By Musa Majad, MSc Development Studies student at the London School of Economics (LSE)
Image: creative commons licence – https://medium.com/center-for-effective-global-action/what-is-cash-good-for-b128accca139).
—
The first objective of the United Nations’ Sustainable Development Goals (SDGs) is to end extreme poverty by 2030, which remains a key aim for many developing countries, donors and international organisations. Social protections play an important role in this, with national governments implementing transfer programmes that can provide some financial support for the poorest households. The dominant method is targeting those individuals who require the most support, often through means-testing, and sometimes applying conditions related to health, education and/or employment. A growing trend is the use of algorithms and automation to make this process more efficient and cost-effective for the state. Jordan followed this path with the support of the World Bank.
In 2019, the Jordanian government launched a formalised cash transfer system called ‘Takaful.’ In 2022, ‘Takaful’ was combined with other cash transfer policies to create the United Cash Transfer (UCT), aiming to reduce income inequality by using algorithms to determine the most vulnerable groups and allocate resources accordingly. Although the policy has delivered some efficiency gains, including improved verification of people’s incomes, easier access to administrative records, and increased payment amounts, the digitisation of social protections has created new barriers, deterring some from applying. Many poor people are excluded from benefits due to the simplification of who counts as economically vulnerable.
Increased efficiency?
The World Bank (the largest lender to the UCT) praised Jordan’s significant shortening of the time required to deliver pandemic response packages (an earlier form of ‘Takaful’), crediting social registries and good use of technology. This is true. The National Aid Fund (NAF) distributed 3 months of cash transfers to 237,000 households, and in 2021, the UCT provided cash transfers to an estimated 62% of Jordan’s poor population. By unifying data into one National Unified Registry (NUR) and using algorithms to automate eligibility checks, there is less pressure on bureaucratic structures to assess and validate claims. However, this is not solely because of algorithms; the government’s decision to roll multiple cash transfer programmes into one also yielded efficiency gains, reduced administrative costs, and avoided the need for multiple rounds of targeting.
Whilst the adoption of algorithms along with streamlining the cash transfer process may have led to a cut in costs, this does not mean that there are no financial implications associated with algorithms. They are data-intensive, and there is a high cost to establishing and updating social registries; the latter is important to reduce the risk of errors and ensure accuracy.
Exclusion and Barriers
The digitisation of the cash transfer selection process creates additional barriers to applying for funds. The government established an online portal for people to submit applications for the UCT; however, this makes it difficult for those without access to digital infrastructure to apply. In Jordan, 35% of the population lacks access to mobile broadband and 21% lack access to mobile phones. The NAF states that 68 branches, 290 registration centres, and mobile stations for rural areas have been made available to counter this, but as the Human Rights Watch report on Jordan’s automated UCT compellingly shows, there have been many Jordanians who have been unable to access these services, and so have to pay a fee to submit the application through mobile phone shops. The report also highlights that there are costs associated with travelling to these sites, which are particularly difficult for those in rural areas.
Additionally, the application requires a lot of information to be provided and a level of literacy that may be challenging for some. For example, a man named Abdullah recounted how the questions were hard as he could not read very well, and it is difficult to provide monthly estimates, as this can vary greatly. Another individual stated that it took him 6 months to gather the correct documentation, including medical reports for his wife.
Furthermore, the use of algorithms has excluded people from accessing cash benefits because accurate data is not always provided, which the system relies on, and it ignores the nuances of economic vulnerability. The UCT consists of a 2-stage process, which starts with assessing applicants on a basic eligibility criterion before using an algorithm that includes 57 socio-economic indicators to rank applicants and determine who should receive the cash transfers. This method relies on “inaccurate and unreliable data about people’s finances. Its formula also flattens the economic complexity of people’s lives.” For example, one factor within the algorithm is that higher rates of consumption indicate a higher household income level; however, this is not always the case. In Amman, around 75% of low-to middle-income households surveyed lived in homes with bad insulation, requiring more energy and money to heat.
Conclusion and implications
The digitisation of cash transfers in Jordan has created new barriers and has been deeply exclusionary. However, most of these impacts are broadly inherent in any targeting system. Exclusion is central to targeting because the state is selecting who is eligible, which results in many deserving people not being able to access the benefits. Algorithms can potentially cause further harm because they are often presented as neutral, objective, and purely technical decision-makers. In reality, however, they are built on human choices: decisions about which data to include, how to define vulnerability, what indicators to prioritise, and where to set eligibility thresholds. When these systems are framed as impartial and data-driven, government decisions can appear rational, scientific, and even inevitable.
This framing can reduce political accountability. If someone is excluded from support, responsibility is easily shifted to ‘the system’ or ‘the algorithm,’ rather than to the policymakers who designed the criteria in the first place. The technical nature of algorithmic systems can also make them difficult to scrutinise, particularly for marginalised groups who may lack the resources or expertise to challenge decisions. As a result, what appears to be a neutral process can obscure underlying value judgments and policy choices, while limiting transparency and weakening avenues for appeal or redress.
Keywords: social protections, poverty, algorithms, AI, Jordan, World Bank, development
Bio: Musa is an MSc Development Studies student at the London School of Economics (LSE), interested in the role of social security and protection policies in reducing poverty and improving development outcomes.