Summary

Governments worldwide are turning to automation to help them deliver essential public services, such as food, housing, and cash assistance. But some forms of automation are excluding people from services and singling them out for investigation based on errors, discriminatory criteria, or stereotypes about poverty. Despite these harms, the allure of tech-based solutions to complex social problems is proving hard to resist.

The World Bank is one of the biggest development actors driving this trend, particularly in the Global South, placing big bets on data-intensive technologies to help governments deliver services. This includes major cash transfer programs that give certain individuals or families financial support. In the Middle East and North Africa alone, eight of ten borrowing countries have received Bank loans to upgrade these programs.

Projects to Upgrade Poverty Targeting in the MENA Region Financed by the World Bank

*Iran and Yemen have active World Bank projects but no known plan to develop or upgrade poverty targeting technology

The Bank has long promoted cash transfer programs that select beneficiaries by trying to estimate their income and welfare. This approach, known as poverty targeting, has attracted intense criticism for undermining people’s social security rights, particularly in the wake of the economic crisis triggered by the Covid-19 pandemic. Poverty targeted programs are prone to error, mismanagement, and corruption, and routinely fail to reach many of the people they aim to cover. While the Bank has acknowledged these problems, it is financing a range of technologies it claims will make poverty targeting more accurate, reliable, and efficient.

This report documents the human rights impact of one such Bank-financed program in Jordan, known as the Unified Cash Transfer Program, but commonly referred to by its original name, Takaful. After screening out families that do not meet basic eligibility criteria, Takaful uses an algorithm to identify which of those remaining should receive cash transfers by ranking their level of economic vulnerability. Drawing on interviews with applicants and beneficiaries, government officials, community activists, and an analysis of World Bank documents, Human Rights Watch found that this algorithm is leading to cash transfer decisions that deprive people of their rights to social security. The problem is not merely that the algorithm relies on inaccurate and unreliable data about people’s finances. Its formula also flattens the economic complexity of people’s lives into a crude ranking that pits one household against another, fueling social tension and perceptions of unfairness.

The harms of Takaful’s targeting algorithm highlight the need for cash transfer programs based on the principles of universal social protection – providing all people with support during periods of their lives when their economic and social rights are particularly at risk, such as when they are children, face disability, become unemployed, or reach old age, and regardless of their income or wealth. On paper, the Bank has endorsed this approach as the “cornerstone of inclusive social policy.” But it continues to fund poverty targeted programs in dozens of countries, despite the wealth of policy options available to finance their transition to universal schemes.

This gap between rhetoric and practice has devastating effects on people’s rights to social security and related rights to food, health, housing, and an adequate standard of living. Although Takaful has extended regular cash assistance to 120,000 households in 2022, this is only a small fraction of the households in Jordan living under the official poverty line, which itself is an inadequate measure of the number of people unable to realize their economic, social, and cultural rights in the face of high inflation and unemployment.

The Failures of Takaful

Takaful’s complex process for evaluating who receives cash transfers begins with a questionnaire that applicants must complete. Applicants enter their name and national ID number, as well as income-related information such as wages, living expenses, and electricity and water meter ID numbers. This information is combined with data from 37 government agencies, such as vehicle registration and business licenses, employment histories, and enrollment in other social protection programs, to create comprehensive profiles of people seeking benefits.

The National Aid Fund (NAF), the Jordanian government’s social protection agency charged with implementing Takaful, told Human Rights Watch that it vets applicants in two stages. First, it assesses whether applicant households have met basic eligibility criteria, such as whether they are headed by a Jordanian citizen and living under the official poverty line. Given the program’s limited budget, however, not all eligible households become beneficiaries. Instead, NAF applies an algorithm that uses 57 socio-economic indicators to estimate their income and wealth, ranking them from least poor to poorest. The agency then selects households deemed most vulnerable to receive a limited number of cash transfers, providing them between 40 to 136 Jordanian dinars (US$56 to $192)[1] per month.

This veneer of statistical objectivity masks a more complicated reality: the economic pressures that people endure and the ways they struggle to get by are frequently invisible to the algorithm. For example, applicant households are required to declare their income and living expenses, such as food, rent, and education. The algorithm uses the self-reported income to calculate their ranking if it is higher than the estimate based on the 57 indicators. Several people told Human Rights Watch that they could submit an application only if they declared an income that matches their living expenses. NAF disputed this, saying that people can list expenses that exceed their income by about 20 percent.

Either restriction fails to recognize how people struggle to make ends meet, or their reliance on credit, support from family, and other ad hoc measures to bridge the gap.  

“The questions asked don’t reflect the reality we exist in,” said Abdelhameed, a father of two who makes 250 dinars a month ($353) working at a coffee shop in Amman. “My rent is 140 dinars [$197]. My phone, electricity and water bills, and other living expenses are 320 dinars [$451]. My income is less than my expenses, but I had to match it on the application. Each month, we just trust in Allah and hope we can cover our expenses. Some months, my siblings abroad help out.”

Like many others whom Human Rights Watch interviewed, Abdelhameed eventually learned that he was able to submit the application if he lowered his expenses to match his income. Forcing people to mold their hardships to fit the algorithm’s calculus of need, however, undermines Takaful’s targeting accuracy, and claims by the government and the World Bank that this is the most effective way to maximize limited resources.

This is one of several dubious criteria that Takaful’s algorithm uses to distinguish between people who are selected for transfers and those who are notified that they are “not among the poorest.” The algorithm may infer that households are less vulnerable the more water and electricity they consume, under an indicator that measures dwelling characteristics.  The value of an old family car, a small business, or livestock also affects the family’s ranking.    

These measures of vulnerability trap people in impossible choices between the realization of their right to social security and other economic and social rights, such as their rights to a decent living, health, and food. Some people told Human Rights Watch that owning a car could have been one of the reasons they were rejected from Takaful, even though they needed it for work, or to transport water and firewood. “The car destroyed us,” said Mariam, a resident of al-Burbaita village in the southern governorate of Tafilah, one of the poorest villages in the country. Her family received Takaful cash transfers in 2021 but was dropped from the program in 2022. “We use it to transport water and for other needs. But sometimes we don’t have the money to fill it up with diesel, so we walk to the street and wait for someone to pass by and agree to pick us up,” she added.  

Human Rights Watch also found that the algorithm reinforces gender-based discrimination. Its calculation of household size, one of the measures of vulnerability, only considers the number of Jordanian members in the household. This formula artificially shrinks the size of households headed by Jordanian women with noncitizen spouses and children because the law does not recognize their right to pass on citizenship to these family members on an equal basis with men, lowering their benefit payments or excluding them from the program entirely. Women in male-headed households are not spared from gendered design choices either: awarding Takaful payments to heads of households, who in Jordan are usually regarded as the husband or father, rather than individual adult members, heightens dependency on male family figures and unduly restricts women’s access to benefits.

Takaful also offers lessons on the importance of reconciling the ambition of migrating government services online with the realities of the digital divide. The online application is daunting for people who lack internet access and digital literacy. “Who needs money?” said Aseel, a 29-year-old worker at a plastics factory in Amman who has benefited from Takaful and helped many other families apply. “The people who really don’t know how [to apply] or don’t have internet or computer access.”

NAF has mobilized its offices and set up temporary registration centers around the country to help people apply for the benefit. But people who are unable to access these resources, or who do not know someone familiar with the application process like Aseel, often end up paying a fee to submit their application through mobile phone shops or maktabehs – stores that sell office supplies, children’s toys and printing and internet services, and have now become a regular port of call for people struggling to navigate the requirements of the poverty-targeted scheme. These businesses also help people cash out benefit payments from their e-wallets – the main payment method that both the government and the World Bank have promoted as an effort to increase financial inclusion.

These online transactions impose hidden costs that cut into an already meagre benefit. Misha’al, 36, tried to apply for Takaful on his own, but the online portal was loading too slowly on his phone. He paid 2 dinars ($2.82) to travel by bus to a maktabeh about 35 kilometers away from his home to submit his application. The maktabeh charged another 3 dinars ($4.23) for the service. When his application was approved, Misha’al paid a mobile phone shop another 3 dinars to withdraw his benefit, on top of the administrative fee of half a dinar ($.70) levied by the e-wallet company. These fees reduced his family’s monthly benefit from 101 ($142) to 92.5 dinars ($130).

In response to these findings, the Bank said that Takaful is “amongst the most redistributive and cost-effective poverty reduction programs currently active in Jordan,” and part of a broader set of programs dedicated to advancing universal social protection in the country. It emphasized that information systems and technology can “facilitate” the delivery of social protection programs, but “are not substitutes” for interactions between institutions and people.

NAF said that the algorithm’s 57 indicators are designed to measure “multi-dimensional poverty,” and that none of them would, on their own, exclude a household from Takaful. Instead, each of these indicators is assigned a certain weight stipulating their relative importance in the targeting process. For example, households with cars that are more than five years old would be less likely to qualify for cash transfers than households that do not own cars, all else being equal. However, the agency acknowledged that owning a car less than five years old or a business worth 3,000 dinars or more ($4,231) would automatically exclude families from the program.  

NAF also confirmed that it requires applicants to declare living expenses that are no more than about 2o percent of their income, because many of them had previously declared they had no income but high expenses. It also emphasized that people have multiple ways to apply for Takaful: online, at one of NAF’s 68 branches or 290 registration centers, or through a mobile registration vehicle that makes stops in rural areas.

NAF declined to disclose the full list of indicators and the specific weights assigned to each indicator, saying that these were for internal purposes only and “constantly changing.”

Key Recommendations

Establishing a universal social protection system in Jordan will address many of the problems identified in this report and provide lasting and meaningful protections for people’s rights, including their rights to social security and privacy. Universal programs, such as benefits for all children or for people that reach retirement age, remove the need for subjective and error-prone assessments of people’s income and welfare, and the intrusive data collection practices required for these assessments. They also simplify eligibility requirements, easing the burden of applying while improving transparency and accountability. The social protection consultancy Development Pathways has estimated that introducing a modest package of universal child, disability and old age benefits in Jordan will cost 280 million dinars ($394.9 million), or less than 1 percent of GDP, during the first year of the program, and increase to slightly above 2 percent by 2035.

A shift in the government’s approach could help many individuals and families break the cycle of precarity and poverty. “If the government is in front of me right now, it would be useful for them to feel the life pressures we feel living in Jordan,” Aseel said. “Everyone has to juggle several jobs. I don’t have time to see my family anymore. I don’t have time to tell my mom how I’m doing because I am too busy. I don’t see my friends anymore because I’m too busy working… people are just so desperate.”   

 

Recommendations

To the World Bank and other development banks:

• Cease developing, pursuing, supporting, or recommending the use of algorithmic decision-making to estimate, rank, or score people’s financial welfare and allocate cash transfers based on these evaluations.  
• Conduct and publish algorithmic audits on all Bank-financed programs that use algorithmic decision-making.
• Ensure that all audits, including the 2023 evaluation of the Takaful targeting algorithm, declare the algorithm’s value weighting and decision-making criteria.
• Ensure that audits examine the rate of exclusion errors, the reasons for such errors, and corrective measures taken; the adequacy of privacy and data protection safeguards to mitigate the risk of data breaches and leakages and inaccuracies in personal data held by the relevant government agencies; and the right to appeal eligibility decisions.
• Focus financing and technical assistance on modernizing information systems that are critical to securing universal social protection, such as population registries and vital statistics databases, and ensure that such assistance is provided in a manner consistent with privacy and related human rights standards.
• For existing social protection loans with poverty targeting objectives, work with borrower governments to revise objectives and budget allocations to prioritize establishing policies and infrastructure that are in line with the Bank’s own vision to achieve universal social protection by 2030.
• Build on the Bank’s 2022 Social Protection and Jobs sector strategy to develop concrete commitments to universal social protection, expanding financing and technical assistance for social protection schemes that cover various life course contingencies, such as universal child benefits, maternity benefits, and pension schemes for all people aged 65 and above.
• In countries with limited fiscal space, advocate for categorical schemes that may be initially restricted in scope (for, e.g., a universal child benefit that targets all children under a certain age, provided a strategy is put in place to progressively expand the scheme to target all children below the age of 18).[2]
• Ensure that social protection programs are designed to meet individual rather than household needs and provide benefits to the individual rather than the head of household.
• Identify and take concrete steps to address the impact of the digital divide on access to social protection.   

To the Government of Jordan:

• Cease using the Proxy Means Testing-based targeting algorithm and methodology to allocate cash transfers and reconfigure Takaful’s infrastructure, including the National Unified Registry, to support the rollout of universal social protection schemes that provide support to individuals rather than households.
• Shift social protection strategy from poverty targeting to schemes that cover all people when they are most at risk in life, such as a child benefit that provides    Pprotection during and after childbirth and an old age benefit that protects people during retirement age.[3] 
• Adopt categorical schemes that may be initially restrictive in light of fiscal constraints, provided there is a concrete strategy and timeline in place to expand these programs to become truly universal (e.g., providing a universal child benefit for all children under a certain age, with the goal of expanding the benefit to cover all children under 18, as well as the benefit amount to an adequate level).[4]
• During the transition period between poverty-targeted and universal social protection schemes, adopt interim measures to expand and improve Takaful, such as by expanding coverage to families of people with temporary Jordanian passports, non-citizen spouses, and children of Jordanian women.  
• Publish all 57 indicators that the targeting algorithm uses to estimate household income, as well as the weights assigned to each indicator.
• Publish the 2023 evaluation of the targeting algorithm as well as yearly audits of the targeting algorithm for as long as it is operational, ensuring that such audits disclose any changes in the value weighting and decision-making criteria, the rate of exclusion errors, the reasons for such errors, and corrective measures taken.
• Make public all previous findings regarding inclusion and exclusion error rates associated with Takaful’s targeting method, and steps taken to address these errors.
• Publish data that will enable meaningful public scrutiny of Takaful’s effectiveness at reaching people living in poverty, including the 2017-2018 Household Expenditure and Income Survey datasets.
• Increase public awareness of the appeals process and the right to appeal through media and public campaigns as well as in-person outreach.
• Expand support for the appeals process, increasing relevant staffing to process appeals and provide in-person support throughout the process.
• Establish a data protection law that meets international data protection law standards, including an independent data protection authority and mechanisms that support the benefits appeals process, such as those enabling social protection beneficiaries to correct inaccuracies in their personal data held by government agencies.
• Reform existing social dialogue mechanisms, such as the Economic and Social Council and the Tripartite Committee or establish new ones to enable regular and meaningful dialogue between independent civil society organizations, trade unions, and government agencies on social protection reforms, integrating their input into the implementation of social assistance schemes as well as broader strategy shifts.[5]  
• Increase the number of registration centers administering social assistance programs across the country, particularly in rural areas.
• Provide subsidies for costs associated with applying online and withdrawing the benefit payment, including e-wallet withdrawal fees and travel costs.
• Dispatch mobile cash units regularly, particularly to areas far from city centers, to help people withdraw their benefit payments.

To other international organizations and donors supporting Takaful:

• Support the Jordanian government in winding down the Proxy Means Testing-based targeting algorithm and approach for allocating cash transfers.
• Support the reconfiguration of Takaful’s infrastructure, including the National Unified Registry, to enable the rollout of universal social protection schemes that provide support to individuals rather than households.
• During the transition period between poverty targeted and universal schemes, publish yearly assessments of the targeting algorithm as outlined above, and analyses of previous and current iterations of the algorithm.
• Conduct regular and meaningful consultations with independent civil society organizations and trade unions on social protection reforms, integrating their input into the provision of technical assistance.
• Increase support and resources to expand the network of registration centers administering social assistance programs across the country, particularly in rural areas.
• Provide support and technical assistance to set up mobile cash units, particularly in areas far from city centers.

 

Methodology

Human Rights Watch conducted 70 interviews between October 2021 and April 2023 for this report. Thirty-six interviews were with individuals or families who had applied for Takaful, and another nine were with people who applied for support without specifying whether it was for Takaful. Seven interviews were with business owners, supermarket employees, and others who explained trends in the price of goods and services and general living conditions.

Human Rights Watch conducted five interviews with employees or owners of maktabehs – stationery stores or printing shops usually located in town centers that help people to apply for Takaful online, sometimes in exchange for a fee; these people shared their experiences helping dozens of Takaful applicants. Human Rights Watch also interviewed staff at five civil society organizations and charities, as well as two community development centers providing services in Tafilah governorate. NAF employees based in Tafilah governorate and Ministry of Social Development (MOSD) employees in Karak governorate were also interviewed.  

All interviews were conducted in Arabic or English in person or by phone. Human Rights Watch explained the purpose of the interviews and obtained informed consent to use the information the people provided. In cases where interviewees asked to not be named or where we assessed that naming them would jeopardize their privacy or security, we have used pseudonyms or withheld identifying information.

Human Rights Watch supplemented these interviews with an analysis of posts and comments published on two Facebook groups between March 2022, around the time people were notified whether they received cash transfers that year, and October 2022.

One of these groups is focused on questions and discussions about Takaful. It is set to private but open to anyone to join, provided they have a Facebook account. The other group, which is public, was also focused on Takaful at the time that Human Rights Watch was conducting research for this report; it has since become a general discussion group.  This analysis, though not statistically representative, helped to identify key issues and concerns among a larger group of applicants and beneficiaries.   

This report also draws on data published by Jordan’s Department of Statistics, as well as publicly available data and reporting on Takaful and general conditions of poverty and inequality in the country. The report relies on analysis of Jordan’s economic outlook and Takaful’s implementation and performance provided by major international organizations, including the World Bank, the International Monetary Fund (IMF), and UNICEF.

The report’s analysis of the broader implications of digital government drew on loan documents covering World Bank-financed social protection programs in the Middle East and North Africa, related Bank reports and policy documents, and research by the social protection research consultancy Development Pathways, the digital rights nongovernmental organizations Access Now and Privacy International, and the US-based Center for Human Rights and Global Justice at New York University.

Correspondence with the Jordanian government, World Bank, and other actors

Human Rights Watch wrote to the World Bank and UNICEF with detailed questions and concerns about Takaful and targeted social protection programs more generally on September 13, 2022, and to Jordan’s National Aid Fund on September 26, 2022. UNICEF and the World Bank responded in writing on October 6, 2022, and October 7, 2022, respectively. Additionally, Human Rights Watch met with staff members of the World Bank’s Jordan country team and Social Protection and Jobs Global Practice on October 12, 2022. NAF did not respond in writing at the time, but Human Rights Watch held a detailed, on-the-record discussion with agency leaders about the program on October 9, 2022, at the NAF headquarters in Amman.

Human Rights Watch shared a summary of its findings with the World Bank and UNICEF on May 8, 2023, and with NAF on May 11, 2023. On May 8, 2023, Human Rights Watch also shared a summary of its findings with other international organizations and donors that may have provided technical assistance to NAF, including the World Food Programme (WFP), the United Nations High Commissioner for Refugees (UNHCR), the United Kingdom’s Foreign, Commonwealth and Development Office (FCDO), the United States Agency for International Development (USAID), the German Agency for International Cooperation (GIZ), and the Dutch Ministry of Foreign Affairs. Responses to these requests for comment are summarized below.

Summary of World Bank’s Response

The World Bank reiterated its “commitment to advancing the implementation of universal social protection (USP), ensuring access to social protection for all persons…while also taking into consideration each country’s context and fiscal situation.”

It acknowledged that delivering social protection programs “requires the use, and sometimes collection of data,” including basic identifying information and socioeconomic data. It emphasized that information systems and technology can “facilitate” the delivery of social protection programs, but “are not substitutes” for interactions between institutions and people. However, it maintained that countries with “more mature delivery systems,” such as social registries, “interoperability with other government databases,” and payment systems, “generally performed better in providing emergency benefits and shock relief.”

With regards to Takaful, the Bank clarified that this was one of several programs geared “towards achieving USP in Jordan.” It confirmed that it has provided the Jordanian government with technical assistance to develop and improve Takaful’s registration, enrollment, payment, and grievance redress mechanisms. It also confirmed that it is working with NAF to refine the targeting algorithm and expects to disclose the results of this evaluation in July 2023.

The Bank clarified that the total value of the Jordan Emergency Cash Transfer Covid-19 Response Project, which helped expand Takaful, is US $1,014.17 million, and that “99.7% goes directly to the beneficiaries as cash transfers.” Only the remaining amount or less than 0.3% of the total project cost, the Bank said, is directed towards operational expenses and ensuring “implementation oversight.”

The Bank acknowledged the “additional cost that beneficiaries bear due to cash-out fees required by payment service providers” and is “exploring redress options.” It also noted that the NAF is exploring ways to notify beneficiaries that are subsequently excluded from the program of their changed status “in a more timely manner, prior to the cessation of payments.”[6]

Summary of NAF’s Response

NAF clarified that it evaluates applicants for Takaful in two stages: 1) determining a household’s eligibility based on administrative data obtained from a centralized data repository known as the National Unified Registry (NUR); and 2) ranking the eligible households according to the severity of their poverty based on income. This ranking is calculated using a targeting formula based on “fifty-seven indicators in several areas,” with each indicator assigned a “certain weight.” These indicators cover “levels of spending on basic needs, family income, housing, health, education, property and assets, family size, and geographical area…based on the concept of multidimensional poverty.” Specific information the NAF provided on various indicators are reflected in the report where relevant.

NAF also explained that “80% of the data in the registration [or application] form comes from the NUR, and the rest of the data is filled in by the citizen.” For workers with formal jobs, NAF obtains information about their income from the Social Security Corporation’s database, through the NUR. Informal workers are required to report their income on the form. NAF also requires applicants to report their living expenses; however, these figures should be no more than about 20 percent of their income. NAF explained that it established this restriction after finding that many applicants were declaring they had no income and high living expenses.

NAF added that people can apply for Takaful online, at one of NAF’s 68 branches and 290 registration centers across Jordan, or through mobile registration stations that make stops in extremely rural areas. NAF also said that people can “request support through multiple channels, such as the online form available through NAF’s website, through phone at NAF’s support center, through the National Call Center, or by visiting any of NAF’s field branches.”[7]

Human Rights Watch also met with agency leaders on May 30, 2023, who provided additional information about the targeting algorithm and clarified other details about the program.

Summary of Other Responses

UNICEF said that “the Takaful Programme is an initiative fully owned and implemented [by] the Government of Jordan,” and that its role is confined to providing “technical advice on international best practices.”[8] (In its October 6, 2022 letter, UNICEF said that it had provided other forms of technical assistance, such as providing IT hardware and equipment and “facilitating the human resources” to support data collection during the registration process).[9]

WFP said it “does not have any active or direct role on supporting NAF’s targeting algorithm,” but has helped NAF set up its digital payment system, home visits with beneficiaries, complaints and feedback mechanisms, and monitoring and evaluation process.[10]

FCDO said that it has provided financial and technical assistance to the social protection sector in Jordan since 2018, and that it helped finance the ongoing evaluation of the targeting algorithm that will be completed mid-2023.[11]  

USAID said that the funding it provided for Takaful was “primarily” to support the distribution of emergency cash transfers to informal workers affected by the Covid-19 pandemic (a sub-program known as Takaful-3). It added that it has had “no direct involvement” in establishing Takaful’s targeting algorithm, and “minimal engagement” with NAF since December 2021.[12]

UNHCR said that it has not provided technical or financial support to NAF since 2019, and has not been involved in the agency’s targeting decisions.[13] GIZ said that it is “not involved in the NAF related cash transfers.”[14]

The Dutch Ministry of Foreign Affairs had not responded as of June 7, 2023.

Human Rights Watch’s letters and the various responses are included in the Annex, and relevant information from those communications are provided throughout the report.

 

I. Public Sector Automation and the World Bank

Governments worldwide are turning to algorithmic systems to improve and streamline their delivery of essential public services. They are using algorithms to distribute cash and food assistance, allocate housing and job support, investigate child maltreatment, and detect social security fraud.[15] Advances in artificial intelligence (AI) and big data have fueled the narrative that these technologies can help governments decide who to prioritize for these services and regulate access, while lowering costs and improving efficiency.[16]

Public sector algorithms are frequently shrouded in secrecy, making these claims difficult to verify.[17] But a growing body of research has documented how some of these algorithms cut people off from income support and other essential benefits in error and with little or no explanation[18]; disproportionately exclude women, racial and ethnic minorities, migrants, and people with disabilities from services[19]; and single people out for investigation based on arbitrary criteria or stereotypes about poverty.[20] In some cases, these harms reveal structural problems that algorithms cannot fix and may even amplify, such as inadequate budget allocations, flawed policies, or longstanding discrimination.[21]

Automated Poverty Targeting

Amid these concerns, international organizations ranging from the World Bank, the IMF, UNICEF, United Nations Development Programme, and the WFP are pouring funding and resources into projects to automate public services, particularly in the Global South.[22]

As the world’s largest provider of social protection loans, the Bank has enormous influence over how borrower governments digitize and automate their social protection systems. The Bank has emphasized that “information systems and technology” can facilitate the delivery of social protection programs, but “are not substitutes” for interactions between institutions and people.[23] It has also warned against “inappropriate technology choices” that are “ill-suited to local needs and context.”[24] But some of the projects it supports are raising serious human rights concerns. Chief among them is the automation of poverty targeted programs, which target cash transfers and other benefits to people based on their socio-economic status.

What is “social protection”?[25] There is no uniform definition of social protection, and it is sometimes used interchangeably with the term social security. Under international human rights law, “social security” refers to a set of individual entitlements that protect against income insecurity throughout people’s lives, including during common life events such as old age, unemployment, sickness, or birthing and caring for dependents. Social security programs that do not require direct contributions from individuals as a condition of eligibility is known as “social assistance.” These programs are mainly financed through taxes and, in some cases, grants or loans from development actors. In this context, social protection refers to a broader range of programs that include all forms of social security as well as other programs that are essential to the realization of economic, social, and cultural rights, including the rights to food, water, education, and healthcare. These include school meals, housing assistance, and personal social services like childcare and support services for older people.   In Jordan, however, “social security” refers to a social insurance scheme funded by contributions from insured persons and employers.[26] To avoid confusion, this report will refer to Takaful and the country’s other non-contributory programs as “social assistance,” and to broader initiatives and approaches essential to realizing people’s economic, social, and cultural rights as “social protection.”

The Bank’s investments in data-intensive systems to improve targeting are growing even though it has recognized the need for a radically different approach to social protection. In 2016, it announced a partnership with the International Labour Organization (ILO) to advance universal social protection – “a nationally defined system of policies and programmes that provide equitable access to all people and protect them throughout their lives against poverty and risks to their livelihoods and well-being.”[27] Unlike poverty-targeted programs, universal schemes provide a range of benefits to all people within specific categories, such as children, people with disabilities, or people over a certain age. The Bank reaffirmed its commitment to these principles in its 2022 strategy on social protection and jobs.[28]

Multiple studies have shown that universal programs are far better at reducing poverty and inequality and protecting people’s rights to social security and related economic and social rights.[29] In contrast, targeted programs are prone to error, mismanagement, and corruption, stigmatizing for beneficiaries, and inadequate at protecting people against income insecurity.[30]

The Bank insists, however, that poverty targeting is one of the most cost-effective ways for governments with limited fiscal space to bridge gaps in their social protection coverage.[31] It has also made significant investments in technology and data-driven solutions that it believes will go a long way to mitigate chronic problems with traditional methods of targeting.[32]

The ILO and others have challenged the Bank’s economic justification of poverty targeting, arguing that governments can raise new revenues or reallocate existing funds to establish universal schemes, and that such spending can spur economic activity that more than pays for its cost.[33] Using Jordan’s Takaful cash transfer program as a case study, this report examines the other plank of the Bank’s case for continuing poverty targeting: that the judicious use of technology will improve targeting accuracy and the delivery of cash transfers.

A key pillar of the Bank’s strategy to automate poverty targeting is to finance the creation of social registries – a system that enables the collection and analysis of vast amounts of data about people to determine their eligibility for social assistance programs.[34] Social registries typically support a type of poverty targeting known as the Proxy Means Test (PMT), which uses an algorithm to rank a household’s level of welfare from poorest to richest.[35] This evaluation is based on a variety of data points about the household that serve as proxies of its income, such as the employment and education history of each household member, their type of dwelling, the assets they own, and even how much electricity and water they consume.[36] Households with a qualifying ranking or score are selected to receive benefits.

The Bank has observed that the PMT mechanism “is something of a Rorschach test for those who think about targeting.” A 2022 Bank report on targeting social assistance found:

Many, especially in ministries of finance or planning, see it as modern, scientific, data driven, replicable, and thus good for preventing patronage politics in social programs and safeguarding their reputation. Some communities find it a black box, a mystery. Analysts and observers have mixed opinions. Some find it a realistic, if imperfect, solution to a problem without perfect solutions; some find it anathema for its inbuilt statistical errors or lack of transparency.[37]

Between 2013 and 2022, the number of countries operating social registries jumped nearly threefold, from 23 to 60.[38] Social registries traditionally rely on data collected from household surveys and mass registration exercises, which are expensive and time-consuming and therefore conducted only infrequently and are limited to certain populations and geographic areas.[39] A 2021 report by the social policy think tank Development Pathways found that this type of data collection produces a “static” snapshot of people’s needs that quickly becomes outdated and inaccurate:

So, if a survey is undertaken in 2010 and a household is assessed for a programme in 2015 without another survey having been undertaken in the meantime, it will be assessed against its situation in 2010. It is as if nothing has changed over time. In effect, therefore, social registries are based on the flawed assumption that we live in a static world, as if household characteristics remain constant in a form of suspended animation. Yet, in the real world, even one specific characteristic of a household can change rapidly.[40]

The Bank has acknowledged the substantial risk of error and exclusion associated with registries that generate “fixed lists of registrants and beneficiaries,” but insists that modernizing them will help achieve the “progressive realization” of universal social protection.[41] It argues that the future of social registries lies in high-tech, data-intensive capabilities that automate eligibility checks, and cycle people in and out of social protection programs based on real-time monitoring of changes in their finances and circumstances. These so-called “dynamic” registries will help governments with limited social protection budgets efficiently substitute people that are no longer eligible for support for those with qualifying needs.[42]

To spur their creation, it is financing multiple projects to upgrade relevant databases and infrastructure and increase data sharing among government agencies.[43] In the Middle East and North Africa alone, a Human Rights Watch analysis of World Bank loan documents found that eight out of ten borrowing countries have taken out loans to introduce or upgrade social registries and related poverty targeting infrastructure. One other country, Yemen, has obtained a loan to finance the targeting of cash transfers to beneficiaries that the government had previously identified for other social assistance programs using PMT.  

 

Table 1: Active Social Registry and Poverty Targeting Projects in Middle East and North Africa

Country	Project Name	Project Approval Date	Relevant Project Objectives
Egypt	Takaful and Karama Cash Transfer Expansion and Systems Building Project	Dec-22	Expanding PMT to target cash transfer and pension programs, and expand the social registry to support PMT.[44]
Iraq	Iraq Social Safety Net Enhancement Project	Nov-22	Upgrading social registry to serve as a “gateway” for all services and benefits provided by the Ministry of Labor and Social Affairs.[45]
Lebanon	Lebanon Emergency Crisis and COVID-19 Response Social Safety Net Project	Jan-21	Expanding cash transfers using poverty targeting and establishing a national integrated social registry.[46]
Jordan	Jordan Emergency Cash Transfer COVID-19 Response Project	Jun-20[47]	Improvements in poverty targeting, rollout of a monthly recertification process, and expansion of National Unified Registry to serve as a “single gateway” for social protection services. Also, funds rollout of digital payments.[48]
Mauritania	Social Safety Net System Project II[49]	Mar-20	Updating and enhancing the social registry, and promoting usage; strengthening and extending Social Transfer Program, Tekavoul.[50]
Tunisia	Digital Transformation for User-Centric Public Services	Jun-19	Creation of a social registry, with “machine learning and big data” integrations to “improve social security compliance and fraud detection.”[51]
Morocco	Identity and Targeting Systems for Social Protection Project	Mar-17	Development of a social registry and a digital ID ecosystem.[52]
Occupied Palestinian Territory	Social Protection Enhancement Project	Jul-17	Creation of social registry to expand social protection coverage while “minimizing errors and fraud.”[53]
Yemen	Emergency Social Protection Enhancement and COVID-19 Response Project	Dec-20	Targeting of cash transfers to households previously identified for benefits through PMT and categorical targeting.[54]

Other “Digital Government” Initiatives

Dynamic registries are part of a broader transition to a “digital public infrastructure” that the Bank deems essential to making not just social protection programs but all government services more inclusive and equitable.[55] Its “Identification for Development” initiative has committed more than US $1.5 billion in financing to help 35 countries build digital identification systems and update their population records – a move the Bank says will help governments quickly verify the identities and needs of people seeking access to public services.[56] A related initiative to digitize “Government-To-Person Payments” has provided 33 countries with technical assistance to build payment systems such as e-wallets, direct deposits, and biometrically verified transactions.[57]

The Bank imagines a future where identification, payment, and data exchange systems like social registries “work together and interoperate seamlessly” to “underpin digital transactions and connections for people, businesses, and governments, including service delivery and operations across the public and private sectors, including financial services, e-commerce, education, healthcare, transportation, social safety nets, taxation, and business and property registration.”[58]

This seemingly limitless expansion of “digital government,” however, has undermined people’s rights to social security.[59] Human Rights Watch and others have found that digital ID systems are prone to error and bias, wrongly excluding people from much needed support.[60] The Bank has promoted India’s biometric ID system, Aadhaar, as a paradigmatic example of digital government that has “significantly streamlined the delivery of public services,” and announced a partnership with the government to replicate this system in other countries.[61] But researchers have found that, in many cases, the technology has also led to people being denied access to these services, whether because of the failure to link their benefits to Aadhaar or biometric authentication errors.[62]

Humanitarian experts have raised similar concerns about digital payment systems. Although intended to increase financial inclusion, they can have the opposite effect for people that cannot provide proof of their identity to authenticate transactions or otherwise lack the digital and financial literacy to access and manage payments.[63]

The Bank has acknowledged that while the digital transformation of social protection can “offer the promise of more efficient and effective inclusion, more needs to be done to ensure that they benefit everyone.”[64] To close the gap, it has offered guidance and technical assistance to governments on how to make digital registration and payment processes more user-friendly, and partnered with the private sector to increase internet access, smartphone adoption, and digital literacy in low- and middle-income countries.[65] Despite these efforts, disparities in access persist: the industry body Groupe Speciale Mobile Association (GSMA) estimates that the number of people not connected to mobile internet rose from 3.2 billion people in 2021 to nearly 3.4 billion in 2022.[66]

The increasing reliance on personal data to administer social protection schemes also poses grave threats to the right to privacy. The last decade has marked the rise of ever-more intrusive forms of government surveillance, enabled by advances in biometric recognition technologies, a thriving market for spyware and data brokers, and the vast reserves of user data amassed by tech companies.[67] Without adequate privacy and data protection safeguards, privacy and digital rights experts warn that social protection databases could become another tool of surveillance and repression, particularly against activists and marginalized groups.[68]  

The World Bank has provided assurances that it conducts due diligence during the project planning and design phases to assess whether a country’s laws and policies meet international data protection standards. Where they fall short, the Bank will include terms in its loan agreements requiring governments to adhere to these standards.[69] Despite these apparent protections, the UK-based NGO Privacy International has questioned whether the Bank enforces these terms. Moreover, of the eight countries it studied where the Bank financed Covid-19 response projects, only four had implemented data protection legislation.[70]

In any event, a coalition of civil society organizations has warned that data protection safeguards may be inadequate in “environments where human rights risks are too high, or where evidence-based policymaking, civil society engagement, rule of law, and rights-based assessments are simply not possible.”[71] In these environments, more drastic action, such as declining or withdrawing support for digital ID and other data-intensive systems, may be necessary.[72]

Privacy and Data Protection in Jordan At the time of publication, Jordan’s parliament was considering a bill that would establish a data protection law.[73] The government first proposed the law in 2014 and the last version was made public in January 2022. An analysis of the bill by the digital rights group Access Now found major loopholes that “cas[t] serious doubts about whether the legislation will protect people’s privacy and personal information.”[74] One of these loopholes is the proposed structure of the Personal Data Protection Board, which is supposed to be an independent oversight authority responsible for enforcing the law and investigating data protection complaints, including against the government. The bill, however, would designate the Minister of Digital Economy and Entrepreneurship (MODEE) as the chair of the board, with the power to appoint board members, and reserves board seats for representatives of security agencies.[75] Jordan Open Source Association, a Jordanian digital rights group, has urged changes to the bill that would protect the independence of the board, including the appointment of board members from non-governmental institutions.[76] Jordanian law also grants government authorities broad discretion to monitor people’s communications and personal data, which the data protection law is unlikely to change.[77] Access Now reported that, in 2018, the Jordanian government exploited its access to personal data held by telecommunications and internet service providers to discourage protests.[78] Without meaningful safeguards to prevent such abuse, the extensive collection and analysis of personal data required to administer the country’s social protection schemes leave beneficiaries particularly vulnerable to surveillance-based persecution and discrimination.[79] “Many of the recent protests in Jordan stem from economic grievances,” said Marwa Fatafta, MENA Policy Manager at Access Now. “The government of Jordan has a history of cracking down on protests and labor unions. If some of these protestors are also applying for government support, the data collected about them during the registration process could be weaponized to restrict their rights or deny them financial support in reprisal.”[80] In response to a request for comment, the World Bank said that “Takaful and NAF abide by the data protection regulations of the GOJ [government of Jordan], which are administered by the MODEE,” and that “all applicants are asked to provide their consent before NAF might access administrative data on them to assess and verify their eligibility.”[81] However, the Bank and NAF have indicated that access to this data is required to verify and evaluate people’s applications, raising doubts about the ability of applicants to provide meaningful and informed consent.

The introduction of dynamic social registries heightens the risks to privacy and access to social protection, making them even more difficult to mitigate. The extensive profiles compiled on beneficiaries to facilitate eligibility decisions increase the pool of personal data vulnerable to surveillance. At the same time, more data does not necessarily lead to better decisions.[82] People may be wrongly excluded from support when aspects of their experience with poverty elude conventional measures of hardship: owning a car or a business, for example, fails to capture whether the family can afford petrol, or whether their business is laden with debt.

In contrast, universal social protection schemes are, by design, more likely compatible with privacy and data protection standards, as they do not require intrusive data collection about people’s income and socio-economic status. Universal child benefit programs, for example, typically require applicants to only submit copies of their children’s birth certificates.[83] The lower evidentiary burden also improves accessibility: UNICEF has found that, “the lower the number of documents required, the higher the level of access for the most vulnerable and disadvantaged children.”[84]

 

II. Case Study: Automated Poverty Targeting in Jordan

In May 2019, the Jordanian government launched the Takaful cash transfer program, part of a five-year strategy to modernize its social protection system.[85] It was established with the help of two World Bank loans approved in 2018 and 2019 totaling US$2 billion, which were intended to help the government implement sweeping economic reforms under a series of International Monetary Fund programs.[86] The reforms, which drew widespread protests, removed fuel subsidies, increased the sales tax, lowered the threshold for income tax, and restructured electricity tariffs to raise prices for residential use.[87] The Bank loans earmarked $200 million to help the government build “better and more efficient” social assistance programs that would ease the hardship caused by resulting price hikes.[88] Improving poverty targeting would be central to this approach.[89]  The Covid-19 pandemic led the Bank and other development partners to finance a significant expansion of Takaful, as part of a $1.014 billion Covid-19 response package.[90]  

Jordan’s development partners have also provided extensive technical assistance to the National Aid Fund, the Jordanian social protection agency responsible for implementing Takaful. A Technical Working Group comprising the Bank, UNICEF, the WFP, and representatives of development aid agencies has provided NAF with equipment and guidance on enhancing the Takaful algorithm, among other forms of support.[91]

Background

Jordan has suffered from persistently high rates of poverty and inequality. In 2018, the government recorded 15.7 percent of the population living under what it deems to be the poverty line, an increase from 14.4 percent in 2010-2011.[92] Independent experts say that the poverty rate was likely higher: a study of government survey data that took into account changes in inflation rates and household size found that the poverty rate in 2017 was 22.2 percent.[93]

Successive economic shocks caused by the Covid-19 pandemic and the conflict in Ukraine have increased pressure on the government to deliver on its social protection reforms. The Ministry of Planning recently announced that the poverty rate jumped to 24.1 percent in the first quarter of 2022.[94] The IMF’s most recent analysis of Jordan’s economic outlook also raised concern that “the labor force participation rate for Jordanians [the proportion of Jordanians in the workforce], at 33.5 percent, is yet to recover to pre-pandemic levels… and is indicative of some economic scarring.”[95] The unemployment rate, the proportion of Jordanians in the workforce without a job, “remained elevated” at 22.6 percent in the second quarter of 2022.[96]

The Jordanian government is betting on Takaful to reverse these trends and previous poverty targeting failures. Since 2004, it has relied on poverty targeting to administer various social assistance programs.[97] Targeted programs prior to Takaful, the 2019 strategy noted, reached only a “minority of poor Jordanians,” because of budgetary constraints and inefficient targeting methods.[98] The government struggled to verify people’s income and accurately target cash transfers because, “[u]ntil recently, the technology for inexpensive and rapid access to administrative records from relevant ministries did not exist.”[99]

The government has since modernized the technology systems it uses to administer social assistance with the help of the Bank and other international partners. It has established an online portal where people apply for and manage Takaful and other social assistance benefits.[100] It has also automated the vetting, selection, and payment of Takaful beneficiaries through upgrades to the NAF’s data systems, including the Management Information System and the NUR.[101] The NUR, which was created in 2013 with $2.46 million in Bank financing, began as a social registry to facilitate poverty targeting.[102] According to the government and the World Bank, the NUR is now being expanded to serve as “a single gateway for Jordanian households to access a range of social services and programs.”[103]

The government plans to consolidate its social assistance programs into a single program and require everyone to apply for and manage their benefits using the Takaful platform.[104] In the meantime, Takaful complements other social assistance benefits targeted at people living below the national poverty line or with specific vulnerabilities, such as orphans, older people, and households headed by people with disabilities.[105]

By some measures, Takaful has been a success. The Bank has praised Jordan for “significantly shorten[ing] the time needed to roll out its pandemic response packages” because of investments in their social registries and “good use of digital technology.”[106] At the beginning of the Covid-19 pandemic, NAF disbursed three months of cash transfers to 237,000 households that were targeted for their reliance on informal sources of income (a program known as Takaful-2).[107] Households with stateless Palestinians who hold temporary Jordanian passports and Jordanian women married to non-Jordanian men were eligible to receive these transfers.[108] In December 2020, the government began another round of emergency transfers but restricted them to Jordanian households (Takaful-3).[109] These transfers, which lasted for 12 months, reached 160,000 Jordanian households.[110]

Household vs. Individual Access to Social Assistance Takaful distributes cash transfers to the heads of households rather than individual members. This design perpetuates longstanding gender discrimination. Jordan’s Civil Status Law designates the husband or father as the head of the household by default and authorizes women to assume the role only if they are divorced, widowed, the oldest unmarried child in a family where the father is deceased, or married to a non-Jordanian.[111] This effectively forces many women to obtain government support through their father or husband.[112] While women married to non-Jordanian men can issue a family book in their name, they cannot have their children added to it.

The Bank has also credited Takaful for expanding regular, non-emergency cash transfers to an additional 120,000 households in 2022 (previously Takaful-1 and now also known as the Unified Cash Transfer program). It argues that this makes Takaful one of the “most redistributive and cost-effective programs currently active in Jordan.” [113]

Analysis of poverty and inequality trends in Jordan by other international organizations, however, raises doubts that Takaful has been an adequate buffer against economic hardship. The ILO has found that, even with Takaful, the country’s social assistance schemes continue to reach “very few Jordanians,” with “less than 10 percent of households, and less than a third of those officially classified as poor” able to access these schemes.[114] While the IMF has praised Jordan for efficient social protection spending, it has echoed these concerns, noting that demand for Takaful has outstripped coverage.[115] In the first six months of 2022, over 400,000 households applied for 120,000 program slots.[116] 

How Takaful Works

To apply for Takaful, the head of a household must fill in an application questionnaire. The questionnaire requires applicants to submit their identity information, such as their name and national ID number, as well as information to facilitate a detailed analysis of their finances, such as sources of informal income, living expenses, and electricity and water meter ID numbers (so that NAF can obtain data about their electricity and water consumption from relevant utility providers). NAF told Human Rights Watch that “80 percent” of the questionnaire is populated with data from the NUR and the rest of the data is provided by the applicant.[117] The head of the household is periodically required to update this information online.[118]

NAF has established a two-stage screening process to assess which households will receive cash transfers.[119] During the first stage, NAF assesses whether applicants have met basic eligibility criteria. Households are eligible if the person applying is a Jordanian citizen with a national ID number and the head of household does not work in the public sector or for the military. Households that receive any other type of benefit from NAF are not eligible, except in certain circumstances, including if one or more family members live with disabilities.[120] NAF also assesses whether households are living under the official poverty line.[121]  

Because demand exceeds the number of cash transfers available, however, not all eligible households are enrolled as beneficiaries.[122] Instead, during the second stage, NAF applies a targeting algorithm modeled after the PMT method to assess the “severity of their poverty,” ranking them from least poor to poorest.[123] Households deemed poorest are enrolled first, followed by those with the next highest level of poverty until all program slots are taken. Those that are not selected for cash transfers are notified via text message or through the online portal that they are “not among the poorest.”[124]

This screening process relies extensively on the NUR, which provides a detailed picture of each household’s interactions with thirty-seven government agencies, including those responsible for registering vehicles and businesses, issuing passports and family books, and administering pensions and social insurance schemes.[125] Eligibility determinations during the first stage rely entirely on data drawn from the NUR about each applicant household.[126] Poverty targeting during the second stage relies on both data from the NUR as well as data reported through the application questionnaire.[127] NAF also conducts in-person or remote visits to people’s homes to verify income and welfare details.[128]

Until 2021, people’s ranking controlled the size of people’s benefit payments.[129] Under Takaful-2 and Takaful-3, benefit amounts varied from 39 to 136 dinars ($55 - $192).[130] NAF told Human Rights Watch that they stopped relying on the algorithm to calculate benefit amounts following complaints that this was arbitrary and confusing.[131] Benefit amounts now begin at 40 dinars ($56), increasing to a maximum of 136 dinars ($192) depending on the number of family members in the household and whether they are facing life events such as disability or old age.[132]  

Poverty Targeting Under Takaful

According to NAF, the targeting algorithm is programmed to estimate a household’s income and wealth based on 57 “socio-economic indicators related to the welfare status and deprivations on the households’ level.”[133] The households are subsequently ranked from least poor to poorest. If the applicant has reported an income greater than the estimate based on the 57 indicators, the higher income is used to determine their ranking.[134]

NAF declined to provide Human Rights Watch with the full list of indicators, stating that this was for internal purposes only and that the indicators and the weights assigned to them were “constantly changing.”[135]

However, NAF explained that these indicators cover “levels of spending on basic needs, family income, housing, health, education, property and assets, family size and geographical area.”[136] It added that some of these indicators could be disaggregated into sub-indicators; for example, “dwelling characteristics” contain a number of sub-indicators, such as the type and value of dwelling.

NAF acknowledged that owning a car less than five years old or a business with 3,000 dinars ($4,231) or more in capital would automatically disqualify families from the program. But these were exceptions to the general rule that none of the indicators or sub-indicators would, on their own, exclude a household from Takaful. Instead, each indicator or sub-indicator is assigned a certain weight in the targeting process, indicating its relative importance in determining the household’s ranking. For example, a household headed by a woman would be more likely to qualify for cash transfers than a household headed by a man, all other indicators and sub-indicators being equal.   

In correspondence and discussions with Human Rights Watch, NAF elaborated on some of these indicators and the weights assigned to them. This report’s analysis of the algorithm relies in part on these explanations, which are up to date as of June 9, 2023. 

Key indicators of the targeting algorithm include:[137]

• Head of household: applicants can be more likely to qualify for cash transfers if they belong to female-headed households, as opposed to male-headed households.
• Illness: applicants can be more likely to qualify if one or more of their household members have a chronic illness or medical condition.
• Household size: applicants can be more likely to qualify if they support a larger family with more children. However, households headed by Jordanian women married to non-Jordanian spouses are treated as a family of one (discussed below).
• Geographic area: applicants can be more likely to qualify if they live in governorates with fewer job opportunities. The weights for each governorate are defined based on the Household Expenditure and Income Survey.
• Cars: Applicants that own cars that are less than five years old are automatically excluded from Takaful because this is deemed to indicate a higher standard of living. Applicants that own a car that is more than five years old are not automatically excluded but can be less likely to qualify. Owning a car for family members with disabilities does not affect the applicant’s ranking, in line with the Law on the Rights of Persons with Disabilities. Other factors considered include the number of cars owned and the value of the car.    
• Business: Applicants that own a business valued at 3000 dinars or more ($4,231) are automatically excluded. For applicants that own businesses of lesser value, they are not considered business owners for the purposes of their ranking. Instead, the value of their business is attributed to their overall income. Owning a business that is inactive does not affect the applicant’s ranking.
• Livestock: The algorithm is programmed to assume that each animal owned generates a certain amount of income for the household. These amounts are based on data collected by the Ministry of Agriculture. 
• Housing characteristics:  The algorithm evaluates characteristics such as the size and the type of dwelling (e.g. a tent or an apartment), the applicant’s ownership share, the type of water or electricity supply (e.g. connected to a public or private network), and how much electricity and water the household consumes. Priority is given to applicants living in tents or tin structures over those living in apartments or houses. Applicants can be less likely to qualify the more electricity and water they consume.
• Living Expenses:  Applicants are required to report their “levels of spending on basic needs,” such as electricity, housing, food, education, and loan repayments. The algorithm relies on this information to calculate the applicant’s ranking in two ways. 1) To estimate their household income, the algorithm takes into account self-reported expenses that are supported by official documents (such as water and electricity bills) and verifiable through administrative data. Other than water and electricity, it is unclear what types of expenses meet these criteria. 2) The applicant is required to report an income that is at least about eighty percent of their total living expenses. The applicant’s self-reported income is used to calculate their ranking if it is higher than the algorithm’s estimate.

Flawed Criteria

Human Rights Watch found that the practice of distilling the economic complexity of people’s lives into a series of indicators using restrictive definitions and measurements results in a flattening of people’s hardships, a loss of nuance, and a failure to fully understand any individual family’s situation.

To be sure, some of these weights mirror well-established indicia of economic vulnerability in Jordan. Assigning a higher vulnerability weighting to female-headed households, for example, reflects the government’s recognition that “Jordan has one of the lowest female labour-force participation rates in the world.”[138] The gender employment gap is widening: according to the Department of Statistics, the unemployment rate for women in Jordan grew from 24.3% in 2020 to 33.1% in 2022, compared to 18.1% to 20.5% for men.[139] Female workers also face systemic pay disparities in the private sector: the ILO has estimated that the median wage for males is 7% higher than that for females, while the World Bank has found that women working similar jobs with similar education and experience as men earn roughly 17 percent less.[140]

But other weights are, at best, inconclusive measures of people’s economic standing, and potentially exclusionary. For example, the algorithm assumes that households with higher electricity and water consumption are less vulnerable than households with lower consumption, all else being equal. In NAF’s view, higher rates of consumption can imply higher household income. But a family’s electricity consumption may be higher because they are less well off: for example, a 2020 study of housing sustainability in Amman found that almost 75 percent of low-to-middle income households surveyed lived in apartments with poor thermal insulation, making them more expensive to heat.[141]

Low-income families may also use older, more energy-intensive appliances because they cannot afford to replace them. The government’s own social protection strategy indicates that less than 4 percent of people in the poorest decile can afford “cost-saving assets” such as solar water heaters.[142]  Nearly everyone Human Rights Watch interviewed about their electricity usage indicated that they were using 500 – 600 kWh per month – well above the median usage of 300 kWh per month.[143]

Several Takaful applicants whom Human Rights Watch interviewed also raised doubts about the reliability of asset-based profiling. Owning a business, a house, or a car made them or people they know less likely to qualify for support, even though these assets had little or no bearing on their economic situation.

A tailor in Amman’s historical downtown, locally known as al-Balad, received support from NAF in 2019, but the agency cut him off the following year. He believes that his business, which he has been struggling to keep afloat since 2011, played a role in the agency’s decision. Like many other small businesses, his tailoring shop was badly hit by the pandemic, forcing him to take out 12,000 dinars ($16,925) in loans to cover his electricity bills, rent, and other basic needs.[144]

An owner of a small shop in the village of Jurf al-Darawish, 150 kilometers south of Amman, was unsuccessful both times he applied for Takaful in 2021. He said that his business selling snacks, coffee, and car accessories was likely a factor. He bought the shop after he was fired from his job at a gas station in al-Hasa in western Tafilah Governorate, about 20 km away from where he lives. He thought that the shop, which is five minutes away from his home, would save him transportation costs and help him to earn more but he is barely turning a profit and struggling to support his family. The shop, which Human Rights Watch visited, was bare and low on inventory.[145]

A vegetable seller with four children in Marka, a neighborhood in east Amman, said that NAF informed him he was not eligible for support because he inherited a house in Tafilah. But he can no longer afford meat, struggles to pay rent, took out loans to cover basic needs, and faced electricity cutoffs. “We have to live in a tent for them to accept us,” he said.[146]

Human Rights Watch also found that asset-based profiling can force some people into an unacceptable tradeoff between their right to social security and the assets they need to exercise other economic and social rights, such as their rights to a decent living, health, and food. Five people Human Rights Watch interviewed believed that owning a car factored into the NAF’s decision to reject their application for Takaful, even though they needed it to work, or to transport water, firewood, and other essentials.

As a rideshare driver and tour guide in Amman, Tariq, 33, depends on his car to make a living.[147] But he believes that his means of livelihood was one of the reasons he was unable to obtain emergency cash transfers under Takaful at the beginning of the Covid-19 pandemic in 2020, when lockdown restrictions cut off the only source of income for him and his family.

Living outside the city can make people even more dependent on their cars for basic tasks. Mariam from Burbaita, whose family received Takaful in 2021 but was dropped from the program in 2022, believes that their car lowered her family’s chances of obtaining the benefit. She said:  

The car destroyed us. Look at it, it’s parked outside. We don’t use it. They consider it to be a transportation car [for transporting goods], but we don’t use it, we use it [only] to transport water and other needs. Sometimes we don’t have the money to fill it up with diesel, so we walk to the street and wait for someone to pass by and agree to pick us up.[148]

An owner of a forty-year-old car in Burbaita, who has not renewed his car registration in over a year and did not qualify for Takaful, agreed with Mariam:

These program decisionmakers in Amman, when they see in a person’s application that they own a car that is categorized as a transporting vehicle ( نقل مشترك) they think it’s something, but in reality as you can see, these cars are very old and we use them not only for transportation in the area, but also to go buy things for our needs and also transport the water that we get from the wadi [valley or ravine] downhill.[149]

Some people who need their car but are also desperate for support have resorted to transferring car ownership to relatives, friends, and neighbors. A community activist in Burbaita said he had six to seven cars under his name as a way to help others in need.  “Everyone comes and registers their car under my name,” he said. “People can’t give up on these cars because it transports their water and food, they can’t eat if they sell it, so they just officially change the name of the owner.”[150]

This tactic is not foolproof: another resident of Burbaita said that, after he transferred ownership of his car to his father, the government told him that its records still had his name on the title.[151] Other people might also sell their car without formally changing the title, unknowingly jeopardizing their ability to obtain benefits. Aseel said that her brother did not qualify for Takaful, even though he is 25,000 dinars ($35,261) in debt, cannot afford rent and struggling with depression. Aseel added that he sold his car to someone else but that it was still registered in his name, raising the possibility that this could have affected his ability to obtain Takaful.[152] 

Existing social inequalities may also transform even seemingly innocuous factors into proxies for discrimination. Household size, for example, appears to be a straightforward indicator of vulnerability – the more people a household must feed, the higher the need.

But this calculation is based on the family book, a government document issued only to Jordanian citizens and which lists the head of the household, the spouse, and all unmarried children.[153] The networks of family and kinship forged beyond this restrictive definition of a household eludes the algorithm. Since Aseel applied for her own family book after her divorce, she is, in the government’s view, a household of one. But she told Human Rights Watch that her home is a “family home.” “All my brothers and sisters come to my home to eat, so my expenses are more,” she said.[154]  

NAF also confirmed that the algorithm’s calculations are based only on the number of Jordanian members in the household, not its total size.[155] This leaves families of Jordanian women married to non-citizens disproportionately vulnerable to exclusion. While Jordan’s citizenship law permits men to confer nationality to their spouse and children, regardless of the nationality of their spouse, it denies women the ability to do the same, except in narrow circumstances where the father is of unknown nationality or stateless, or where paternity is unclear.[156] As a result, NAF’s algorithm treats most of these families as one-person households, no matter their size. Even if they are targeted for cash transfers, NAF confirmed that they would only be entitled to the minimum benefit amount: a meagre 40 dinars ($56) per month.

Encoding discriminatory restrictions into the country’s largest cash transfer program has pushed Hiba, a 36-year-old Jordanian woman, Abdullah, her 59-year-old Syrian spouse, and their three children deeper into poverty. Although her family received two payments of 136 dinars ($192) under Takaful-2, they were not eligible for Takaful-3 or the Unified Cash Transfer program.[157] “I was eligible the first time, why am I not eligible now?” Hiba wondered. “I don’t want anything except a roof over my head and my children.” She recalled a NAF employee, in rejecting her appeal, telling her that “your children are non-Jordanian, so we give you once and that’s enough.”[158]

In response to these findings, NAF emphasized that there is no single indicator or sub-indicator that would automatically exclude applicants from the program, except in cases where they own a car that is less than five years old and businesses worth 3,000 dinars ($4,231) or more.[159] Water and electricity consumption, the agency added, was “a sub indicator within a main indicator that is dwelling characteristics.”[160] Weights assigned to asset-based indicators depend on factors such as the value, age, and the type of asset owned. NAF also said that about 25 percent of Takaful beneficiaries own cars.[161] People who own businesses with less than 3000 dinars in capital are also not excluded from the program; instead, this is factored into the algorithm’s estimate of the household’s income.[162]

NAF also acknowledged that Takaful and other NAF programs were targeted only at Jordanian households with a national ID, but said that the Zakat fund, an Islamic charity assistance program, was available to families “regardless of the national ID.”[163] However, the Zakat fund only covers a small proportion of households struggling to make ends meet; benefit amounts are also typically small.[164]   

Technical Experts’ View of the Takaful Algorithm Human Rights Watch asked researchers from the Distributed Artificial Intelligence Research Institute (DAIR), which leads community-driven AI research, to review the following technical description of the algorithm provided by the Jordanian government: Takaful formula score uses 57 socioeconomic indicators related to the welfare status and deprivations on the households’ level, including the gender of the household head, recognizing the additional vulnerabilities FHHs [female head-of-households] face. The methodology categorizes poor households to 10 layers, starting from the poorest to the least poor, then each layer includes 100 sub-layers, using statistical analysis. Thus, resulting in 1000 readings that differentiate amongst households’ unique welfare status and needs… A technical simulation by the Bank team, using data from the Household Income and Expenditures Survey (HIES), shows that Takaful’s targeting methodology approximates well the poverty level of households. Nonetheless, NAF has slightly amended the targeting variables, to better address the crisis response, this included amending the assets thresholds and the informal income value.[165] Dr. Alex Hanna, Director of Research at DAIR, said: “These are technical words that don’t make any sense together. They are giving a bizarre description of what sounds like a neural network [a type of algorithm modeled after the human brain], but we really don’t know what kind of algorithm it is. Are the developers manually assigning weights to various indicators? Or are the weights learned and they are choosing to adjust them? We don’t know.”[166] Nyalleng Moorosi, senior researcher at DAIR, agreed. “This is a poor description of model architecture and I think they are using this language as technical obfuscation,” she said. “If you have good data, AI will work; if you have bad data AI will not work as well; and this is an area that doesn’t tolerate error very well. You are talking about people’s lives.”[167] While it is unclear whether the Takaful algorithm uses AI to generate households' ranking, the World Bank has expressed optimism that governments can leverage advances in AI to improve targeting accuracy.[168] In Tunisia, the Bank is financing the use of an AI-based technique known as machine learning to improve targeting accuracy and detect social security fraud.[169] These developments could pave the way for increased “predictive optimization” in social protection systems – the phenomenon of using AI to “predict future outcomes” and “mak[e] decisions about individuals based on those predictions.”[170] Other examples of predictive optimization that have raised human rights concerns include a risk scoring algorithm that was used in the Netherlands to assess an individual’s likelihood of committing welfare fraud, and the Correctional Offender Management Profiling for Alternative Sanctions (COMPAS) software, used in parts of the United States to assess a criminal defendant’s risk of committing future crimes.[171] Computer scientists from Princeton University and Cornell University have warned that these computer models are prone to inaccuracy and error since they are ill-suited to the complex task of understanding and regulating human behavior.[172]

Inaccurate or Incomplete Data

Computer scientists have long invoked the adage, ‘garbage in, garbage out,’ to warn of the limits of automated systems – these systems work only as well as the data they are provided and trained on, and shoddy data leads to shoddy, even dangerous results. In Takaful’s case, flaws in the data collected and compiled on people applying to the program can distort people’s ranking and decisions about who is ultimately selected as a beneficiary.

NAF relies on several sources of data to assess and rank the poverty level of eligible households. The NUR, NAF said, contains the majority of the data its algorithm uses to evaluate people’s financial and living situation.[173] Synced with the databases of thirty-seven government agencies, the registry monitors whether beneficiaries and their family members have, for instance, bought or sold a vehicle or a business, found a job in the formal labor market, or applied to other social protection schemes. The application questionnaire is designed to solicit information about what a NAF official described as “things we don’t know,” such as some living expenses and sources of informal income.[174]

Human Rights Watch interviews with applicants, along with information made public and provided by the government, show that this automation-assisted process of targeting beneficiaries is marred by inaccurate or incomplete household data that fails to provide a true picture of people’s needs.

Some of these inaccuracies stem from arbitrary restrictions that hinder proper data collection about people’s financial situations. Applicants, maktabeh employees who help people apply, and a NAF employee in Tafilah Governorate told Human Rights Watch that the online portal only accepts applications that list living expenses that are equal to or less than the household’s income. When people declare that they have no income, or expenses that exceed their income, the online portal generates an error message, preventing them from submitting their application.[175]

NAF officials Human Rights Watch met in Amman disputed this, saying that they will process applications so long as declared expenses do not generally exceed people’s income by more than 20 percent.[176] The agency said that it introduced this cap after an audit revealed that applicants were declaring that they had no income while citing high living expenses, perhaps in the hope of getting more support.[177] One maktabeh employee in Dhiban, 70 kilometers south of Amman, told Human Rights Watch that he has successfully submitted applications that list expenses exceeding income by 18 to 20 dinars ($25-$28), but only if the applicant has a medical condition.[178]

The pandemic-fueled rise in the cost of living, alongside stagnant wages, has made these restrictions nearly impossible to follow. In January 2023, the Consumer Price Index, which measures the average change over time in prices paid by consumers for goods and services, recorded a 3.77% increase compared with a year earlier, with fuel and electricity prices recording the highest increase (31.82%). Rent and dairy products also increased by 5.17% and 7.88% respectively.[179] In October 2022, the price of vegetable and dry goods had climbed by 7.21%.[180] In the meantime, the government has postponed a planned increase in the minimum wage – which is currently 260 dinars ($367) per month for Jordanian workers – until at least 2024.[181]

Abdelhameed, a father of two who makes 250 dinars ($353) a month working at a coffee shop in Amman, told Human Rights Watch that his family’s living expenses were frequently twice that of his income. “My rent is 140 dinars ($197). My phone, electricity and water bills, and other living expenses are 320 dinars ($451) … Each month, we just trust in Allah and hope we can cover our expenses. Some months, my siblings abroad help out.”[182]

“How can you expect someone to put expenses that are exactly like their salary, when their salary is 220 to 240 dinars ($310 - $338)? That is not realistic,” said Eyad, a maktabeh employee in Baqaa refugee camp a few kilometers north of Amman who has helped people apply for Takaful since its rollout in 2019.[183]

To overcome this restriction, ten applicants told Human Rights Watch that they inflated their income or lowered their expense figures on their application. “Ibrahim” (not his real name), who lives with his wife and four children in Zarqa and was laid off from his job as a pharmacist’s assistant in 2018, recalled, “I put my income as 300 dinars ($423) even though it’s not true to avoid the error message.”[184] Misha’al, 36, who earns 350 dinars ($494) per month, said he listed 350 dinars in expenses even though this is not nearly enough to support his family of five.[185] Aseel said that she lowered the figure she spent on food to 100 dinars ($141), even though this only covered some necessities, such as rice, oil, pasta, and beans.[186]

These workarounds are a band aid on a gaping flaw in Takaful’s targeting methodology, which fails to capture how people scramble to fill the gap each month, such as by asking family for assistance, buying on credit, selling household items, or taking on odd jobs. Aseel juggles a full-time job at a plastics factory and side gigs as a photographer: she postponed two semesters of business school to accommodate her work schedule, sometimes freelancing from 8 p.m. to 2 a.m. after her day job ended.[187] Suleiman, a father of five in Dhiban, and Motasem, a father of two in Ghor al Mazra’ along the southern edge of the Dead Sea, have racked up loans to scrape by. “You keep on taking loans and that’s how you fall into debt and then you become imprisoned,” said Motasem. (Jordanian law authorizes the imprisonment of people who fail to repay debts. The moratorium the government imposed on debt imprisonment expired on May 1, 2023.)[188]

Another potential source of inaccuracy and error is the application questions themselves, some of which require a methodical accounting of expenses and living conditions that some applicants found difficult to understand, or out of touch with how they get by and make do.

Abdullah struggled to provide a monthly estimate of how much he spends on his daughter’s education, since this figure varies according to how much money he has on hand, his daughter’s needs, and how many days of school she attends.[189] Atef was puzzled by the question asking him to provide his electricity and water meter numbers and a copy of his electricity bill: Families who rent, he said, often pay their landlords directly for their electricity, while others may share meters with other households.[190] This proved problematic also for Motasem when NAF asked him to submit copies of his electricity bills, since they were in his landlord’s name.[191] Abdelhameed, the employee at a coffee shop in Amman, summed up their dilemma: “The questions asked don’t reflect the reality that we exist in.”[192]

Case Study: The Multi-Dimensional Problem of Accessing Social Protection in Jordan This report focuses on how digitizing and automating one of Jordan’s cash assistance programs interferes with people’s social security rights, but the struggle to access social protection is multi-dimensional. Discriminatory laws, cumbersome administrative processes, and unresponsive bureaucracies are among some of the other barriers that people commonly experience. “Fatima” (not her real name) lives in East Amman.[193] She was married to an Egyptian man, who died in September 2022 after a long battle with respiratory illness. They have four children. The law does not permit Fatima to pass on Jordanian citizenship to her spouse or children, which severely limits their access to public services. Her children are eligible for non-citizen ID cards that grant them limited access to public services and the right to work, but getting the card is an expensive and time-consuming process requiring a medical test, interviews at various government offices, and extensive paperwork.[194] Fatima could only afford to obtain these cards for three of her children. Fatima first applied for Takaful during the Covid-19 lockdown in 2020, after hearing about the program on social media. At the time, Takaful permitted families of Jordanian women who have non-citizen children to receive assistance. This version of the program, known as Takaful-2, also relied on the targeting algorithm to generate benefit amounts. Fatima’s application was approved a month after she applied, and she received two to three installments of around 130 dinars, or $183 (she cannot recall the exact benefit amount or number of installments). She believes the size of the benefit payment was calculated based on the needs of four family members only, because neither her husband nor one of her daughters had a non-citizen ID card or Jordanian passport. “The amount was little especially during the lockdown, everyone was at home so there were a lot of expenses,” she said. At the time, her husband also required oxygen cylinders to help him breathe, further straining the family’s finances. After Takaful-2 ended, the program’s coverage narrowed and the government suspended non-Jordanians’ access to the benefit. When Fatima re-applied, the targeting algorithm processed her application as that of a single-person household, as she was the only family member with a Jordanian passport. This made her far less likely to qualify for support. Fatima was also perplexed by the requirement that she could only declare living expenses on the application form that were more than about 20 percent of her household’s income. Her family’s expenses far exceeded their income of 300 dinars ($423). “The way I understand it is that they’re saying that 300 dinars are enough to live,” she said. Fatima was rejected all three times she re-applied for Takaful.  “After that [the third] attempt, I stopped applying, there’s nothing I can do,” she said. She suspected that she did not qualify because she was married to a non-citizen, but there was no way to know. She submitted an appeal online. NAF affirmed its original decision but did not explain why she was not selected for cash transfers. This experience deepened Fatima’s disillusionment with government programs and reminded her of her struggle to renew her health insurance. She explained: They asked me to stamp my insurance from social security [the Ministry of Social Development] to show that I don’t receive a salary from them, then the land department [the Department of Land and Survey] to prove that I don’t own any, then the Driver and Vehicle Licensing Department in Sahab to prove that I don’t own any cars, in addition to a medical report to prove that my husband was sick. Why do they ask for all of these papers? They asked for a lot of papers and I had to go around to different areas, and needed money for transportation so I just canceled it [the insurance]. Everything in the country is complicated, nothing is easy here.

Unpredictable Benefit

Several people told Human Rights Watch that they experienced temporary disruptions to their benefit or were dropped from the program altogether for reasons that eluded them. This unpredictability hinders the ability of beneficiaries to budget effectively, contrary to the goal of the government and the World Bank to create social protection programs that build long-term financial resilience.[195] It also conflicts with the Bank’s guidance that cash transfers should be “predictable and regular.”[196]  

Khaled and Manal, a couple in Amman, both 42 years old, said that their payments stopped abruptly for two months in 2022. They believe that the payments were halted because their son got a temporary job paying the minimum wage, 260 dinars ($367) per month.[197] The payments re-started after they updated their online account to reflect that their son had lost his job. NAF disputed that something like this could occur, stating that it permits beneficiary households to stay on the benefit for a year in the event that a household member becomes formally employed.[198] This disruption, whatever its cause, increased the couple’s skepticism of the program. “They don’t tell you when the payment is and what amount it is, so we don’t depend on it,” Manal said.

A day laborer in Marka said that his hardships worsened after he was unexpectedly cut off from Takaful. “I haven’t paid rent in seven months…there’s no food…There’s no income at all,” he the day laborer said. “They cut off the electricity so I connected illegally because I can’t live like this…It’s been a month since we’ve tasted chicken.”[199] The man said he believed that he lost the benefit because his son secured a job which, if true, would be inconsistent with NAF’s policy of a one-year grace period.

Some households that received emergency cash transfers under Takaful-2 or Takaful-3 did not qualify for subsequent cash transfers, which have more stringent targeting requirements. Walaa, a resident of Baqaa Camp, said that her family’s income is not nearly enough to cover their living expenses. Although she qualified for cash transfers in 2020 and 2021, she did not qualify in 2022. Walaa struggled to make sense of this rejection:

The difference between when we first applied and now? Honestly, the expenses increased. At first, we were four people when we applied, and the second time, we were five people, and this increased our expenses. My youngest daughter – 4 months old – she refused my milk, so I had to get her formula. This time we are in a smaller house than the one before it even…our income doesn’t cover our expenses. Sometimes, we can’t pay what we need to pay.[200]

The destabilizing effects of benefit disruptions are also evident among the scores of beneficiaries that flood NAF and social media forums with appeals for help after their benefit is cut. In October 2022, NAF told Human Rights Watch that it had received approximately 100,000 appeals since January. The agency attributed the high volume of complaints to the tightening of eligibility and targeting restrictions.[201]

In the two Facebook groups where discussions about the program took place, fifty of the 395 posts and comments that Human Rights Watch collected and analyzed between April and October 2022 indicate that the user had lost their benefit.[202] Many of these users did not understand why they had been dropped from the program, and some expressed confusion and despair at the loss of support.

The Bank has implicitly acknowledged that beneficiaries are sometimes caught off guard when they are dropped from the program. It told Human Rights Watch that “NAF is exploring methods of notifying beneficiaries that are subsequently excluded from Takaful of their changed status in a more timely manner, prior to the cessation of payments.”[203]

Inadequate Verification

NAF relies on home visits to verify the data submitted by applicants, correct errors, and adjust people’s ranking during the poverty targeting process as needed. These visits might surface, for example, that an applicant who claims to be living in a tent is in fact living in an apartment.[204] The NUR is updated accordingly, and the algorithm re-ranks the applicant based on the new input. In practice, this safeguard is not foolproof and can introduce human bias. Three families shared with Human Rights Watch that they were asked to provide stereotypical evidence of their poverty. During a virtual home visit, Khaled and Manal said that the interviewer asked them to focus their camera on “something that shows that you are really in need, something messy.”[205] Osama, 50, also said that the interviewer scrutinized the state of his furniture. “If it’s good, they will reject [your application].”[206] A woman in al-Qadsiya said that a NAF representative visited their house and took pictures, including the ceiling, “focusing on whether or not the house is good enough to be lived in.”[207]

NAF confirmed that it may “photograp[h] parts of the house” during these visits and that it considers providing households with additional support if the visit surfaces new evidence of the “severity of poverty and their standard of living.” It also said that these visits take place with the applicant’s consent.[208]  

NAF has also conceded that the applicant profiles generated by the NUR may be out of date if people do not update their information at the relevant government agencies (for example, if an applicant forgets to update the Ministry of Industry about the closure of their supermarket).[209] Outdated registry data can interfere with the accurate assessment of claims. A man in Burbaita, whose application for Takaful was denied, said government records showed that he owned a car, even though he had transferred ownership fifteen days earlier.[210] 

Appeals

Mechanisms for appealing and sharing grievances about automated decisions should be clear, transparent, and accessible, ensuring that people can quickly correct errors, address questions, and remedy unfairness. NAF told Human Rights Watch that people can “amend [their] data and request support through multiple channels, such as the online form available through NAF’s website, through phone at NAF’s support center, through the National Call Center, or by visiting any of NAF’s field branches (there are 42 main branches and 26 secondary branches across Jordan).”

However, many Takaful applicants are not aware that NAF provides them the right to appeal eligibility and targeting decisions, or how to submit an appeal. According to NAF and UNICEF’s 2020 survey of Takaful applicants, 75 percent were unaware that they had the right to share feedback and complaints about the registration process, and 67 percent did not know that NAF’s Takaful helpline existed.[211]

Human Rights Watch’s analysis of the two main Facebook groups focused on Takaful also indicates that many people find the appeals process confusing and unclear. Fifty-seven of the 395 posts and comments between April and October 2022 raised questions about the appeals process, such as whether they were eligible to appeal a determination, and how long they had to wait for a response. Some posts and comments also asked other users and group administrators to explain to them text messages they received from NAF after they submitted an appeal.

Several people Human Rights Watch interviewed recounted frustrating experiences submitting appeals. Misha’al, who lives in Dhiban, spent half a day trying to find out why his Takaful payments stopped, but to no avail.[212] His local MOSD office told him that he had to lodge an appeal at the NAF office in Madaba, about ninety minutes away from his home. When he arrived at the NAF office, it was overcrowded and dozens of people were waiting ahead of him in the queue. He gave up on trying to appeal and went home.

Atef tried to appeal the second time he was rejected from Takaful but received conflicting responses.[213] The NAF supervisor in his area told him that he had been rejected because the employee conducting home visits could not find his home. When he tried to call NAF, an employee told him it was “too late” to appeal. The NAF supervisor agreed to fax his appeal to the agency’s office in Amman, but he has not heard back. 

These challenges may stem in part from the lack of resources available to staff helplines and appeals mechanisms.  NAF said that it has assigned 107 employees to implement Takaful.[215] Even if all 107 employees were assigned to handle the 100,000 Takaful-related appeals it received between January and October 2022, each employee would have to process 934 appeals within this period.

According to WFP, it helped NAF establish a call center to handle appeals and feedback but this call center, which handles almost 30,000 calls monthly, is only staffed with 13 operators.[216] NAF told Human Rights Watch that it has “trained a team of employees capable of responding to all grievances, and the response rate reached 98%.” However, this response rate does not reflect whether staff are equipped to meaningfully address people’s concerns, questions, and feedback.[217]  

Digital Divide

Takaful’s registration process is digital by default. The government has established an online portal for people to submit and update their application, and applicants are notified whether they have qualified through the portal or via text message. But these digital processes are difficult or impossible to navigate for people who are unable to afford a smart phone or a computer, or who do not know how to use one. According to GSMA, 35 percent of Jordan’s population still lack access to mobile broadband, and 21 percent lack access to cell phones entirely.[218]

“Who needs the money? The people who really don’t know how to apply or don’t have internet and computer access,” said Aseel, who helped six other families in her community apply for Takaful.[219] A day laborer in Marka explained why it is unrealistic for him to apply online: “I need to have a touch telephone to apply for support. Where am I supposed to get one from? It costs 200 dinars ($282), who has 200 dinars to spend on a telephone?”[220]

The evidentiary requirements imposed by poverty targeting complicate the goal of making online registration inclusive. The application questions, some of which are critical to Takaful’s targeting accuracy, require a level of literacy that increases the challenge of applying online. According to a May 2020 survey of 450 Takaful applicants and beneficiaries conducted by UNICEF, while 57.4 percent of respondents rated the questionnaire “easy” or “very easy,” a significant minority appeared to struggle with at least some of the questions: 21 percent rated the difficulty level as “medium,” and 21.7 percent found it “hard” or “very hard.”[221]

A separate survey of 3,203 families commissioned by NAF and UNICEF in February and March 2020 also indicated geographical disparities in people’s comfort level with applying online: 74 percent of respondents in the south reported that they applied for the program at a registration center established by NAF, compared to 28 percent of respondents in the central region, where Amman is located.[222] Human Rights Watch also found that people living outside the city of Amman are more likely to need in-person support. Three of the eight residents of the capital city that discussed the online application process said they needed in-person support, compared to seven of eight living on the outskirts of Amman or in another governorate. 

Abdullah, who received Takaful in 2020 but was rejected when he applied again in 2021, was among the applicants who found that the “questions were very difficult” because he “wasn’t able to read very well.”[223] Atef, who lives in Zarqa and applied for Takaful twice, agreed, lamenting that “even Tawjihi [the national secondary education examination] is not that hard.”[224] The adjustments people frequently need to make to their income and expense figures to ensure that their application will be considered further complicate the process.

The time and effort needed for people to assemble supporting documents that verify their eligibility prolong the registration process, potentially delaying access to the benefit. A man in Burbaita who applied for Takaful through the NAF office in Tafilah said that he spent around six months gathering the necessary documents to complete his application, including a medical report demonstrating that his wife had been diagnosed with cirrhosis.[225] He began receiving Takaful payments only a year after he applied. Motasem was unable to enroll in Takaful-2 until he persuaded his landlord to formalize their arrangement in a rental agreement.[226] When he submitted the agreement, NAF asked him to list a house number. “There are no house numbers in rural areas. So, I had to put a random house number,” Motasem said.  

The government has made a significant effort to help people overcome these challenges, but this may still be insufficient to bridge the digital divide. NAF has 68 offices around the country serving a population of 11.15 million, or one office for every 169,000 people.[227] To enhance its coverage, the agency said it set up an additional 290 registration centers, and dispatched mobile registration bus units to border towns and extremely rural areas.[228] But the agency told Human Rights Watch that it is planning to wind down these services, as it believes that more people are becoming comfortable with applying online.[229]  

An activist In Burbaita confirmed that he had seen a NAF vehicle visiting tents in the area to register people for Takaful, but that these visits were infrequent: he first spotted the vehicle in 2021 and did not see it again until January 2023.[230] Moreover, he said that people in the area travel to the NAF office in Tafilah to follow up on their application and provide supporting documentation, which is around 25 km away. “People who don’t have cars have a hard time because they have to wait for anyone with a car to go with them,” he said. “If not, they have to walk to the street and find a car that is willing to drive them all the way up.”[231]

Difficulties applying for and managing the benefit online have spawned an informal economy of intermediaries – usually maktabehs and mobile phone shops that line town centers across the country – that help people apply and withdraw their benefit for a fee. “People who approach maktabehs are people who don’t know how to use technology,” said Eyad, an employee at a maktabeh in Baqaa’ camp.[232]

These costs add up. Since Takaful’s online portal was loading too slowly on Misha’al’s smart phone, he traveled from his home in Dhiban to a maktabeh in Madaba – around a 45-minute drive – to submit his application.[233] He paid 2 dinars ($2.82) for the bus to Madaba, and another 3 dinars for the maktabeh’s assistance. When his application was approved, Misha’al paid a mobile phone shop 3 dinars to withdraw his benefit, on top of the administrative fee of half a dinar ($0.70) levied by the e-wallet company.

These fees also reflect the financial burden imposed by the mobile payments process, which was established to increase financial inclusion.[234] Conventional banking services are out of reach for most low-income Jordanians: According to the Department of Statistics, only 4.7 percent of women and 24.4 percent of men in the lowest wealth quintiles own and use bank accounts.[235] To expand access to payments, the government and the World Bank have invested heavily in e-wallet applications.[236] People can open e-wallet accounts on their mobile phones with any major cell phone provider in Jordan, and use their e-wallet to pay bills, buy goods and services, and deposit or withdraw cash. NAF told Human Rights Watch that most beneficiaries access their benefit payments through e-wallets.[237]

According to NAF, 83 percent of beneficiaries reported cashing out their payments as opposed to paying bills using the e-wallet application or conducting online transfers.[238] NAF also told Human Rights Watch that all e-wallet payment service providers charge a standard rate of half a dinar per withdrawal.[239] One beneficiary added that her e-wallet provider capped the amount per withdrawal to 96 dinars ($135), requiring her to make and pay for additional withdrawals to access their entire benefit payment.[240]

The cost of accessing e-wallet payments is higher for people living far away from cash-out points. A May 2020 survey conducted by UNICEF found that the distance between people’s homes and cash-out points was the second most common challenge facing beneficiaries when they tried to withdraw their payments.[241] A NAF employee in Tafilah Governorate said that the nearest bank or cash out point in Jurf al-Darawish, for example, is in al-Husainya, which is 13 km away, or al-Hasa, 25 km away. “In these villages, the e-wallet arrangement isn’t very convenient for people because of lack of access to service providers. Most people here receive [the benefit] through their bank accounts,” the employee said.[242]  

Nayef, who lives in Jurf al-Darawish, was enrolled in Takaful in 2021. He traveled 25 km to al-Hasa, or 26 km to Tafilah, to withdraw his benefit payment at a mobile phone shop or a bank. Taking the public bus to Al-Hasa took him between one to two hours. If he visited Tafilah instead, he left at 7 in the morning and did not return till 2 in the afternoon.[243]  

A man in Burbaita said that, whenever he receives a text alert from NAF that his benefit payment is ready, he tries to arrange with drivers of public buses to make a stop at his neighborhood so that he can commute to the Cairo Amman Bank in al-Eis to withdraw the money. If the bus does not come, he tries to hitchhike or, as a last resort, rent a car for the day, which costs 15 dinars ($21).[244]

To withdraw her household’s payment, “Shahnaz” (not her real name), who lives in Dhiban, had to arrange and pay for a bus to a mobile phone shop in the town center, which charges 3 dinars ($4.23) to assist her with the withdrawal.[245] Suleiman paid 3 dinars ($4.23) for a bus ride from his home in Dhiban to a mobile phone shop in Madaba, which also charged 3 dinars ($4.23) to process the withdrawal.[246] Ibrahim, who lives in Zarqa, said he would pay 5 dinars ($7) to travel to a mobile phone shop to make withdrawals, and sometimes more during peak hour traffic.[247]

Abdullah was incredulous at the costs of withdrawing a benefit that barely made a dent on his family’s living expenses. He said:

When they give me three months [of the benefit], it was already little and for them to take fees? The most I ever received was 124 dinars ($175), and then when I want to withdraw, I got 121 dinars…What do 121 dinars ($171) do? Nothing. The poor always stay poor.[248]

In response to a request for comment, the World Bank acknowledged that, “in some cases…there are minimal costs for withdrawing the benefit due to the regulations of private sector agents,” and it is working with NAF and the Central Bank of Jordan to address this challenge. The Bank’s assessment, however, did not appear to account for the travel costs and mobile phone shop fees that some people incur to cash out their payments. NAF also told Human Rights Watch that it is working with the Ministry of Interior to crack down on maktabehs that charge excessive fees to help people submit their application. “Half a dinar to one dinar [$0.70 - $1.41] is okay, but beyond that, they shouldn’t [charge],” the NAF official said.[249]

Mistrust and Suspicion

Multiple studies have found that targeted cash transfer programs risk creating the perception that the government agencies administering them are being arbitrary or unfair, generating suspicion of people who benefit and undermining trust in the government.[250] The World Bank has admitted that Proxy Means Testing, the targeting method that Takaful is modeled after, “can seem mysterious and arbitrary to some.”[251] In some cases, researchers have documented the frustration and resentment people experience when they do not understand why others receive benefits that are denied to them, fueling a breakdown in social trust that has been linked to spikes in crime, conflict and violence.[252]

Fourteen people that Human Rights Watch interviewed expressed skepticism of Takaful’s targeting process. “There are people who have everything and good income and they get support, and there are people who cannot find something to eat, and they don’t get it. It’s all random,” said Walaa.[253] “There is a guy who receives 560 dinars ($790) in retirement, and I applied for him and he still received it [Takaful],” said Atef. “There are people who don’t deserve the support who get it.”[254] A longtime resident of Faisaliyah observed that people were discouraged from applying when many people they knew were turned down from the program. “They lost hope,” the resident said.[255]   

The program’s haphazard restrictions on income, expenses, and assets have heightened the impression that the government is out of touch with people’s needs and hardships. Eyad, the maktabeh employee, blamed this disconnect partly on the government’s reliance on automated targeting. “It should not be just these algorithms that are assessing people’s needs, they should be individually assessing cases,” he said.[256] Aseel echoed this sentiment, urging NAF to “put in more effort to review applications and see the conditions and circumstances that people live in.” She said:

If the government is in front of me right now, it would be useful for them to feel the life pressures we feel living in Jordan. Everyone has to juggle several jobs. I don’t have time to see my family anymore. I don’t have time to tell my mom how I’m doing because I am too busy. I don’t see my friends anymore because I’m too busy working…people are just so desperate.[257]

The struggle to understand Takaful’s selection of beneficiaries has fueled the perception that the program benefits people and families with political connections. Nine of the fourteen people told Human Rights Watch that they believed the targeting process was tainted by nepotism and corruption. This view was shared among people who were rejected from Takaful or had benefited, as well as people who helped others apply. “There’s a lot of inequality, said Shahnaz, whose family was finally enrolled in Takaful after submitting multiple applications. “People who have connections get money.”[258] A maktabeh owner in El-Eis said, “some people who come here [to his maktabeh] to apply are those who are certain that they will get it, after they have been referred by someone, like a relative in the government.”[259]

Human Rights Watch was unable to verify these allegations. But the perception of corruption among interviewees raises doubts that the government’s pivot to poverty targeting has made its social protection system more transparent and accountable – a key goal of its most recent social protection strategy.[260]

 

III. The Way Forward

Redesigning Jordan’s social assistance schemes to reflect the principles of universal social protection will alleviate or fix many of the problems identified in this report and establish lasting and meaningful protections for people’s rights, including the right to social security. Because eligibility requirements for universal programs are category-based (e.g., benefits are provided to all households with children under a certain age or all older persons above the statutory retirement age), they are easier to understand, apply for, and verify.[261] They also eliminate the need for subjective and error-prone assessments about people’s income and welfare. Transparent and objectively verifiable criteria may assuage mistrust and suspicion about the government’s management of social protection programs.

Implementing universal schemes would still require the government to conduct identity and documentation checks to verify that applicants are who they say they are, and other basic details such as their age and residence. The government should work with the Bank and its international partners to ensure that these processes maximize the inclusion of people who experience challenges obtaining identification, avoid relying on risky biometric technologies such as facial recognition, and incorporate robust privacy and data protection safeguards.[262] Designed in line with these standards, universal schemes would meaningfully reduce the evidentiary burden on people applying for benefits and sensitive data collected about them. 

The social protection consultancy Development Pathways has estimated that introducing a modest package of universal child, disability and old age benefits in Jordan will cost 280 million dinars (US$394.9 million), or less than 1 percent of GDP, during the first year of the program, and increase to slightly above 2 percent by 2035.[263]

During the transition to universal schemes, the government should work with the Bank and international partners to adopt interim measures to expand and improve Takaful, such as by expanding coverage to families of people with temporary Jordanian passports, non-citizen spouses, and children of Jordanian women. To enable meaningful public scrutiny of the program, it should conduct and make public regular audits of the targeting algorithm for as long as it is operational. These audits should, at a minimum, assess the rate at which the targeting algorithm excludes households from cash transfers in error, the reasons for such errors, and the corrective measures taken.

This report’s findings also highlight the need for the World Bank to take concrete steps toward realizing its vision of achieving universal social protection by 2030.[264] It should immediately cease developing or recommending projects to establish poverty targeted programs, including those to introduce or upgrade social registries and other poverty targeting infrastructure. Instead, it should focus its financing and technical assistance on information systems and other technologies required to support universal social protection schemes, such as population registries and vital statistics databases, ensuring that these are developed in line with privacy and data protection standards. For existing projects with poverty targeting objectives, it should work with borrower governments to revise these objectives and budget allocations to prioritize universal social protection.

 

IV. Legal Standards

International human rights law recognizes and protects the right to social security. Article 9 of the International Covenant on Economic, Social and Cultural Rights (ICESCR) and Article 22 of the Universal Declaration of Human Rights (UDHR) recognize the right of everyone to social security. Jordan ratified the ICESCR in 1975.[265]

Jordan has also ratified the ILO’s Social Security (Minimum Standards) Convention, 1952 (No. 102), which establishes minimum standards for the provision of social security benefits during life cycle contingencies such as old age, unemployment, and disability,[266] and the ILO Equality of Treatment (Social Security) Convention, 1962 (No. 118), under which it has accepted obligations of equality of treatment between nationals and non-nationals in specific branches of social security.[267]

The Convention on the Elimination of All Forms of Discrimination against Women (CEDAW), which Jordan ratified in 1992, requires states parties to ensure that women are treated equally with men, including with regard to their rights to social security, family benefits, and the right of rural women to benefit directly from social protection, and to modify laws, regulations and practices that discriminate against women.[268]

Jordan’s Constitution does not explicitly provide for the right to social security. Article 23 of the Constitution recognizes elements of that right, including the right of workers “supporting families” to “[s]pecial compensation” in the event of dismissal, illness, and work-related emergencies, and upon old age.[269] The National Aid Fund Law, No. 36 of 1986, outlines the responsibilities of NAF, including their responsibility to provide social assistance to eligible people or families.[270] However, the law limits the provision of support to Jordanian nationals, in conflict with the government’s obligation to ensure equal treatment of nationals and non-nationals.[271]   

As a UN specialized agency, the World Bank has an obligation to respect and promote human rights.[272] The Bank has itself recognized that “creating the conditions for the attainment of human rights” is a “central and irreducible goal” of its development agenda.[273] Although the Articles of Agreement, which govern the Bank’s activities, restrict its decision-making to “economic considerations” and prohibits political interference, the Bank has interpreted this mandate to include the protection and advancement of human rights through reducing poverty.[274] As a result, its lending decisions “are based on the quality of the project, and the effectiveness of the programs in reducing poverty.”[275]

Normative standards on the right to social security provide guidance to the Jordanian government on how it should implement its domestic commitments on social security in line with its international human rights obligations. These standards are also relevant to the World Bank’s role in advancing economic and social rights through as the structuring and implementation of social protection loans.

The Committee on Economic, Social and Cultural Rights (CESCR), the treaty body that monitors implementation of and issues authoritative interpretations of the ICESCR, has recognized that the right to social security is “of central importance in guaranteeing human dignity for all persons.”[276] It is an essential precondition of the right to an adequate standard of living and other rights, including the right to adequate food established under Article 11 of the ICESCR.[277] According to the Committee, states have a “core obligation” to “ensure the satisfaction of … minimum essential levels” of the right to social security. This obligation encompasses the duty to use “maximum available resources” that ensure “minimum essential level of benefits to all individuals and families that will enable them to acquire at least essential health care, basic shelter and housing, water and sanitation, foodstuffs, and the most basic forms of education.”[278]

The Committee recognizes both universal and targeted non-contributory schemes as possible ways to implement the right to social security.[279] But this report and multiple other studies indicate that providing a range of benefits to people during periods of their lives when their economic and social rights are particularly at risk, such as childbirth, unemployment, and old age, is necessary to discharge a state’s obligation to ensure “minimum essential level of benefits” to everyone.[280] Targeted programs remain a viable option for expanding social protection coverage once effective universal programs have been established.

In the absence of universal programs such as universal child and old age pension benefits, Takaful in its current form does not sufficiently meet Jordan’s obligations to ensure the right to social security. The World Bank’s approach to social protection financing, which heavily favors poverty targeting despite mounting evidence of its inadequacy and arbitrariness, is not only inconsistent with its mandate to pursue evidence-based measures for effectively reducing poverty, but also at odds with its obligation to respect and promote the right to social security.

The problems documented in this report also show that the use of algorithmic decision-making to facilitate poverty targeting poses heightened risks to the right to social security. Relying on flawed proxies of household need and inaccurate household data to identify beneficiaries – as Takaful does – fails to ensure that qualifying conditions for benefits are “reasonable, proportionate and transparent.”[281] The frequent upheaval of people’s benefits is at odds with the Committee’s recommendation that the “withdrawal, reduction, or suspension of benefits should be based on grounds that are reasonable, subject to due process, and provided for in national law.”[282] Misplaced confidence in the accuracy of poverty targeting algorithms can also lead to under-investment in appeals and grievance mechanisms, undermining people’s ability to challenge errors and unfairness in the targeting process.

In 2019, the UN Special Rapporteur on extreme poverty and human rights also highlighted other risks of automated benefits programs, warning that the “robotic application of rules” may unfairly withdraw or reduce benefits when people experience sudden changes in life circumstances, eliminate opportunities for “meaningful questioning or clarification” of eligibility determinations, and create a dehumanizing and “degrading” environment for beneficiaries.[283] Human Rights Watch has found that these problems exacerbate the mental health impacts of poverty, interfering with the right to the highest attainable standard of mental health established under Article 12(1) of the ICESCR.[284]

 

Acknowledgments

This report would not have been possible without the people who shared their experiences with Takaful and the daily toll of poverty and inequality in Jordan. Human Rights Watch is deeply grateful for their stories and insight.

This report was researched and written by Amos Toh, senior researcher in the Technology and Human Rights Division at Human Rights Watch. Hiba Zayadin, senior researcher in the Middle East and North Africa (MENA) Division, and Sarah Saadoun, senior researcher in the Economic Justice and Human Rights (EJRD) Division, provided extensive research and editorial guidance. Additional research was provided by Charbel Salloum, senior research assistant in the MENA Division, Sara Kayyali, former senior researcher in the MENA Division, Ekin Ürgen, associate in the Technology and Human Rights Division, and an independent researcher in Jordan.  

The report was edited by Frederike Kaltheuner and Anna Bacciarelli, director and program manager of the Technology and Human Rights Division, respectively; Fred Abrahams, associate program director; Adam Coogle, deputy director in the MENA Division; and Michael Bochenek, senior counsel, Children’s Rights Division. Lena Simet of the EJRD Division, Brian Root, Senior Quantitative Analyst, Rothna Begum of the Women’s Rights Division, Hye Jung Han of the Children’s Rights Division, and Samer Muscati of the Disability Rights Division, and various researchers in the MENA Division provided expert review. Chad Anderson, an independent expert on social protection, graciously reviewed a draft of the report.

This report was prepared for publication by associates for the Technology and Human Rights and Middle East and North Africa Divisions; Travis Carr, publications officer; Jose Martinez, senior coordinator; and Fitzroy Hepkins, administrative manager.