Site icon AfricaBusiness.com

Africa’s Informal Businesses Need an AI Dividend They Can Count

By Dr. Gleb Tsipursky

A trader receives an automated suggestion to buy more stock. An artisan gets a polished marketing message from a phone-based assistant. A small farmer receives a recommendation about crop treatment. These tools can help, but the question that matters is what happens after the recommendation. Did the owner earn more after fees? Did a mistake cost a day’s revenue? Could the owner challenge bad advice?

African governments and technology companies are investing in AI skills and services, with promises of broader access and greater productivity. Those promises need a practical economic test in stalls, workshops, farms and informal service businesses. Operators should be able to see whether a tool earns its keep after fees, errors and the time they spend learning to use it.

Treat informal operators as partners in deciding which problems AI should solve. Measure their net gains alongside indicators of use such as downloads and engagement.

What the Kenyan pilots tell us

In September 2026, the International Labour Organization described SABY, a WhatsApp-based business adviser that builds on its Start and Improve Your Business materials. The Kenya-centred pilot also attracted entrepreneurs elsewhere in Africa without a targeted rollout in those markets. By July 2026, according to the ILO, 838 entrepreneurs had used it and exchanged more than 64,000 messages. The ILO describes users asking about stock records, staffing and growth decisions. [1]

Those figures show how participants engaged with the tool. Measuring whether entrepreneurs increased profits or avoided costly mistakes would require separate evidence. The practical design lesson is to put assistance in a familiar channel, let people communicate in ordinary language and voice notes, and reinforce human business-development training.

A second Kenya-centred example comes from agriculture. In February 2025, Safaricom and Opportunity International announced FarmerAI, a chatbot pilot with an original 2025 enrolment target of 800 to 1,000 Kenyan potato farmers reached through SMS and WhatsApp. The service was designed to offer locally relevant information about weather, fertiliser, pests and market prices. The announcement described the pilot’s design and enrolment target. Actual participation, yields and incomes would need to be measured separately. [2]

The practical conditions differ between a market stall and a smallholding. A trader may buy stock with borrowed money and absorb the entire cost of an inaccurate demand forecast. An artisan may rely on referrals and have limited time to check AI-written customer terms. A farmer faces physical uncertainty that a plausible-looking answer cannot eliminate. All three need ways to verify important advice with people and records they trust. Their language, connectivity, cash flow and availability change what counts as a useful tool.

How to measure the gains

This is why a rollout should begin with an operator council. Recruit women and men, younger and older workers, vendors with different levels of literacy, people who share phones and enterprises with different margins. Pay participants for the time they spend testing. Ask them to identify repetitive tasks that cost money, such as comparing supplier prices, recording debts, preparing product descriptions, translating customer requests or finding trustworthy market information. Let them reject features that add steps without improving trade.

Then run a small pilot with a comparison group and a simple business ledger. Record baseline weekly sales, margin, time spent on the task, data and subscription charges, the cost of correcting errors and the number of decisions needing human help. Revisit those measures after the same period for participating and comparable non-participating enterprises, taking seasonal fluctuations into account. A comparison group and seasonal adjustments can improve an evaluation. Establishing causation also requires accounting for other differences between businesses, such as motivation, resources and market access. The result to investigate is the additional net benefit after the real cost of use.

Imagine a tailor in Nairobi who uses an assistant to prepare quotations. In this hypothetical example, the assistant saves 30 minutes on each batch of orders, but one mistake in the quantity of fabric needed consumes most of the savings. A sensible evaluation captures both the time saved and the cost of replacing the fabric. It also asks whether the tailor learns to spot the next error without outside help. That skill stays with the tailor.

Customer experience matters as much as productivity. Give operators an easy way to flag inaccurate translations, misleading prices, inappropriate product recommendations or advice that ignores local practice. Record whether vendors fix those problems and whether owners can stop using a tool without losing access to a sales platform, lender or adviser. A pilot in which people cannot decline participation gives a distorted picture of acceptance.

Payment design deserves similar scrutiny. A service that appears inexpensive in a promotional demonstration may cost more when operators pay for mobile data, transactions or repeated requests. The GSMA’s March 2026 study of mobile-money affordability examined five African markets and found that transaction-based taxes can increase payment costs and encourage some users to return to cash. Its findings concern mobile money. For AI services, the practical lesson is to measure the full cost from the user’s side of the transaction. [3]

The African Development Bank’s September 2026 call for research identified another set of questions worth testing. It sought a consultant to examine how AI could affect informal employment through direct assistance, platform decisions, competition and displacement. The call sets out a research agenda. Measuring those effects requires evidence from the research itself. [4]

What should determine whether a pilot expands

A national programme can set common expectations across different tools, including transparent fees, usable language options, access to a person when a recommendation is disputed, privacy controls that fit the business and public reporting of net benefit. Local trainers, trade associations, cooperatives and small-enterprise advisers can turn those expectations into practical exercises. Teach operators to compare a proposed action against their own records, identify when a recommendation is outside their experience and document corrections.

Release funding for expansion when independent evaluations show gains that survive fees, correction costs and time invested. Report differences by gender, business size, location and access to devices. Where the evidence is weak, redesign the service or stop the pilot.

Africa’s AI strategy will earn credibility when informal businesses can show, in their own records, that a tool improved the way they earn a living and strengthened their ability to make decisions.

About the author

Gleb Tsipursky, PhD, is a behavioural scientist, CEO of Disaster Avoidance Experts and author of The Psychology of AI Adoption at Work: From Resistance to Results (Georgetown University Press, 2026). Author’s book page

Photo credit: Courtesy of Dr. Gleb Tsipursky.

Sources

1. International Labour Organization, September 2026. An AI business adviser in your pocket: What we are learning from the ILO’s SABY pilot

2. Opportunity International and Safaricom, February 2025. Opportunity International and Safaricom launch new AI chatbot for smallholder farmers

3. GSMA, March 2026. Mobile money taxes and affordability in Sub-Saharan Africa: Evidence on user behaviour and market impacts

4. African Development Bank, September 2026. Call for expressions of interest on AI and informal employment in Africa

Exit mobile version