The Frontier Is the Lab: A Note on Building AI for Finance in Emerging Markets

Promise and peril in roughly equal measure and why the people building here are solving the real problem, not a lesser version of it.

I want to set out plainly what I believe about this work, because I think the prevailing way of talking about AI and finance in emerging markets gets the emphasis wrong in both directions. The optimists oversell it as a frictionless leapfrog while the skeptics dismiss it as a sandbox where the real technology gets a watered-down trial run. Both are wrong, and the truth sitting between them is more interesting than either.

Here is the short version. The opportunity is genuinely enormous, larger than the rich-world conversation appreciates. The risks are genuinely serious, more serious than the boosters admit. And the people building carefully in this space are not working on a junior version of the problems being solved in London and San Francisco they are working on the harder version, and getting it right produces better technology for everyone. I hold all three of those at once. This is a note on why.

The promise is real, and it's not what gets advertised

Start with the opportunity, because it's the part that's easy to caricature.

The lazy version of the promise goes: billions of underbanked people, plus cheap AI, equals a vast new market. True as far as it goes, and useless as a guide to building anything. The real promise is more specific and more demanding. It's that AI can finally do, at a cost that works, the unglamorous translation work that has kept hundreds of millions of capable economic actors outside the formal financial system.

Consider who that system has historically excluded. Not the unsophisticated that's the condescending misreading. It has excluded people and businesses whose financial reality is perfectly real but illegible to the institutions that allocate capital. A trader who has run a profitable business for fifteen years but has never produced an audited statement. An SME whose entire operating history lives in mobile-money flows and supplier relationships rather than a general ledger. A founder who understands their unit economics intimately and has never had reason to format them into a model an investor can read.

These people are not data-poor. They are legibility-poor. The information exists; it just isn't in the shape the formal system demands. And until very recently, the only way to translate it was an expensive human an analyst, a loan officer, a relationship manager which meant it simply didn't happen at the bottom of the market. The economics didn't work.

That is the actual promise of AI here: it collapses the cost of translation. It can read the messy, partial, informal, real financial life of a person or business and render it into something the formal system can act on. Done well, that doesn't just serve a market it widens the perimeter of who gets to participate in the economy at all. That is a genuinely big deal, and it is worth being excited about without embarrassment.

The peril is real and it lives in the same machine

Now the other half, which the excited version skips, and which builders cannot afford to.

The same capability that includes people can harm them, and it's not a separate risk you can bolt a safeguard onto it's the same machine running. An AI that decides who is creditworthy from alternative data is also an AI that can encode bias into a loan decision and hand it to someone with no way to understand or contest the "no." A tool that produces a clean valuation from thin inputs is also a tool that launders uncertainty into false confidence making a guess look like a fact to a founder or an investor who can't see the difference. A system that scales financial advice to millions is also a system that scales a flaw to millions before anyone notices.

And the populations most served by these tools are often the least equipped to absorb the downside. A wrong credit decision, an over-confident recommendation, a model that quietly discriminates these land hardest on people with the thinnest financial cushion and the least recourse. The asymmetry is brutal: the upside of inclusion is spread across many; the downside of a bad system is concentrated on the vulnerable.

This is why I have no patience for the framing that treats safety, accountability, and humility as drag on innovation here as the boring compliance tax you pay to do the exciting building. In this domain, they are the building. A financial AI that isn't honest about what it doesn't know, that can't be held to account, that has no human answerable for its consequential calls, is not an innovative product with a compliance gap. It is a liability-generating machine pointed at the people least able to bear it.

Why the frontier is the lab, not the afterthought

Put those two halves together and you arrive at the conviction that animates this whole publication, and most of my own work.

Building AI for finance in emerging markets is not a constrained, lesser version of the "real" work happening in data-rich economies. It is the work, in its least forgiving form. The rich-world version of these problems is often hidden by abundance: when data is clean, plentiful, and mostly true, you can get away with systems that are quietly overconfident, opaque, and unaccountable, because the inputs are good enough that the flaws rarely surface. Strip the abundance away which is what the frontier does and every weakness in your system gets exposed immediately.

You cannot lean on clean data, because there isn't any. You cannot hide behind false precision, because the inputs are visibly shaky. You cannot pretend the human doesn't matter, because the local human holds context no dataset contains. You cannot treat accountability as optional, because the people affected have the least slack to absorb your mistakes. The frontier forces you to build systems that are honest about uncertainty, designed around human judgment, and accountable by construction not because a regulator made you, but because nothing else works here.

And those, it turns out, are simply better systems. The discipline the frontier imposes meet the data where it lives, calibrate your confidence to your inputs, keep an accountable human in the loop, treat the vulnerable user as the design center rather than the edge case produces AI that is more robust everywhere, including in the rich markets that thought they'd outgrown the need for it. Rich markets have thin-data pockets too: new ventures, private companies, novel situations, fast-moving conditions. A system built to be humble and accountable when the data is bad is more trustworthy when the data is good. The frontier isn't where you go to build a simpler thing. It's where you're forced to build the right thing.

What I'd ask of anyone building here

So if you're building in this space, here is what I'd put to you, as someone doing it alongside you rather than commenting from a distance.

Hold both halves. Don't let the size of the opportunity talk you out of the seriousness of the risk, and don't let the seriousness of the risk talk you out of building at all the cost of not building good tools here is also borne by the excluded. The mistake in both directions is the same: treating promise and peril as a dial you tune toward one end. They're not. They're two properties of the same system, and your job is to build something that captures the first while genuinely containing the second.

Be honest about what your system doesn't know. Keep a named human answerable for the calls that matter. Design for the person with the least slack, not the most. Treat legibility as the thing you're providing and accountability as the thing you're guaranteeing. Do that, and you're not building a charity-grade version of fintech for a poor market. You're building the version of financial AI that the rest of the world will eventually realize it needed too.

The frontier is not the place where the real work gets a discount. It's the place where the real work can't be faked. That's exactly why it's worth doing here and worth doing well.

Frontier Finance AI covers the collision of artificial intelligence and capital markets across Africa and Asia. These are the author's personal views.