The global conversation about artificial intelligence adoption tends to begin with a headcount: How many people have used AI? How many have reliable internet access? How many can reach a frontier model?
Those questions are useful. But for lower-income and early-capacity economies, they are incomplete. They measure the spread of access and awareness more clearly than they measure whether technology is improving how a hospital treats patients, how a municipality delivers services, how a cooperative reaches a market, or how a small business gets paid.
For frontier markets, the more important question is not, “How quickly can everyone adopt AI?” It is:
What is the minimum level of effective adoption, among the right institutions, required to produce broad social benefit?
I call this Minimum-Effective-Diffusion, or MED.
Diffusion is not the same as benefit
Microsoft’s 2025 AI Diffusion Report provides a valuable map of global AI use and the foundations that support it. Its broad conclusion is persuasive: adoption rises where electricity, connectivity, computing, skills, and language access are stronger.
The report also makes the scale of the divide visible. Countries building frontier models are concentrated in one part of the world, while countries with weaker foundations appear at the bottom of adoption rankings.
A terminology distinction matters here. Microsoft uses frontier to describe countries at the leading edge of model development. I use frontier markets in the development and investment sense: economies with structural constraints and early-stage digital capacity. MED is a framework for the latter.
Where I believe the diffusion conversation can evolve is in the relationship between use and impact. A country can rank low in individual AI adoption and still generate meaningful gains if a smaller number of high-leverage institutions use digital systems well. Conversely, a country can show growing consumer experimentation without changing a single consequential workflow.
Diffusion is an input. Social benefit is the objective.
What Minimum-Effective-Diffusion means
MED is the threshold at which effective, productive, institutionally embedded adoption begins to create system-level benefit.
It is different from maximal diffusion, where nearly everyone uses a technology, and nominal diffusion, where access exists but is shallow or irregular. MED focuses instead on a limited set of capable, strategically central actors:
- municipal governments;
- hospitals, clinics, and health networks;
- schools and vocational programs;
- agricultural cooperatives;
- small and medium-sized enterprises;
- payment, identity, and logistics providers.
These actors matter because they sit inside workflows that many other people already depend on. When they reduce processing time, improve reliability, lower leakage, or expand access, non-users benefit too.
- Institutional adoptionGovernment, health, MSMEs, cooperatives
- Routine workflowsTools become part of actual service delivery
- Operational gainsLess friction and leakage; more trust
- Market expansionMore people can participate in useful services
- Public benefitValue reaches people who never touched the tool
The 10–30% range I proposed in the working paper should be understood as a planning heuristic, not a universal empirical law. The threshold will vary by sector, network structure, institutional capability, and the strength of complementary infrastructure. The central claim is directional: which actors adopt, and how deeply the technology enters their workflows, can matter more than raw user share.
The lesson from mobile money
Mobile money is a useful analog because its welfare effects did not depend on every citizen becoming a sophisticated financial-technology user. What mattered was the creation of a reliable service layer: agents, payment rails, transfers, savings behavior, and connections between households.
Research by Tavneet Suri and William Jack found that access to Kenya’s M-PESA system increased consumption and helped lift an estimated 194,000 households—about 2% of Kenyan households—out of extreme poverty. The gains were especially pronounced for female-headed households. Their finding was not simply that more people had phones. It was that a service embedded in economic life changed resilience, saving, and occupational choices. See the study in Science.
This is the pattern MED is designed to identify: infrastructure becomes consequential when it supports a trusted service and a repeatable workflow.
Coverage is necessary, but it is not sufficient
Infrastructure still matters. Reliable electricity, connectivity, devices, and data capacity define what is possible. But coverage figures alone can hide the larger adoption problem.
The GSMA’s 2024 State of Mobile Internet Connectivity report estimated that only 4% of the global population lived outside mobile-broadband coverage, while 39% lived within coverage but did not use mobile internet. That gap demonstrates the difference between availability and meaningful use.
The same is true of electricity. A connection that is unreliable, unaffordable, or disconnected from productive activity cannot deliver its theoretical benefit. Access is a prerequisite. Integration is the multiplier.
Why institutions become the leverage point
In frontier markets, institutions—formal and informal—are often gateways of trust, service distributors, workflow hubs, and economic bottlenecks at the same time.
Consider a clinic. One clinician using an AI assistant casually is individual adoption. A health network integrating a carefully governed tool into triage, scheduling, stock management, or follow-up is institutional adoption. The second case can improve the experience of thousands of patients who never know AI was involved.
The same distinction applies to a municipality processing permits, a cooperative coordinating crop information, or a small-business network using interoperable payments. The unit of analysis should not be “AI users.” It should be the workflow, the institution responsible for it, the population it reaches, and the outcomes that change.
Haiti’s diaspora is part of the system
For Haiti, the model must include the diaspora. It is a source of capital, skills, information, trust, and demand for cross-border services. World Bank data show that personal remittances represented 16.3% of Haiti’s GDP in 2024. That scale makes the diaspora more than an external stakeholder; it is part of the country’s operating environment.
A diaspora-integrated service can create a two-sided MED effect. Better payments, identity verification, logistics, health coordination, or professional services can reduce friction for people in Haiti while also increasing transparency and trust for people supporting family members or doing business from abroad.
A different AI strategy for frontier markets
If we optimize for MED, the sequencing of investment changes. The first priority is not universal access to the newest model. It is making a small number of essential systems capable of using digital tools safely and repeatedly.
That means investing in:
- reliable electricity and connectivity at critical institutional nodes;
- payment rails, digital identity, and logistics infrastructure;
- clean, governed, locally relevant data;
- digitized workflows before AI automation;
- staff capacity and accountability;
- diaspora-linked services where they strengthen trust and participation.
It also changes measurement. Success should include time saved, errors prevented, leakage reduced, service reliability, user trust, and the number of people reached through an improved workflow. Consumer account counts can remain part of the picture, but they should not stand in for public value.
The question that matters
Minimum-Effective-Diffusion does not reject broad diffusion. It changes the order of operations.
Institutional capacity precedes durable consumer adoption. Trust precedes participation. Value creation precedes market growth.
Haiti and other frontier markets should not be evaluated only by how far they are from universal AI usage. They should also be evaluated by how effectively they can activate the institutions that already connect people to health, education, government, finance, agriculture, and commerce.
Mass benefit does not require mass adoption. It requires the right diffusion, through the right institutions, into the right workflows.
That is the practical promise of Minimum-Effective-Diffusion—and a more useful place to begin.
Sources and further reading
- Microsoft AI Economy Institute, Global AI Adoption in 2025.
- GSMA, The State of Mobile Internet Connectivity 2024.
- Tavneet Suri and William Jack, “The long-run poverty and gender impacts of mobile money,” Science, 2016.
- World Bank, Haiti country data.