AI data labelling: a pathway or peril to refugee self-reliance?

Executive summary

Humanitarian budgets are collapsing at the same moment that AI systems are generating major demand for high-quality human-labelled data. Considering flexibility and low entry requirements of AI data labelling, social enterprises have moved quickly to position refugees at this intersection, framing this digital work as a scalable, dignified pathway to economic self-reliance. Drawing on organisational documents, social enterprise impact reports, and first-hand worker testimonies, the policy brief examines whether data labelling constitutes a meaningful livelihood solution for displaced populations or whether it reproduces new forms of digital precarity within global AI supply chains.

The analysis focuses on four interrelated dynamics: the dual identity of social enterprises, which pitch their impact-oriented services to tech clients while selling the same work as empowerment to humanitarian donors; the self-reliance narrative promoted by social enterprises; and the lived realities behind this narrative, including both the positive aspects of giving refugees access to AI work and structural labour harms embedded in this labour.  While social enterprises are providing hope and some insulation for refugees entering the labelling market, they cannot overcome structural harms of the market, leaving refugees to bear the cost of that failure. This paper recommends that UNHCR publish binding operational guidelines for social enterprises, drawing on the Fairwork Framework, to ensure that AI-related digital work can support refugee livelihoods without entrenching the very exploitation it claims to address.

1. Context

Global emergencies are intensifying: the number of state-based armed conflicts reached a 70-year record in 2024 – numbering 61 (PRIO 2025). At the same time, global warming neared a 1.3 anthropogenic increase (World Weather Attribution 2025), and authoritarian repressions keep rising under the flag of a right-wing turn (Human Rights Watch 2024), leading to growing displacement all over the world. Despite this increasing polycrisis, in 2025 UNHCR’s budgets have diminished by 30% year to year and are expected to further fall by 15% in 2026 (Le Poidevin 2026). The greatest impact of these shortfalls has been felt on refugees themselves, whose numbers soared to over 36 million in 2025 (UNHCR n.d.).

By 2023, two-thirds of refugees lived in poverty, and 62% resided in host countries that severely restrict labour market access in practice, (UNHCR 2023, 1) despite the 1951 Refugee Convention’s formal guarantee of the right to work. The International Labour Organization (ILO) thus framed the gig economy as a “stepping stone to jobs and income”, (ILO 2025a, 19) while UNHCR highlights that web-based economic activities afford displaced individuals the vital opportunity to earn a living (Easton-Calabria, Hackl 2023, 7). In 2022 UNHCR and ILO launched an 18-month project in 8 countries called “Promotion, Inclusion and Protection of Refugees in the Gig Economy,” aiming to equip displaced populations with new skills and knowledge connected to digital hygiene and cyber risks (ILO n.d.).

Remote digital work like AI data labelling is thus uniquely attractive because it bypasses physical borders and discriminatory national labour laws, offering low barriers to entry and flexible, piece-rate work (Clark 2021, 770). For displaced populations, it theoretically offers a “virtual entry” into the global workforce, presenting an immediate income path for those routinely blocked from stable, local jobs (Altenried 2024, 1120). For tech companies, this transnational digital matching grants access to an “extremely scalable workforce” on a “pay-as-you-go” basis (De Stefano 2016, 4-5).

2. The hidden infrastructure of AI: what data labelling is

“AI isn’t magic; it’s a pyramid scheme of human labour” – Adio Dinika, a researcher at the Distributed AI Research Institute explained to The Guardian (Bansla 2025). AI market has been rapidly expanding with UNCTAD projecting for it to grow by 25 times – from 189 billion to 4.8 trillion USD between 2023 to 2033 (UNCTAD 2025). Microsoft estimated that one in six people are using generative AI worldwide with Global North leading the way with almost a quarter of the population being GenAI users (Microsoft 2026, 2). Its major branches like large language models (LLMs), developed for chatbots like ChatGPT, and computer vision systems, used to identify objects in self-driving cars, require massive amounts of high-quality data. If AI is trained on raw, unlabelled data, it absorbs inaccurate and different in format information, ruining its output (Jenkins 2024). To prevent a drop in quality, tech giants rely on millions of human data labellers, who are important for teaching AI systems to do tasks that are hard to describe clearly, such as finding emotions or bias in text (CWA Union 2025). These workers draw boxes around objects in images, transcribe audio, tag particular words and mistakes in AI-generated text and review datasets to ensure consistency and accuracy (SuperAnnotate 2025). Labelling is quite easy to complete and only requires language knowledge (often non-English languages on a native level), basic computer literacy and common sense (Kaneti 2025, 611).

3. The social enterprise model and its self-reliance pitch

Social enterprises entered the humanitarian sector in 2010-s, offering humanitarian agencies a mechanism to embrace market logics while maintaining moral legitimacy (Portales 2019, 57). By presenting themselves as hybrid entities that combine the efficiency and innovation of for-profit businesses with the passion and values of non-profits, they now act as a “bridge” between these two worlds, creating a “blended value” that appeals to both corporate clients and humanitarian donors (Battisti 2019, 140).

3.1 Hybrid structure and promise of social enterprises

If anyone visits the website of a social enterprise with AI data labelling programs, they find a platform created to speak to two entirely different worlds simultaneously. To tech giants, mostly based in the Global North, the enterprise pitches seamless, high-quality data labelling services at competitive rates, promising to build the future of AI. To humanitarian donors and foundations the pitch is more about resilience and showing refugees economically and mentally benefiting by entering the global digital economy.[i] To succeed in disseminating both pitches, social enterprises make their whole structure hybrid. The first half is a commercial agency that targets corporate tech contracts, showing they can successfully work in a private setting. The second one is a non-profit foundation that benefits from tax reductions and charity donors that give out grants specifically to NGOs. It is responsible for the social mission part and takes on digital literacy training and providing infrastructure, like laptops, to their refugee workers.[ii]

By utilizing these two faces, the social enterprise positions itself as a cushion between the client AI data labelling platform and refugees. However, the main website page of the corporate agency of EmpowerAI does not make major emphasis on datasets being produced by refugees: the visitor would need to scroll and click to learn about this.[iii] The reason behind it is that the social impact pitch is deemed unattractive to corporate clients that are “accustomed to no compensation floor and thriving on a rampant race to the bottom”. When one enterprise tried to establish stricter rules on fair compensation, they faced so many difficulties in getting new clients that the hired consultants recommended it to “tone it [social impact narrative] down” (Kaneti 2025, 618, 620).

3.2 Selling “self-reliance” to the market and the worker

The narrative of self-reliance has been present in refugee support field, being grounded by UNHCR in their 2005 “Handbook of Self-Reliance”, aiming to empower refugees “to meet essential needs <…> in a sustainable manner and with dignity” (UNHCR 2005, 1) It has later become a running theme in most social enterprise reports under an umbrella narrative of “work instead of aid” (Sama 2022, 12). They set an inspiring tone by making the title “Empowering Change in Conflict-Affected Communities” and featuring “inspiring case studies” that “captured positive changes our initiatives brought to communities in need” (Humans in the Loop 2024, 1). Social enterprise operations are thus driven by a goal of “economic self-sufficiency of disadvantaged populations,” while a longer-term vision is “refugees [being] able to overcome economic & social barriers and become pioneers in the tech sector” (Subul 2024, 10-11). Here social enterprises reframe refugees from “tech user[s] to tech empowered,” (Techfugees n.d.) actively promoting a narrative change of the role refugees can play in digital economy. Considering the ambition, the effects are supposedly going beyond giving short-term solutions, instead aspiring to provide “financial independence and career development”, serving as an opportunity to “re-start their careers” (Humans in the Loop 2024, 3). However, analysed social impact reports provide no longitudinal data on whether participants transition to stable employment. Instead, it highlights “success stories” of workers who leveraged their annotation gigs with the only “ripple effect” being that participating refugees started “investing more into their families, education and communities after securing digital work” (Sama 2022, 24).

4. The realities of AI data labelling programs for refugees

To understand how social enterprises succeed or fail in fulfilling their promise of self-reliance to refugees, it is necessary to consider both the bright and the dark side of AI data labelling programs.

4.1 The bright(-er) side of data labelling

Hope given through market access

As it was mentioned above, in many host countries displaced people routinely confront obstacles, including the denial of work permits, unrecognized educational credentials, language barriers, widespread discrimination and geographic isolation in camps or peripheral urban areas (Altenried 2024, 1120). Hence, AI data labelling can give much-needed accessibility to income that does not necessarily require a 9-5 office presence. It makes it accessible to refugees caring for families, those with physical or mental disabilities or facing religious restrictions on mixed-gender workplaces (CWA Union 2025). Another benefit is transportability of data labelling work as regardless of refugee’s next destination, it will still be available online, which is a ray of stability important for ever-changing livelihood of displaced people (De Stefano 2015, 5). Moreover, from the psychological perspective, earning money independent of social allowance can in itself be empowering with refugees feeling “on top of the world” when they first got it (Berhane Gebrekidan 2024, 7). A job in a prospective field of AI serves as motivation and a point of pride for people who lost their homes and had to restart their lives in a new country.

Social enterprise mediation

Importantly, social enterprise model offers an additional insulation for refugees that would otherwise be thrown into the ocean of data labelling work without any safety nets. Refugees are usually unaware about how to register and engage on digital platforms or do not even possess the needed infrastructure (Kaneti 2025, 616). Most social enterprises are at least trying to negotiate better conditions for their workers, for instance, setting fairer pricing models, arranging clearer task scopes and deadlines, and implementing specific ethical guidelines (Ibid, 615). They are also taking on most of the administrative burden: finding clients and negotiating connection with refugees in spite of their documentary situation. This support is especially vital for refugees who often lack official documents or bank accounts due to host state restrictions and other bureaucratic obstacles. So, social enterprises focus on mediating bank transfers from the clients, which refugees themselves call “very convenient”. While it might take some time: “We sometimes do not receive our work allowances until after two months” (Al-Hammada 2024, 11), this is still much better than not receiving anything at all. In case of office work in host countries, social enterprises can also offer the space and infrastructure (computers, Wi-Fi etc.) needed for data labelling. Finally, as mentioned above, they are conducting trainings that go beyond data labelling, important for those without education, computer or soft skills, which can thus improve refugees’ employability and future work opportunities.[iv]

4.2 The dark side of AI data labelling for refugees

Non-transparency and communication hurdles

When a refugee starts work on the first project, the enterprise gives instructions from the client. They are individual for each project and can be 30-40 pages which workers must learn to avoid mistakes. However, workers have no direct line to the client, so any unclear cases are communicated back through a long chain: enterprise team manager to senior staff to client (Ibid, 9). This layered communication frequently produces misalignment between what the written policy says, what the trainer explained, and what the client actually wanted. The consequences then fall on refugees. When a completed and submitted project fails to meet client expectations, workers are expected to redo the work under revised guidelines, importantly doing it for free. As one Syrian refugee data worker in Lebanon put it: “Why do we have to show flexibility and re-implement the work even though it is not our fault, as the instructions were not clear from the beginning? Is it fair to work twice on a project for one wage?” (Ibid, 10).

Non-disclosure agreements (NDAs) remain an issue for most data labellers as well, as they are forced to sign these documents to formally protect data of the client company. NDAs function as instruments of forced silence and intimidation, making it punishable to form unions to try improving working conditions of labellers. While some social enterprises are trying to take on more advocacy functions and negotiate the unionization of workers, it remains extremely difficult due to clients justifying NDAs as necessary to protect their intellectual property and corporate confidentiality (Equidem 2025, 42).

Algorithmic control and NDAs

The mental strain of meeting targets is serious – in most projects the lowest quality threshold stands at 90-97% (al-Hammada 2024, 7). A score below this could trigger warnings, lost incentives or even placement on a “performance improvement plan,” which workers described less as support than as a formal corridor toward dismissal (Berhane Gebrekidan 2024). To make matters worse, workers might have to face triggering text or pictures, often connected to their war-torn homelands, as they are often hired to label in their native language. While most times social enterprises are trying to avoid such projects, they are often left unaware about concrete tasks and end goals of their clients (Kaneti 2025, 617). “I hope this isn’t for weapons” is a valid concern among labellers, however the non-transparency of end client makes it impossible to be certain about the usage of labelled data. It might indeed be for weapons (Miceli 2024).

The time pressure remains massive as well. If the worker takes too much time on a task, their rating will fall, resulting in less or lower-paid tasks in the future (Hao, Hernandez 2022). It leaves them in a vicious cycle of trying to complete tasks fast but retain the highest quality. One data worker summarized this pressure: “How can you read this essay, grade it accurately, and then provide feedback in five minutes? The time per task was not adding up, and then the timer put on pressure” (CWA Union 2025). This is usually the point that clients of social enterprises are not willing to negotiate on, as famously said: “time is money,” and the labelled datasets are expected by the company on time, regardless of how much work and rework it actually requires (al-Hammada 2024, 7).

Psychological effects

In an attempt to mitigate negative effects on workers, most social enterprises introduce wellness breaks and support sessions with a psychologist. Despite flexibility being one of the appeals of this work, in reality, those working in the office have a standard 8-hour day and highly restricted wellness breaks. Workers in similar facilities reported having just “a 20-minute wellness break twice a week” and noted that they “have to beg for a wellness break on the group chat for it to be granted” (Berhane Gebrekidan 2024, 14). Psychological support is also unprofessional. Instead of receiving help, workers state: “I always go to the counsellors and come out more triggered and more crazy” because some counsellors “think you are faking it or trying to manipulate them so you can get a leave permit when you cry” (Ibid, 32-33).

Economic precarity and dependence in the absence of choice

When there is “only one project every few months <…> Many annotators feel compelled to accept any available work” (al-Hammada 2024, 32-33) The task can also come in the middle of the night or when they are unavailable. Some workers resort to using a “browser extension that sounds an alarm when a new task appears” and keep it “on loud even when [they] sleep, to wake [them] up in the middle of the night” (Hao, Hernandez 2022). However, this additional payment may be fundamental for the whole family, so missing or declining it is not an option. Yet this work is still poorly paid. Labelling objects on 10,000 images over several days is compensated with just 140 euros – an amount that barely covers ten days of food in a country like Lebanon (al-Hammada 2024, 9). In a year the “salary”, already positively negotiated by one data labelling social enterprise, amounted to just a bit above 2000 euro (Kaneti 2025, 615), which is far from enough for survival of even one person.

Non-transferable skills

Finally, the long-term effects of trainings conducted by social enterprises for refugees might not be as useful as promoted. Instead of acquiring high-level coding or web development skills that foster genuine upward mobility, the social enterprise only taps into these spheres on a surface level, primarily focusing on training refugees to carry out routine, low-skill microwork. The trainings range from turning on a computer to English language or soft skills, however they rarely go into developing any of these skills higher than pre-intermediate level (Humans in the Loop 2025, 15). While this is important for some displaced data labellers, many other are not only literate but also have a higher education (al-Hammada 2024, 8). Transferability of these skills thus remains doubtful, trapping refugees in the cage of labelling microtasks for specific clients who have narrow rules and skills needed for usually short-term projects.

5. International policy landscape: Emerging good practices

The regulatory vacuum that historically enabled these harms is beginning to fill. The following instruments constitute a growing toolkit that UNHCR can leverage in its frameworks for refugee platform work. These examples were chosen to represent relevant policies in different regions and are not meant to cover the entire regulatory landscape of platform work and are listed by scale of coverage.

Considering the existing legislation, it is also important for UNHCR to take into account special vulnerabilities of refugees when developing own policy for work conditions of social enterprise programmes in data labelling, recommendation for which will be presented in the last part of this paper.

6. Policy recommendations

1. Prioritize integrated pathways to sustainable livelihoods

To strengthen the long-term impact of digital livelihood programs, UNHCR should move beyond short-term output metrics and tie recommendations to longitudinal evidence that refugee participants progress toward stable, formal employment. Innovation grants can be impactful if directed toward programs that build genuinely transferable skills, such as advanced coding, ensuring that refugees, including those with higher education, are supported toward meaningful economic mobility instead of entry-level microwork.

2. Adopt minimum labour standards via the Fairwork Framework

Current UN initiatives have made valuable progress on digital literacy and cyber risk awareness. To build on this foundation and ensure that digital livelihood programs deliver on their self-reliance promise, UNHCR should formally endorse the Fairwork Framework (Fairwork n.d.) as a baseline for recommendations to social enterprises across five dimensions:

• Fair pay

UNHCR should work with social enterprises towards replacing piece-rate structures that produce poverty-level earnings with living wage thresholds adapted to host-country cost of living. Enterprises should be required to publish transparent, disaggregated income data as a condition of humanitarian endorsement, replacing the current practice of curated success stories with longitudinal evidence that the work actually generates economic self-reliance.

• Fair conditions

Because employer-provided mental health support has been documented as often inadequate and even harmful for workers, UNHCR should mandate that trauma-informed mental health care for refugee data laborers be funded and managed entirely independently of employer control with particular attention to the double exposure risk for refugees labelling content from their own countries of origin.

• Fair contracts

UNHCR is encouraged to work with social enterprises to review the use of NDAs in refugee data labelling contexts, exploring whether worker protections and client confidentiality can be balanced through alternative grievance mechanisms. Proactive disclosure of client identities and dataset purposes, where feasible, would help workers make informed decisions about the work they undertake.

• Fair management

UNHCR should call on social enterprises to introduce meaningful human oversight of algorithmic performance systems, including the 90-97% accuracy thresholds enforced under time pressure, and to ensure that no worker faces income reduction or dismissal risk for raising concerns about task content. Task timelines should be set in consultation with workers, instead of dictated by client delivery schedules alone in order to improve both working conditions and output quality.

• Fair representation

UNHCR should require social enterprises to build worker representation into their governance structures, viewing them as central voices with real influence over pricing negotiations, task standards and client selection. These should include, among others, regular anonymous worker surveys, introduction of Beneficiary Advisory boards which can receive concerns from workers, as well as allowing to form unions.

7. Literature list

1. Al-Hammada, R. (2024). “If I Had Another Job, I Would Not Accept Data Annotation Tasks.”
2. Altenried, M. (2024). Mobile workers, contingent labour: Migration, the gig economy and the multiplication of labour. Economy and Space.
3. Bansla, V. (2025). How thousands of ‘overworked, underpaid’ humans train Google’s AI to seem smart. Internet source. URL: https://www.theguardian.com/technology/2025/sep/11/google-gemini-ai-training-humans
4. Battisti, S. (2019). Digital Social Entrepreneurs as Bridges in Public-Private Partnerships. Journal of Social Entrepreneurship.
5. Berhane Gebrekidan, F. (2024). Content moderation: The harrowing, traumatizing job that left many African data workers with mental health issues and drug dependency. Data Workers Inquiry.
6. Clark, T. (2021). The Gig Is Up: An Analysis of the Gig-Economy and an Outdated Worker Classification System in Need of Reform, 19 Seattle J. Soc. Just.
7. CWA Union. (2025). Ghost Workers in the AI Machine: U.S. Data Workers Speak Out About Big Tech’s Exploitation. Internet source. URL: https://cwa-union.org/ghost-workers-ai-machine#workers-value
8. De Stefano, V. (2015). The Rise of the “Just-in-Time Workforce”: On-Demand Work, Crowdwork, and Labor Protection in the “Gig Economy”.
9. De Stefano, V. (2016). The Rise of the “Just-in-Time Workforce”: On-Demand Work, Crowdwork, and Labor Protection in the “Gig Economy”. Comparative labor law.
10. Department of Employment and Workplace Relations. (2024). Fair Work (Digital Labour Platform Deactivation Code) Instrument 2024.
11. Easton-Calabria, E., & Hackl, A. (2023). Refugees in the digital economy: The future of work among the forcibly displaced. Journal of Humanitarian Affairs, 4(3).
12. Equidem. (2025). Scroll. Click. Suffer. The Hidden Human Cost of Content Moderation and Data Labelling.
13. EUR-LEX. (2024). Directive (EU) 2024/2831 of the European Parliament and of the Council of 23 October 2024 on improving working conditions in platform work.
14. Fairwork. Principles. URL: https://fair.work/en/fw/principles/
15. Hao, K., Hernandez, A. (2022). How the AI industry profits from catastrophe. Internet source. URL: https://www.technologyreview.com/2022/04/20/1050392/ai-industry-appen-scale-data-labels/
16. Human Rights Watch. (2024). World report 2025. New York.
17. Humans in the Loop. (2024). Impact Report 2023.
18. Humans in the Loop. (2025). Impact Report 2024.
19. ILO. (2024). Fair Work (Digital Labour Platform Deactivation Code) Instrument 2024 under the Fair Work Act. Internet source. URL: https://digitallabour.ilo.org/legislation/fair-work-digital-labour-platform-deactivation-code-instrument-2024-under-fair-work-act
20. ILO. (2025a). Exploring the gig economy: Challenges and opportunities A self-guided resource.
21. ILO. (2025b). The Bihar Platform Based Gig Workers (Registration, Safety and Welfare) Act, 2025. Internet source. URL: https://digitallabour.ilo.org/legislation/bihar-platform-based-gig-workers-registration-safety-and-welfare-act-2025.
22. ILO. (2025c). Gig Workers Act 2025. Internet source. URL: https://digitallabour.ilo.org/legislation/gig-workers-act-2025
23. ILO (2026). Convention concerning decent work in the platform economy. ILC.114/Convention No. 193. URL: https://www.ilo.org/sites/default/files/2026-06/ILC114-Instrument%20C.193-EN.pdf.
24. ILO. Promotion, inclusion and protection of refugees and host communities in the digital and gig economy. Internet source. URL: https://www.ilo.org/projects-and-partnerships/projects/promotion-inclusion-and-protection-refugees-and-host-communities-digital.
25. Jenkins, T. (2024). Garbage in Garbage out. Internet source. URL: https://www.applogicnetworks.com/blog/garbage-in-garbage-out
26. Kaneti, M. (2025). Social Enterprise Solutions for Migrants, Microwork, and AI Ethics: The Case of HiTL. Human Service Organizations: Management, Leadership & Governance, 49(5), 609-627.
27. Le Poidevin, O. (2026). Exclusive: Cash-strapped UN refugee agency to cut more jobs, even as crises mount. Internet source. URL: https://www.reuters.com/world/europe/cash-strapped-un-refugee-agency-cut-more-jobs-even-crises-mount-2026-05-18/
28. Malaysia Parliament. (2025). Rang Undang-undang Pekerja Gig 2025. D.R. 27/2025.
29. Miceli, M. (2024). “I hope this isn’t for weapons.” How Syrian data workers train AI. Internet source. URL: https://untoldmag.org/i-hope-this-isnt-for-weapons-how-syrian-data-workers-train-ai/
30. Microsoft. (2026). AI Diffusion Report. Global AI Adoption in 2025—A Widening Digital Divide.
31. Portales, L. (2019). Basics, Characteristics, and Differences of Social Entrepreneurship.
32. PRIO (2025). New data shows conflict at historic high as U.S. signals retreat from world stage. Internet source. URL: https://www.prio.org/news/3616
33. PRS. (2025). The Bihar Platform Based Gig Workers (Registration, Safety and Welfare) Act. No 8.
34. Sama. (2022). Environmental & Social Impact Report 2022.
35. Subul. (2024). Annual Impact Report 2024.
36. SuperAnnotate (2025). What is data labeling? The ultimate guide. Internet source. URL: https://www.superannotate.com/blog/guide-to-data-labelling
37. Techfugees. From tech user to tech empowered (#Empowerment). Internet source. URL: https://www.notion.so/covidrefugees/2-From-tech-user-to-tech-empowered-Empowerment-11392e4039c8492d8cef375417c01166
38. UNCTAD. (2025). 2025 Technology and Innovation report. Internet source. URL: https://unctad.org/publication/technology-and-innovation-report-2025
39. UNHCR. (2005). Handbook for Self-Reliance. Geneva.
40. UNHCR. (2023). Refugees’ Access to Jobs and Financial Services. Background Guide. Challenge Topic #3.
41. UNHCR. Refugee Data Finder. Internet source. URL: http://unhcr.org/refugee-statistics.
42. World Weather Attribution. (2025). Unequal evidence and impacts, limits to adaptation: Extreme weather in 2025. Internet source. URL: https://www.worldweatherattribution.org/unequal-evidence-and-impacts-limits-to-adaptation-extreme-weather-in-2025/

Endnotes

[i] The difference is mostly visible in these two websites of the same social enterprise: The NGO part: https://humansintheloop.org/impact/foundation/, the corporate part: https://humansintheloop.org/

[ii] See, for example, two entities of this social enterprise:  Na’amal. URL: https://naamal.org/, Na’amal Agency. URL: https://agency.naamal.org/.

[iii] See the website: HITL. URL: https://humansintheloop.org/.

[iv] Look, for instance at the following social enterprise sources: Techfugees. Inclusion. URL: https://techfugees.com/inclusion/,  Medium. Na’amal, DOT, and The Conrad N. Hilton Foundation Launch Follow-Up to Successful Digital Skills Programme in Ethiopia. Internet source. URL: https://na3amal.medium.com/naamal-dot-and-the-conrad-n-9c2268fd56b2