A study of Indonesia's Kartu Prakerja programme finds that on-demand cash and training assistance can significantly boost self-employment and income among those who genuinely receive it. However, the programme's flexible online design enabled third-party 'jockey' agents to intercept benefits, meaning up to 70% of recorded recipients may not have received the programme in full – highlighting a key trade-off between accessibility and verification in safety net design.
Administering unemployment insurance programmes in a developing country is hard (Gerard et al. 2025). In developed countries, unemployment insurance is usually tied to unemployment spells, but with the high levels of informal employment typically seen in developing countries (Ulyssea 2020), it is a challenge to both verify an initial job separation and whether someone has found a new job. Given these challenges, while many developing countries have developed cash assistance programmes to combat long-run poverty, there are few programmes that provide short-run assistance to households – across the income distribution – to address short-run employment shocks.
An innovative approach to a challenging problem
We studied an innovative, on-demand assistance programme that aimed to address this policy gap. Launched in April 2020, the Kartu Prakerja (‘pre-employment card’) programme offered Indonesians an electronic job training credit of Rp 1,000,000 (US$70) that could be used on the programme’s online training marketplace (Hanna, Olken, Sumarto, Maulana, Alatas, and Satriawan forthcoming).[1] Upon completing the training, beneficiaries received transfers equal to IDR 2,400,000 ($170) over four months.
The programme design aimed to address the challenges of providing timely, short-run cash assistance, combined with job training, in a setting with high levels of informality. The programme was flexible, such that anyone could apply online as long as they were not already on social assistance or a student, and enrolment was simple. Moreover, people could apply when they needed it, conditional on not receiving the programme before (for example, in its first year of 2020, people could apply in any of the 11 application windows).
Importantly, as anyone can apply, assistance was not conditioned on employment status and beneficiaries received all four cash payments, regardless of their work status. However, the application system provided a mechanism that could potentially screen for those who really needed it: upon application, there was a lottery for selection. If one was not selected, one could decide whether it was worth applying again.
Finally, the programme delivered assistance quickly to address people’s needs in real time: beneficiaries quickly received the training credit that they could easily use online. Upon completing the training, beneficiaries received their payments, either through a premium e-wallet or bank account. Most people, around 87% in our surveys, received the payment through e-wallet accounts.
The programme scale was enormous. In its first two years alone, more than 23 million people (13% of Indonesian adults) applied over 22 enrolment windows, with over 11 million people winning a lottery and enrolling.
We measured the receipt and impact of the programme using all 20 randomised application windows from the programme’s first two years, along with a combination of administrative data, national sample surveys, and our own online survey of applicants.
Who received assistance?
We first ask: who received the programme? Did the programme actually reach those whom it was intended to reach?
We matched data from Indonesia’s nationally representative household surveys (SUSENAS), which asked households whether they had received the programme, with administrative data on programme recipients according to the government’s internal records. We found a large discrepancy between these two datasets: only about 30% of individuals that are recorded as having received the programme in the administrative data actually self-report that they did so when asked about the programme in the national household surveys.
What drives this discrepancy?
We explore several hypotheses. First, we ask if people are forgetting they received the programmes. It would be surprising if people forgot, given that the registration process required going through numerous ‘Prakerja’ web pages. Also, we do not see the same level of under-reporting in other government programmes, including another programme launched around the same time (Sumarto et al. 2026). In fact, Prakerja beneficiaries would need to be more than 2.5 times more likely to forget than beneficiaries of other government programmes, which seems implausible.
Another hypothesis is that someone else applied in the beneficiary’s name, either with or without the beneficiary’s permission. The demographics suggest that this is possible. Among beneficiaries in the administrative data, those who report that they did not receive the programme in the surveys tend to be older, more rural, less educated, less likely to use the internet, i.e. people who may have had challenges with the online application. And yet, ironically, they also appear to complete the training and receive the cash transfer faster in the administrative data, despite being less internet literate.
One explanation consistent with these patterns was the rise of agents – or ‘jockeys’ – that emerged to help people apply (Putri 2020). In some cases, these agents deducted a fee to ‘apply’ for an individual; in other cases, they may not have told people that they had won the programme.
The ease of the system – e.g. online application, use of e-wallet accounts – may have facilitated the jockey’s emergence. In the first three windows, the government used a third-party facial matching programme to compare the faces of the selfies uploaded with those on the identity cards and then also compared them with administrative records from the Ministry of Home Affairs. For these windows, the discrepancy between administrative data and independent household survey data was low. In May 2020, however, Indonesia’s anti-corruption agency (KPK, Komisi Pemberantasan Korupsi) provided an opinion that additional costs to perform facial recognition were unnecessary and instructed the programme to discontinue the checks, at which point the larger rates of discrepancies occurred.
Did the programme improve employment and consumption outcomes?
Using the national sample surveys, we study 20 lotteries from the first two years to estimate the programme impact. We find no impact on per capita consumption, even among households surveyed within the period when the cash transfers were active. Moreover, we observe no changes in employment or earnings.
It is possible that the programme had meaningful impacts for those that received it, but because – as described above – many households who received the programme according to the administrative data did not actually receive it (or receive it in full) the effects are muted on net. Our online survey, conducted with applicants in 2022, suggests that this is the case.
Our online survey is a selected group, i.e. people with real and active phone numbers and who responded twice to our surveys. Unlike in the nationally representative household surveys, the online survey recipients do indeed report receiving the programme if they win: about 92% of those who receive Prakerja in the administrative data self-report receiving it in our survey.
For this selected group, we find large effects: the programme led to a 19% increase in being self-employed or business owners. We see no change in overall employment, suggesting a shift in the type of employment. Total monthly income from all jobs increased by about IDR 215,000 (19%), suggesting the shift in jobs may have been profitable. Moreover, we observe increases in net asset purchases and a reduction in self-reported depression, suggesting an increase in overall household well-being.
Lessons for policy
Kartu Prakerja was designed as a safety net programme, but was launched during the COVID-19 crisis. In economic crises, programmes often face trade-offs between getting the programme to those who may desperately need it now versus having all the safeguards needed to minimise the chance that the ‘wrong’ people get it. Safety nets in the US during COVID-19 also faced similar trade-offs (Khetan et al. 2024, Griffin et al. 2023, Aman-Rana et al. 2023).
In the case we study, our results suggest that this type of on-demand assistance, consisting of cash assistance and online job training – can be effective. But programme flexibility – on-demand internet-based application, ease of payments via e-money, and fewer of the cumbersome administrative checks that impede programme up-take – may have led to 70% of people not fully receiving the programme. Thus, it is important to strike the right balance between the verification to ensure that only authentic beneficiaries are receiving assistance and the flexibility needed so that people in need can actually access the programme. Notably, the programme did subsequently reintroduce facial recognition and live selfie verification from 2022 onwards, suggesting that this lesson has since been acted upon.
References
Aman-Rana, S, D W Gingerich, and S Sukhtankar (2023), "Screen now, save later? The trade-off between administrative ordeals and fraud," Unpublished manuscript.
Gerard, F, G Gonzaga, and J Naritomi (2025), "Job displacement insurance in developing countries," in R Hanna and B A Olken (eds), The Handbook of Social Protection: Evidence and New Directions for Low and Middle Income Countries, MIT Press.
Griffin, J M, S Kruger, and P Mahajan (2023), "Did FinTech lenders facilitate PPP fraud?" Journal of Finance, 78(3): 1777–1827.
Hanna, R, B A Olken, S Sumarto, A Maulana, V Alatas, and E Satriawan (forthcoming), "On-demand assistance: Experimental evidence from Indonesia," American Economic Journal: Economic Policy.
Khetan, U, J Leder-Luis, J Wang, and Y Zhou (2024), "Unemployment insurance fraud in the debit card market," Unpublished manuscript.
Putri, C A (2020), "PMO Prakerja: 'Joki' daftar Kartu Prakerja bukan kriminal," CNBC Indonesia.
Sumarto, S, E Satriawan, B Olken, A Banerjee, A Tohari, V Alatas, and R Hanna (2026), "Community targeting at scale," Unpublished manuscript.
Ulyssea, G (2020), "Informality: Causes and consequences for development," Annual Review of Economics, 12(1): 525–46.