AI for social good: Supporting workers and businesses in LMICs

People in many countries face persistent challenges to employment. Even when jobs exist, specific opportunities can be hard to identify, difficult to get, and might not pay enough. Workers looking for jobs need employable skills and knowledge of where jobs are. Businesses and entrepreneurs, on the other hand, often struggle to grow, limiting how many jobs they can create. How can AI tools support workers and businesses to address these challenges?

Researchers in the J-PAL network have been using randomized evaluations to learn how to help people get jobs and how to help businesses grow for over two decades, drawing out insights to inform policies and programs from Brazil to sub-Saharan Africa and beyond. Researchers have also started testing how new AI interventions can support workers and firms to improve economic outcomes for people in low- and middle-income countries.

AI to support workers

AI has the potential to support workers and businesses in LMICs by both uncovering new evidence-informed solutions and making existing solutions more effective. 

One clear pathway is by increasing worker performance. Early research on equipping workers with AI tools tends to show positive impacts on productivity. 

For example, in ongoing research in the Czech Republic, small and mid-sized businesses whose workers received training on using AI in operations, marketing, customer management, product development, and financial management saw their profits grow compared to similar firms. Across Boston Consulting Group offices around the world, providing strategy consultants with access to a company-specific AI platform improved productivity and quality on knowledge-intensive tasks.  

Within businesses, this early evidence suggests that AI tools lead to larger gains for workers with lower skills or less experience. In the Boston Consulting Group example above, AI-driven improvements were larger for consultants who were performing below average before the study began. 

Similarly, a quasi-experimental study with customer support agents in the Philippines and the United States finds that providing agents with chatbot-based suggestions for client responses and links to internal documentation increased their performance. Improvements were largest among lower-skill agents, while higher-skilled agents receiving chatbot access experienced a small dip in work quality. 

In the US, giving software developers at large companies access to AI assistants increased the number of tasks they completed compared to developers without access. Less-experienced developers saw larger productivity gains. 

AI to connect workers and businesses

AI tools may help match workers to suitable jobs more efficiently by reducing search barriers for both job-seekers and employers. 

For example, in Ghana, an automated application review tool screened prospective teachers more effectively than “human-only” and “human-with-AI-assistance” alternatives, resulting in a higher rate of offers and hires at the interview stage. Three ongoing projects will evaluate tools designed to improve matching: a career counseling and recommendation tool in Kenya; a platform to help hiring managers assess applicants in Mexico; and a system using administrative data to recommend applicants for existing vacancies in France.

AI to support businesses

However, improving performance and matching only go so far if businesses aren’t growing and expanding their hiring. 

AI and other innovative tools can help banks and other lenders identify entrepreneurs and businesses with high growth potential, providing them with resources to grow and create jobs. For example, in Egypt, an AI-based credit scoring system that incorporated borrowers’ personality traits helped lenders allocate larger loans to client businesses that were more likely to grow with more capital, while flagging others who were unlikely to see their profits grow with the larger loan.

AI may help business owners and leaders make better decisions through customized advice, creating more jobs as they grow their existing enterprises or launch new ones. While giving traditional business training to entrepreneurs tends to have modest effects, evidence shows that more customized or consulting approaches can have greater impacts. Using AI may enable personalized and more effective support to help improve business practices. Among new sellers on a Chinese e-commerce site, for example, those receiving business training materials that an AI tool selected for them based on real-time performance data experienced higher revenues than sellers who were not offered training. 

However, emerging evidence also shows not all entrepreneurs may benefit from general purpose tools. For example, an early study from Kenya found that small business owners who were offered access to an “AI business assistant” chatbot did not improve business performance on average. But researchers also found that, among this group of entrepreneurs, businesses that were already more successful before being introduced to the AI tool increased their revenues and profits—yet the chatbot reduced revenues and profits for those who were less successful. Researchers suggest that the difference in outcomes stemmed from how the business owners picked and implemented specific pieces of AI advice.

Risks of exacerbating inequality

How will the role of workers change in economies shaped by AI? Researchers within and beyond the J-PAL network are assessing AI’s broader economic implications, including the possibility that AI’s widespread use could cause firms to hire fewer people in LMICs or lower their wages. For example, one study identified relative declines in employment for early-career workers in AI-exposed fields in the United States. As AI adoption spreads in LMICs, automation may reduce the need for workers with certain skills, with some groups benefiting while others lose out.

If not designed and monitored with care, AI-based matching and targeting systems to improve jobs and business growth may also exacerbate inequality. These AI tools identify hidden patterns in data that predict successful trainees or candidates. Depending on how the underlying model is trained, it may reinforce biases against marginalized groups. The existing evidence is mixed: in one case, giving reviewers an AI-generated score of applicants reduced discrimination against women. In another study that leveraged a Fortune 500 company’s historical hiring data, a tool that predicted candidate quality worsened discrimination unless it was carefully designed to consider target groups who are less represented in the data.

Careful design may allow AI to instead close gaps—helping underserved groups access credit, prepare for disasters, and maintain housing.  

Looking forward 

Further research should explore how to leverage AI tools for improved skills, productivity, and wellbeing, and avoid inadvertent harm. Priority questions include:

• Skills and training: How can AI tools improve training for job-seekers and workers? Can they make training on hard or soft skills more accessible and customized?
• Workforce organization: How can AI reduce discrimination and bias in the workplace and in hiring?
• Small and medium enterprise (SME) profitability and sustainability: What are the best models of AI interventions to increase the profitability, resilience, and sustainability of small enterprises?
• Welfare effects: What are the job displacement effects of AI, and how do they vary across LMIC sectors and labor markets?

AI adoption presents both opportunities and challenges for LMICs trying to support workers and businesses and foster economic growth that benefits more people. To chart this emerging frontier, J-PAL’s forthcoming AI Evidence Playbook will summarize what we know—and what we still need to learn—about AI’s role across sectors.

Beyond labor markets and business growth, AI offers enormous potential in many other sectors. The next blog in this series will focus on applications relevant for health outcomes—stay tuned!

Read other posts in this series.