AI Without Industrialisation: Can Africa Leapfrog the Traditional Development Model?

Every economy that became rich in the past two centuries climbed the same ladder: workers left low-productivity farms for factories, factories taught skills and paid wages, and the savings built everything else. Much of Africa has never climbed that ladder — and with automated factories in Asia, it may never get the chance. Artificial intelligence raises a provocative possibility: that productivity could arrive through phones, software and services instead of assembly lines.

To test the idea, we built an original model of AI exposure for five sectors that employ roughly 1,570,000 Mozambicans: banking, education, retail, logistics and public administration. The headline finding is counter-intuitive. AI is not about to destroy Mozambican jobs — but nor is it about to transform productivity. Of the work that AI could theoretically perform or assist, only about 6% is reachable today. The algorithm is ready; the economy around it is not.

Mozambique's AI problem is not too much automation. It is too little digitisation for AI to work on.

The classic route out of poverty — from Britain to South Korea to China — ran through manufacturing. Factories absorbed millions of unskilled workers, paid them more than farms could, and forced investment in roads, power and schooling. Economists call the movement of workers from low- to high-productivity activities structural transformation.

Mozambique's labour market shows how little of that has happened. The National Statistics Institute's informal-sector survey counted about 13.5 million people working informally in 2021 — the overwhelming majority of the workforce — and three-quarters of them were in farming and livestock. Formal wage jobs number perhaps a million, and a third of those are in the state. Gas, aluminium and coal generate export revenue but few jobs.

The economist Dani Rodrik has described a pattern of premature deindustrialisation: poorer countries are reaching their manufacturing peak at lower incomes and smaller employment shares than earlier industrialisers did. Automation in Asian factories makes the low-wage manufacturing route narrower still. If the factory door is closing, the question is whether a different door — digital services and AI — can open.

Two widely cited studies frame the global picture. The IMF estimated in January 2024 that almost 40% of jobs worldwide are exposed to AI — about 60% in advanced economies, 40% in emerging markets and 26% in low-income countries. The International Labour Organization's refined index, published in May 2025, finds that one in four workers globally is in an occupation with some exposure to generative AI: 34% of employment in high-income countries, but only 11% in low-income ones. Clerical occupations remain the most exposed.

The gradient is not a mystery. Exposure follows the occupational mix: rich countries have more clerks, analysts and professionals; poor countries have more farmers, traders and labourers whose work is physical. Lower exposure is often read as protection. It is better read as a measure of how little of the economy runs on information that a machine can process.

AI needs three things before it can raise anyone's productivity: electricity, connectivity and digital data. Mozambique has made real progress on the first: the World Bank reports electricity access rose from 31% in 2018 to 60% by the end of 2024. Connectivity lags. DataReportal counts 19.1 million mobile connections — 53% of the population — but only about 7.1 million internet users, or 19.8%.

There is, however, a precedent for leapfrogging. Mobile money arrived before banks did. By late 2025 Mozambique had more than 24.6 million mobile wallet accounts against 6.7 million bank accounts, served by over 400,000 agents. Millions of people skipped the bank branch entirely. The question is whether AI can follow the same path — delivered through the phone, in Portuguese and local languages, to people who never sat at a desk.

Global indices measure what AI could do to a job. They do not ask whether the job sits in an office with power, a computer and digitised records. Our model — the Economic Intelligence Mozambique AI Exposure Model — adds that second step.

• Step 1Break each sector's work into five task families: routine clerical, language and content, analysis and judgement, interpersonal and care, and physical and manual.
• Step 2Score each task family for generative-AI exposure — from 75% for routine clerical work to 3% for physical work — following the gradient of the ILO index, in which clerical tasks are the most exposed.
• Step 3Multiply task shares by scores to get theoretical exposure: the share of working time AI could perform or materially assist.
• Step 4Multiply by two Mozambique-specific readiness factors: digital access (share of workers with a powered, connected device at work) and workflow digitisation (share of the sector's information already in digital form). The result is realisable exposure.
• Step 5Split realisable exposure into substitution (AI does the task) and augmentation (AI helps a person do it faster or better).

Task shares and readiness factors are our editorial estimates, grounded in sector data and reporting; they are shown in full below so readers can challenge them — and the calculator later in this report lets you change every one.

The results split the five sectors into two worlds. Banking is Mozambique's most AI-exposed sector in theory (53% of working time) and, because it is already digital, in practice too (40%). It is the only sector where realisable exposure is close to theoretical exposure — and where substitution of routine back-office and contact-centre work is a near-term reality.

Public administration has the largest pool of exposed work in absolute terms — about 86,000 job-equivalents — because so much of government is forms, registers, letters and approvals. But with an estimated 35% digital access and 30% of workflows digitised, only about 4.6% of its working time is reachable. Teachers are highly exposed on paper (39%), mostly through augmentation — lesson planning, marking, explaining — yet almost none of it is reachable in classrooms without power or internet. Retail and logistics are dominated by physical and face-to-face work, and their low exposure is genuine.

Across the five sectors, about 421,000 job-equivalents of work are theoretically exposed. Only about 26,100 are realisable today — and of those, roughly 12,400 are substitution rather than augmentation.

Pick a sector preset, then change the assumptions. What happens to public administration if half its workflows are digitised? How much substitution would banking face if exposure were higher than we assume? The calculator applies exactly the same formula as our model.

Because readiness, not technology, is the binding constraint, the future depends on how quickly power, connectivity and digital workflows spread. We modelled three scenarios for 2035, holding employment and task mix constant so that only readiness changes.

In a Stalled world, readiness inches up and realisable exposure reaches about 39,000 job-equivalents — still a fraction of what is possible. In a Leapfrog world — where government registries are digitised, most schools are powered and connected, and traders run their businesses through phone-based tools — realisable exposure rises to about 178,000 job-equivalents, roughly 42% of the theoretical total. Most of it is augmentation: about 98,000 job-equivalents of work made faster or better, against 80,000 of substitution.

That is the real leapfrog case. Not a factory replaced by a chatbot, but a teacher, a clerk and a trader each doing more with the same hours — the productivity growth that factories used to deliver, arriving through services instead.

Partly — and not in the way the headlines suggest. AI will not create the mass wage employment that factories once did, and it cannot substitute for the roads, power and schools that industrialisation forced countries to build. What it can do is raise the productivity of the service and public-sector work that already employs Mozambique's growing urban workforce — if the plumbing is in place.

The lesson of mobile money is that leapfrogging is possible when the technology goes to where people already are. The lesson of our model is that, for AI, the place people already are is still mostly offline and on paper. Mozambique's real AI strategy is a digitisation strategy. The countries that understand that first will be the ones that leapfrog.

• May 2026INTIC presents Mozambique's draft national AI strategy for public consultation.
• Aug 2026Consultations with operators, including Vodacom, on the AI strategy, data protection and e-commerce rules.
• 2026EDM targets at least 420,000 new electricity connections by year-end.
• 2030Government target of universal electricity access.