhar adoption were, de-duplication of records, efficiencies and savings for the exchequer[24]. Post-implementation as well, a key part of the UIDAI narrative has been cost savings[25].
However, the examples set out here highlight that eliminating inclusion errors can come at the cost of excluding genuine and deserving citizens. The cases above are from jurisdictions with significantly greater regulatory capacity and digital literacy than India[26]. Even so, beneficiaries faced significant barriers in navigating the new systems in North Tyneside and Indiana. It is not unfeasible that the same concern may arise on an even greater scale in India. We emphasise that many welfare transfers are geared towards the most vulnerable citizens (e.g., NSAP for BPL households), and hence, even a single case of exclusion can have very grave consequences. Digitisation should not come at the cost of increased exclusion errors.
The purpose of this post is not to undermine the ability of privatisation or digitisation to make a genuine contribution towards effective social protection systems – but speaks to the importance of designing such efforts with the end-users needs in mind. Digital welfare systems can work – when they are preceded by a pivot of principles to reducing exclusion.
[1] Aston, P. (2019). Report of the Special rapporteur on extreme poverty and human rights. UN Human Rights Office of the High Commissioner. Retrieved from https://srpovertyorg.files.wordpress.com/2019/10/a_74_48037_advanceuneditedversion-1.pdf
[2] World Bank. (2018). PRINCIPLES ON IDENTIFICATION FOR SUSTAINABLE DEVELOPMENT: TOWARD THE DIGITAL AGE. World Bank. Retrieved from http://documents1.worldbank.org/curated/en/213581486378184357/pdf/Principles-on-identification-for-sustainable-development-toward-the-digital-age.pdf
[3] 1. Ontario Works 2. Ontario Disability Support Program (ODSP), and 3. Assistance for Children with Severe Disabilities (ACSD)
[4] Known as the Service Delivery Model Technology, which was in place between 2002 and 2014.
[5] An off-the-shelf information system is ready to use immediately upon purchase and is created to be widely used across sectors.
[6] Office of the Auditor General of Ontario. (2015). 2015 Annual Report of the Office of the Auditor General of Ontario (pp. Chapter 3, Section 3.12). Retrieved from https://greydocs.ca/urm/groups/public/documents/communication/gc_260387.pdf
[7] The same software was adopted in Maryland, Minnesota and the District of Columbia in the United States and caused serious technical problems in health insurance exchanges in these jurisdictions.
[8] SAMS: More Than A “Glitch”. (2015). Retrieved 9 September 2020, from https://opseu.org/news/sams-more-than-a-glitch/12985/
[9] SAMS: More Than A “Glitch”. (2015). Retrieved 9 September 2020, from https://opseu.org/news/sams-more-than-a-glitch/12985/
[10] General Manager. (2015). Impact of Social Assistance Management System (SAMS) implementation in Toronto. Toronto Employment and Social Services. Retrieved from https://www.toronto.ca/legdocs/mmis/2016/gm/bgrd/backgroundfile-86902.pdf
[11] Office of the Auditor General of Ontario. (2015). 2015 Annual Report of the Office of the Auditor General of Ontario (pp. Chapter 3, Section 3.12). Retrieved from https://greydocs.ca/urm/groups/public/documents/communication/gc_260387.pdf
[12] Supplementary Nutrition Assistance Program (SNAP – food stamps), Medicaid (health insurance), and Temporary Assistance for Needy Families (TANF – a cash transfer program), among others.
[13] Eubanks, V. (2018). Automating Inequality: How High-Tech Tools Profile, Police, and Punish the Poor (1st ed.). New York: St. Martin’s Press.
[14] Rohde, D., & Cooke, K. (2012). The Unequal State of America: Indiana’s rocky road to welfare reform. Reuters. Retrieved from https://www.reuters.com/article/us-equality-indiana-welfarereform-idUSBRE8BJ0J620121220.
[15] Housing Benefit and Council Tax support | North Tyneside Council. Retrieved 10 September 2020, from https://my.northtyneside.gov.uk/category/93/housing-benefit-and-council-tax-support#:~:text=Apply%20online%20for%20Housing%20Benefit,number%20from%20your%20claim%20form.
[16] Marsh, S. (2019). One in three councils using algorithms to make welfare decisions. The Guardian. Retrieved from https://www.theguardian.com/society/2019/oct/15/councils-using-algorithms-make-welfare-decisions-benefits
[17] North Tyneside Council. (2019). Risk Based Verification. Retrieved from https://my.northtyneside.gov.uk/sites/default/files/meeting/related-documents/5c%20Risk%20Based%20Verification%20Report.pdf
[18] Sharma, A. (2019). The digital way: Growth with welfare. Financial Express. Retrieved from https://www.financialexpress.com/opinion/the-digital-way-growth-with-welfare/1724789/
[19] Jha, S. (2020). Bihar, Jharkhand transfer via apps Rs 1,000 each to 21 lakh migrant workers. Financial Express. Retrieved from https://www.financialexpress.com/india-news/bihar-jharkhand-transfer-via-apps-rs-1000-each-to-21-lakh-migrant-workers/1970284/
[20] Economic Times. (2018). ‘Aadhaar-enabled DBT savings estimated over Rs 90,000 crore’. Retrieved from https://economictimes.indiatimes.com/news/economy/finance/aadhaar-enabled-dbt-savings-estimated-over-rs-90000-crore/articleshow/64949101.cms?from=mdr
[21] Odisha Plus. (2019). Social registry in mind; seventeen schemes now on DBT. Retrieved from https://www.odisha.plus/2019/07/social-registry-in-mind-seventeen-schemes-now-on-dbt/
[22] The Hindu. (2019). Odisha to come up with social registry for all welfare programmes. Retrieved from https://www.thehindu.com/news/national/odisha-to-come-up-with-social-registry-for-all-welfare-programmes/article29872337.ece
[23] Mohanty, D. (2020). Odisha plans social registry to weed out ghost beneficiaries using Aadhaar. Hindustan Times. Retrieved from https://www.hindustantimes.com/india-news/odisha-plans-social-registry-to-weed-out-ghost-beneficiaries-using-aadhaar/story-d8XNv9Jv8zVpp4tDbHyohP.html
[24] UIDAI. (2015). Basic Knowledge of UIDAI and Aadhaar. Retrieved from https://uidai.gov.in/images/training/module_1_basic_knowledge_of_uidai_and_aadhaar_16032015.pdf
[25] Economic Times. (2018). ‘Aadhaar-enabled DBT savings estimated over Rs 90,000 crore’. Retrieved from https://economictimes.indiatimes.com/news/economy/finance/aadhaar-enabled-dbt-savings-estimated-over-rs-90000-crore/articleshow/64949101.cms?from=mdr
[26] BBC. (2019). India’s on a digital sprint that is leaving millions behind. Retrieved from https://www.bbc.com/news/world-asia-india-49085846
