Two billion people work in the shadows of the formal financial system. They're not criminals or tax evaders—they're gig workers, street vendors, smallholder farmers, and cash-based entrepreneurs who power economies from Lagos to Manila to São Paulo. They earn income, pay expenses, and manage households. Yet traditional banks treat them as unbankable.
The reason is simple: conventional credit underwriting requires documentation these workers don't have. No W-2s. No tax returns. No credit history. In the eyes of legacy systems, they're invisible.
But artificial intelligence is changing everything. By analyzing alternative data—from mobile money transactions to smartphone usage patterns—AI models can now assess creditworthiness with accuracy that rivals or exceeds traditional credit scores. And it's not theoretical. Fintech lenders like Tala, Branch, and Jumo have collectively issued billions in loans to informal workers across Africa, Asia, and Latin America, proving the model works at scale.
For financial institutions, this represents one of the last major untapped markets. For society, it's a path to financial inclusion for populations that traditional banking left behind. Here's what executives need to know about this transformation—and why waiting is no longer an option.
The Market Is Massive—And Growing
Let's talk numbers. The informal economy generates roughly $10 trillion in annual economic activity globally. In developing markets, it typically accounts for 40-60% of GDP. In sub-Saharan Africa, over 85% of employment is informal. Even in the United States, 36% of workers participate in the gig economy.
These aren't marginal players. They're the majority.
The credit gap is staggering. The IFC estimates a $380 billion financing gap for micro and small enterprises alone. Add individual consumer credit needs, and you're looking at a market measured in trillions. Traditional banks have captured maybe 10% of it.
Why? Because the economics didn't work. Underwriting a $500 loan manually costs $50-100—making small-dollar lending unprofitable. High perceived risk meant banks either charged usurious rates or simply declined to serve this market entirely.
AI flips this equation. Automated underwriting drops costs to under $2 per application. Machine learning models trained on alternative data can predict default risk accurately enough to make these loans profitable at reasonable rates. Suddenly, the math works.
What AI Sees That Traditional Credit Bureaus Miss
Here's the insight that's powering this revolution: everyone leaves digital breadcrumbs, even if they've never had a credit card.
Consider a motorcycle taxi driver in Nairobi. No formal employment. No credit history. But he has a smartphone. Every day, he:
- Receives mobile money payments from passengers
- Pays for fuel using mobile money
- Tops up his phone regularly
- Maintains consistent location patterns along his routes
- Communicates with a network of contacts
Traditional credit bureaus see none of this. AI sees all of it.
The same principle applies to a market vendor in Jakarta, a freelance designer in Buenos Aires, or a small farmer in rural India. They all generate data exhaust that reveals creditworthiness—if you know how to read it.
The Data Sources Revolutionizing Underwriting
The most successful AI lenders draw from multiple alternative data categories:
Mobile Money Transactions: Services like M-Pesa in Kenya, GCash in the Philippines, and Paytm in India have created rich transaction histories. AI analyzes income consistency, expense patterns, savings behavior, and payment reliability. In many cases, mobile money data predicts default better than traditional credit scores because it shows real-time financial behavior.
Telecommunications Patterns: Call and SMS behavior reveals surprising insights. Someone with a diverse contact network, consistent usage patterns, and regular top-ups demonstrates stability. Sudden changes in communication patterns can signal financial stress before it shows up in payment delinquency.
Psychometric Data: Game-based assessments and behavioral surveys measure personality traits, risk tolerance, and financial decision-making. Companies like EFL and First Access pioneered this approach, finding that psychological factors can predict default as effectively as income data.
Geolocation Data: Where someone goes reveals who they are. Consistent commute patterns suggest stable employment. Regular visits to commercial areas indicate business activity. Movement patterns distinguish legitimate borrowers from potential fraudsters.
E-commerce and Marketplace Data: For platform sellers, transaction history is gold. Mercado Libre uses seller performance to underwrite loans. Shopify offers capital advances based on sales data. High ratings and consistent sales strongly predict loan repayment.
Utility Payments: Electricity and water bills are universally paid first when money is tight. Payment history on utilities is one of the strongest predictors of loan repayment—often better than credit card payment history.
The magic happens when you combine these signals. No single data source provides complete picture, but together they paint creditworthiness with remarkable clarity.
The Pioneers Are Already Winning
While traditional banks debated whether this market was worth pursuing, fintech startups proved it absolutely is.
Tala launched in 2014 targeting Kenyan workers with no bank accounts. Using only smartphone data, it underwrites loans in minutes. Default rates are comparable to prime credit card portfolios in developed markets. The company has since expanded to the Philippines, Mexico, and India, disbursing over $2 billion in loans.
Branch took a similar approach across Kenya, Nigeria, Tanzania, and India, reaching 5 million customers. Most had never had access to formal credit before.
Jumo pursued a different strategy: partnering with mobile network operators across Africa and Asia to underwrite loans using their data. By going B2B2C, Jumo reached tens of millions of customers without building its own distribution.
Kredivo in Indonesia focuses on e-commerce, providing instant credit at checkout for online shoppers with thin credit files. It's become one of Southeast Asia's largest digital lenders.
The results speak for themselves. These companies collectively serve millions of customers who were "unbankable" by traditional standards. Loss rates are manageable—typically 8-12% for mature portfolios, well within profitable ranges given the interest rates and scale. And customers are intensely loyal because for many, it's their first positive experience with formal financial services.
Why Banks Are Finally Paying Attention
For years, banks dismissed informal economy lending as a social impact project, not a real business. Three factors changed the calculus:
First, the proof points are undeniable. When fintechs demonstrate profitability serving this segment at scale, it's no longer speculative.
Second, competition is heating up. Fintechs are winning customers that banks assumed would eventually "graduate" into traditional banking. But why would someone who gets instant, hassle-free credit from a fintech switch to a bank that previously rejected them?
Third, technology costs have plummeted. Cloud infrastructure, pre-built ML models, and API-driven data access mean banks can now deploy these capabilities at a fraction of historical costs. What required $50 million and two years five years ago can be piloted with $2 million and six months today.
Smart banks are moving. Equity Bank in Kenya partnered with Safaricom for M-Pesa-based lending. ICICI in India offers instant overdrafts based on account analysis. Nubank in Brazil started by serving thin-file customers and became the country's largest digital bank.
The window for banks to capture this market is open, but closing. Wait too long and fintechs will own the customer relationship—with banks relegated to being infrastructure providers.
The Path Forward: Start Small, Learn Fast
Here's what executives getting this right understand: you don't need to underwrite the entire informal economy on day one. You need to learn.
Start with one segment. Pick motorcycle taxi drivers, or market vendors, or gig workers in a specific city. Keep it focused.
Deploy a pilot with 500-1,000 loans. Small amounts—$50 to $500. Short terms—30 to 90 days. The goal isn't profitability yet. It's data collection and model validation.
Accept that you'll lose money initially. Early cohorts might see 20-30% default rates as models learn. That's the cost of education. Price it into the business case.
Iterate rapidly. Every loan is a data point. Every default teaches the model something. Monthly model updates in the first year aren't excessive—they're essential.
Graduate successful borrowers quickly. Someone who repays a $100 loan on time should immediately qualify for $200, then $500, then $1,000. Reward good behavior instantly. That's how you build customer lifetime value.
Expand deliberately. Only after 6-12 months of proven performance in your pilot segment should you expand geographically or to new customer types.
The institutions winning in this space share a common trait: they're comfortable with experimentation and rapid iteration. They've adopted a startup mentality within their organizations, because that's what this market demands.
The Risks Are Real But Manageable
Let's be clear-eyed about the challenges.
Credit risk is higher. Even with great models, default rates will exceed traditional lending. Expect 8-15% losses at maturity, compared to 2-4% for prime lending. The key is pricing for this risk while remaining competitive with alternatives.
Fraud is sophisticated. Synthetic identities, first-party fraud, and organized fraud rings all target digital lending. You need robust fraud detection—device fingerprinting, biometric verification, network analysis—not just credit models.
Data quality varies wildly. Alternative data is messy. Mobile money providers might have inconsistent APIs. Telecom data includes errors. Your models must be resilient to missing and imperfect information.
Regulatory complexity is significant. Data privacy rules differ across markets. Interest rate caps exist in some jurisdictions. Consumer protection requirements vary. You need legal expertise in every market you operate.
Model risk requires ongoing management. Unlike traditional credit models validated once and used for years, these models need continuous monitoring and frequent recalibration. Behaviors change. Economic conditions shift. Models must adapt.
But none of these risks are fatal. They're management challenges, not fundamental barriers. Dozens of companies have navigated them successfully. The playbook exists.
The Strategic Questions Leadership Must Answer
Before committing resources, executives should pressure-test their conviction with hard questions:
Do we have the risk appetite? This isn't prime lending. Loss rates will be higher, especially initially. Can your organization stomach 20% losses in year one while models learn? Is your board prepared for that?
Can we move fast enough? The winners in this space make decisions in weeks, not quarters. They launch imperfect products and iterate. If your organization requires 18-month planning cycles and perfect launches, you'll lose to more agile competitors.
Do we have the right talent? You need data scientists who understand limited-data environments. Product managers who've worked on mobile-first experiences. Risk officers comfortable with alternative data. If you're hiring traditional consumer lending specialists, you're bringing the wrong playbook.
Build or partner? White-label platforms like Jumo or Oradian offer proven infrastructure. Data science consultancies can accelerate model development. Pure build strategies take years. Most successful banks use a hybrid approach—partner for speed, build for differentiation.
What's the path to profitability? Be realistic. Year one is investment. Year two approaches breakeven. Year three+ delivers returns. If you need immediate profitability, this isn't your market. But if you can take a 3-5 year view, the economics are compelling.
How does this fit our brand? Serving the informal economy can be positioned as financial inclusion (positive) or subprime lending (negative). The narrative matters. Are you prepared to champion this publicly?
The Competitive Endgame
Here's what the financial services landscape looks like in five years if current trends continue:
Fintechs will dominate informal economy lending in major emerging markets. They'll have billions of loans' worth of data, proven models, and loyal customer bases. Some will become banking platforms themselves—Nubank's path writ large.
Platform companies—Grab, Gojek, Mercado Libre, Shopify—will offer embedded finance to their ecosystems, using proprietary transaction data traditional lenders can't access.
Progressive banks will have hybrid models: partnerships with fintechs for distribution and data, proprietary AI for decisioning, balance sheet for funding. They'll be profitable but sharing margin with partners.
Legacy banks will be shut out entirely, or relegated to providing funding for others' platforms, capturing the lowest-margin piece of the value chain.
Which scenario your institution lives in depends on decisions made now. The technology exists. The market is proven. The only question is whether you'll participate meaningfully or watch from the sidelines.
Why This Matters Beyond Profits
Strip away the business case for a moment. Two billion people working in the informal economy aren't edge cases—they're the majority of the global workforce. Excluding them from financial services isn't just leaving money on the table. It's perpetuating inequality and limiting economic mobility for most of humanity.
Access to credit isn't just convenient—it's transformative. A market vendor who can access a $200 loan to buy inventory at bulk prices increases her margins by 20%. A motorcycle taxi driver who can afford bike repairs stays on the road earning income instead of losing days to broken-down equipment. A small farmer who can buy seeds at planting season instead of harvest season doubles crop yields.
Financial inclusion at scale doesn't just help individuals—it accelerates economic development for entire countries.
The institutions that crack this market aren't just capturing profit. They're enabling economic mobility. That's brand value money can't buy.
The Bottom Line
AI-powered underwriting for the informal economy isn't emerging—it's here. Billions in loans have been successfully deployed. The model works. The question facing financial institutions is whether they'll participate in capturing a multi-trillion-dollar market or cede it to more aggressive competitors.
The barriers to entry are falling. The technology is proven. The playbook is established.
What's required is strategic conviction and willingness to operate differently than traditional lending. Start small. Learn fast. Accept that early losses are investments in learning. Partner where it accelerates your path. Build where it differentiates you.
The institutions that move decisively in the next 12-24 months will define this market for the next decade. Those that wait will spend the 2030s explaining to boards why they missed the opportunity.
The informal economy is no longer invisible. The only question is whether your institution has the vision to see it—and the courage to serve it.
---
Cho-Nan Tsai is a three-time CTO and Professor of AI and Machine Learning at USC. He advises financial institutions across North America, South America, and Asia on AI transformation strategies.