Why the missing middle is still missing, and where it might finally move

Caleb Crawford

A working brief on small-business and early-stage finance in emerging markets. September 2026.

Read online at syndeo.dev/notes/missing-middle/

The small businesses that employ most people in the Global South cannot affordably borrow, and the startups that might grow into the next generation of employers in this part of the world cannot raise sufficient capital. It's a problem with an unimaginably massive scale, but the numbers have been so large for so long that they have stopped being alarming. The formal financing gap for micro, small, and medium enterprises (MSMEs) is now put at about $5.7 trillion, or $8 trillion once informal firms are counted, and roughly 70% of MSMEs in emerging markets say they lack adequate finance (IFC, October 2025; IFC). That gap is about 19% of the combined GDP of emerging-market and developing economies. Loans to small and medium enterprises (SMEs) amount to about 12% of GDP in high-income countries, 7% in middle-income countries, and 3% in low-income ones (Carvajal and Didier, World Bank, 2024). Roughly 30% of formal private firms across 109 economies are credit constrained, and constrained firms are the smaller ones (Islam and Rodriguez Meza, World Bank, 2023). In ten African countries the share is 53% (IFC, 2023).

I've spent the past year researching why no platform exists to facilitate this kind of lending at scale, and what can be done about it from a technology perspective. My (admittedly naive) initial question was, "Why can't a US investor with $5k invest in a startup or small business abroad instead of putting that money into TSLA or the S&P 500?" I've since learned that the answer to that deceptively simple question encompasses, just for starters, spheres of international finance, underwriting, foreign exchange risk, macroeconomics, and even geopolitics.

This note is my attempt to lay out that landscape as it stands today: what this gap is, why it has survived several decades of programs designed to close it, who the players are (starting with the governments of these countries) and what each can and cannot do, the unavoidable risks, and the handful of places where the economics might genuinely be changing. I'm putting this research out in the open to hear from people who are also trying to solve these problems.

What "missing middle" means#

The phrase describes firms that have outgrown microfinance but are too small, too informal, or too unusual for a bank. The Aspen Network of Development Entrepreneurs, which has done more than anyone to define the segment, calls them small and growing businesses: commercially viable, five to 250 employees, typically seeking growth capital between $20,000 and $2 million (ANDE). The Dutch Good Growth Fund, a fund of funds active in seventy countries since 2014, uses the same idea from the other side: SMEs that "have outgrown micro financing but do not yet have access to regular financial services" (DGGF). Deals between roughly $20,000 and $250,000 sit in the emptiest band: they are "too large for microfinance institutions' capacity" and "development finance institutions rarely go into this territory" (CSIS).

The definition has important implications. It means that the missing middle includes both a shopkeeper who needs working capital and a software company that needs equity. Both businesses are shut out for overlapping reasons even though they have structurally different financing requirements. It also means the gap is defined in large part by firm and financing size, and the businesses that fall within it are precisely those for which the cumbersome machinery of development finance has the least reach.

Why it persists#

These are the main reasons I have identified, and all five of these problems stack on top of each other. One could make a case that several of these problems individually fit what the design and planning theorists Horst Rittel and Melvin Webber called a "wicked problem," one that "can be considered to be a symptom of another problem" (Rittel and Webber, 1973). The tricky part is that any serious attempt to close the gap has to answer all of them.

1. The cost of a credit decision is essentially fixed, regardless of loan size#

Appraising, documenting, and monitoring a $30,000 loan takes roughly the same staff time as a $3 million one. According to UNCDF, "the costs to appraise and monitor investments for an SME are often the same as for larger transactions," which is why the segment sits "below the radar" of commercial banks, DFIs, and most impact investors alike (UNCDF, 2021). In microfinance the same arithmetic shows up as interest rates: operating expense is "the largest determinant of the rate the borrowers end up paying," and smaller loans need more operating cost per dollar lent (Rosenberg and others, CGAP, 2013).

The clearest measurement I've found is from East African agricultural SMEs. On loans between $15,000 and $1.75 million, commercial banks earn a net 3.2%, non-bank lenders lose 1.6%, and social lenders lose 15% before subsidies. Lenders name "high risks and origination costs as primary impediments" (Aceli Africa, 2024). The rational response of any lender is to write fewer, larger loans to the borrowers it already knows, because that's all the margin allows.

2. The information a lender needs mostly does not exist#

A bank in Lagos cannot look up most applicants. Credit-bureau coverage of adults in 2020 was 13.9% in Nigeria, 13.5% in the Philippines, and 36.4% in Kenya (World Bank Doing Business 2020, Nigeria, Philippines, Kenya). Without information, lenders ask for collateral, and in the World Bank's last measurements the collateral demanded was a jaw-dropping ~200% of the loan value or more: 228% in Nigeria and 255% in India in 2014, and 209% in Mexico in 2010 (World Bank Enterprise Surveys). Only 22.5% of firms in Sub-Saharan Africa have a bank loan or line of credit, against 40% in South Asia and 51% in Latin America, and firms in all three regions fund two thirds to three quarters of their investment from their own cash (World Bank Enterprise Surveys). When a loan does go bad, a creditor in Sub-Saharan Africa recovers about 20.5 cents on the dollar, against 70.2 in high-income OECD countries (Doing Business 2020, regional profile). Underneath all of it is informality: 84% of workers in Africa are informal (ILO via WIEGO), which means the paper trail a credit model needs usually doesn't exist.

3. Emerging and frontier markets are especially vulnerable to currency fluctuations#

A lender that raises dollars and lends naira is making a bet on the naira whether it wants to or not. From June 2023 to February 2024, the naira lost 69.5% of its value against the dollar (BusinessDay, 2024), and it lost another 40.9% during 2024 (BusinessDay, 2025). The Egyptian pound lost more than a third of its value on the day it was floated in March 2024 (AFP via Malay Mail). The Kenyan shilling fell 26.8% in 2023 (Cytonn). Outside Africa, the pattern is the same: Argentina's peso lost 54% of its dollar value overnight in December 2023, Sri Lanka's rupee lost 45% during 2022, and Pakistan's lost 28% over its 2023 fiscal year (Reuters, 2023; Central Bank of Sri Lanka, 2023; State Bank of Pakistan, 2023). These types of fluctuations are common in many emerging and frontier markets, and any structure that lends into them has to price for it.

The consequences disproportionately land on the smallest balance sheets first. Watu, one of the largest asset financers in East Africa, saw its 2024 profits fall 85% on currency depreciation and impairments before recovering in 2025 (TechCabal, 2026). Lipa Later, a Kenyan buy-now-pay-later company that had raised $12 million of equity, entered administration in March 2025 with dollar debt on its books after the shilling had moved from about 100 to about 170 per dollar over the life of the business (TechCabal, 2025). The old CGAP guidance on microfinance FX risk still applies: hard-currency indexing "creates devaluation risk for the MFI's loan clients, except in those rare cases where clients' income is denominated in hard currency," and hedges are often "unavailable or problematic because of the small size of the transactions" (CGAP, 2006).

Economists have an apt name for this underlying condition, which affects nearly all of these markets but is technically nobody's fault: "original sin," or the inability of most countries to borrow abroad in their own currency. Eichengreen, Hausmann, and Panizza traced this issue to the structure of global portfolios rather than to any single country's policy (NBER, 2003). Only two African sovereigns, Botswana and Mauritius, carry an investment-grade rating (Cleary Gottlieb, 2023), which sets a ceiling on how cheaply anyone inside those borders can borrow, whatever their own record.

4. Well-meant policy has an unfortunate habit of making it worse#

Kenya capped lending rates at four points above the policy rate in September 2016. The number of loan accounts fell 26.1% within nine months while the average loan size rose 36.7%, which the central bank read as "lower access to small borrowers," and it estimates that rationing MSMEs out of credit cut 2017 growth by 0.4 percentage points (Central Bank of Kenya, 2018; IMF, 2019). Kenya wasn't an outlier. At least 76 countries cap lending rates, and the same review finds the pattern in Latin America: in Nicaragua, a cap led microfinance institutions to cut lending and some to leave rural areas, and in Colombia, caps "severely affected small firms because of their higher transaction costs" (Maimbo and Henriquez Gallegos, World Bank, 2014). Zambia introduced its cap in 2013 with the stated aim of "increasing access to finance for SMEs." Over the next three years, annual growth in microfinance lending fell from 63% to 38%, lenders made up the lost interest with fees (which rose from 5.8% to 18% of interest income), raised their average loan sizes, and closed branches. The cap was repealed in under three years (Ferrari, Masetti, and Ren, World Bank, 2018).

Lending quotas fail more quietly. The Philippines required banks to set aside 8% of their loan books for micro and small firms and 2% for medium ones. While that mandate was in force, the MSME share of bank lending fell from 30% in 2002 to 16.4% in 2010, and the fine for missing the target worked out to about $9,300 per percentage point, which the ADB's researchers called "miniscule" (Khor, Jacildo, and Tacneng, ADB, 2015). By 2018, the big banks were lending 2.54% of their funds to micro and small firms against the 8% quota, and simply paying the penalty (BusinessWorld, 2018).

Debt relief does similar damage from the other direction. India's 2008 farm-loan waiver forgave more than $16 billion owed by some 60 million rural households. Banks responded by pulling back from the places where they had been forced to write off the most: credit outstanding fell 13% to 16% for each standard-deviation increase in a district's exposure to the waiver, and new lending came to 36 cents per dollar written off in high-exposure districts, against $4 in low-exposure ones, with no measurable gain in productivity, wages, or consumption (Giné and Kanz, World Bank, 2014). The borrowers were farm households, but the lender's reaction is the part that carries over to small firms.

The largest example isn't a credit policy at all. After 2008, regulators tightened anti-money-laundering and sanctions enforcement, and global banks responded by cutting ties with the small banks that cost the most to vet. The number of active correspondent banks worldwide fell about 22% between 2011 and 2019, and 41% in small island states (BIS, 2020). When the World Bank surveyed 110 jurisdictions, the authorities said the customers hit hardest were money transfer operators, then small and medium domestic banks, then small and medium exporters (World Bank, 2015). Compliance was one driver among several, since banks also cited plain business strategy (BIS Quarterly Review, 2020), but the direction was the same.

The pattern repeats across every one of these cases. Each rule was written with large banks and large borrowers in mind, and when it raised the cost or lowered the return of serving a customer, lenders dropped their smallest customers first. Regulation is still necessary in all of these markets. The lesson I take from these cases is that someone should check what a new rule does to the economics of a $5,000 loan before it takes effect. The same governments have also done a good deal that worked, which is where the section on the players starts.

5. On the equity side, there is nowhere to exit#

Venture capital in these markets is small and, more importantly, hard to get out of. Africa attracted $4.1 billion of tech funding in 2025, and $1.64 billion of that, 41%, was debt (Partech, 2026). In 2024, the continent's share of global venture funding fell below 1%, the median VC deal was $2.5 million, and there were 26 VC-backed exits all year (AVCA, 2025). The United States recorded an estimated 1,259 venture-backed exits in the same year (PitchBook-NVCA, 2025). Latin America saw $4.5 billion across 751 deals, with sixty companies that have each raised more than $150 million still waiting to list or be acquired (LAVCA, 2025). Southeast Asia's activity has reset to 2016 levels (Cento Ventures, 2025). Across emerging markets as a whole, private equity and venture funds returned a pooled net 8.47% a year over ten years, about 400 basis points below global public equities (Cambridge Associates, 2026). For the pension funds and endowments that supply most venture capital, those returns alone are reason enough to keep their money in the US and Europe, before political or currency risk even enters the conversation.

The accelerators and studios that sit at the front of this pipeline help, but less than their brochures suggest. The Global Accelerator Learning Initiative, which compares accepted ventures with rejected ones, finds that accelerated ventures in emerging markets gained about $15,000 more revenue and $14,000 more equity over a year than their rejected peers, that accepted ventures start out well ahead of rejected ones, and that "for many individual accelerator programs, accepted ventures do not end up outperforming rejected ventures" (GALI, 2017; GALI, 2021). Venture studios look better on follow-on funding, with about half of studio-built African companies raising a next round against roughly a third for accelerator alumni, on a base of forty-odd studios and 160 companies (GIZ SAIS, 2026). All of this is useful, but it's small relative to the size of the gap.

The players, and what each can actually do#

Host-country governments#

The largest player in most of these markets is the government of the country itself. It acts through its treasury, which spends on guarantees, subsidized loans, and tax breaks (fiscal policy), through its central bank, which sets interest rates, manages the currency, and regulates banks (monetary policy), and through the laws and public systems that only a state can build. Section 4 above covered the rules that backfired, and many of the guarantors described below are also government-owned. The rest of the record is more mixed, and in places more encouraging.

What governments spend. The programs governments run themselves are enormous. India's Emergency Credit Line Guarantee Scheme backed ₹3.68 lakh crore (about $45 billion) of loans between 2020 and 2023, 95% of them by count to MSMEs, and a fifth version worth ₹2.55 lakh crore opened in May 2026, with airlines among the eligible borrowers (SBI Research, 2026; SBI, 2026). Indonesia's subsidized KUR program lent Rp280 trillion (about $18 billion) to 4.92 million borrowers in 2024, with a non-performing ratio of 2.19% (ANTARA, 2024). The best evidence on whether this kind of spending works comes from Chile. In a randomized trial of 10,072 SMEs during the 2021 lockdowns, firms offered a loan backed by FOGAPE, the state guarantee fund, saw their total credit rise one-for-one with the loan, with 5% higher employment and 24% higher sales (Gertler and others, 2026). A later study using central-bank data found that the guaranteed loans postponed defaults by two years without reducing them, and that banks used what they knew to lend to the safer firms. Even so, the program's welfare gains exceeded its fiscal cost by 21% (Chittaro and Sánchez, 2026).

The failures are predictable enough to list. Brazil's audit court found that the country's 2020 emergency credit lines spent R$13.20 of budget for every R$1 of additional payroll, partly because guarantees went to firms that, in the auditors' words, "já teriam acesso ao crédito," or would have had access to credit anyway (TCU, 2021). South Africa announced a R200 billion loan guarantee in 2020 and disbursed R18.2 billion of it, because banks carried the first loss, earned no margin on the loans, and still asked owners for personal sureties (Parliamentary Budget Office, 2022). More than half of Indonesia's KUR borrowers came back for another subsidized loan, and only 3% moved on to commercial credit. When the government phased out subsidies for repeat borrowers, the share of new borrowers rose from 50% to 93% (Alibhai and others, World Bank, 2025). The designs that held up covered only part of each loss, set a rule for when borrowers graduate, and published enough data for auditors to measure additionality. Most of this money also goes out in loans of a few thousand dollars: the average ECLGS guarantee to an MSME was about $2,500, and the average KUR loan about $3,500.

What central banks steer. Central banks have two tools here: rules about whom banks lend to, and the stability of the currency itself. India requires banks to lend 40% of their adjusted net credit to priority sectors, including 7.5% to micro enterprises, and banks cleared that sub-target with 9.2% in the 2024 fiscal year (RBI; SBI Research, 2025). The rule has had real effects. When India widened eligibility in 1998, newly eligible firms grew their sales and profits, with no sign that the new credit was replacing other borrowing (Banerjee and Duflo, 2014). Banks can also buy their way into compliance, though. Trading in priority-sector lending certificates reached ₹8.95 lakh crore in 2024, about 14% of the credit base the targets apply to (SBI Research, 2025). Indonesia's central bank took a softer route in 2021, setting a system-wide target of 30% for inclusive lending by 2024 and rewarding banks that meet it with lower reserve requirements. Those incentives reached Rp255.77 trillion by June 2024, while non-performing MSME loans rose to 4.04% and banks became more selective with micro borrowers (Bank Indonesia, 2024). Direct lending by the central bank has the worst record. Nigeria's put more than ₦10 trillion into sector interventions before announcing in November 2023 that it would stop, and an internal review in 2025 found nearly ₦4.7 trillion still unpaid. The governor has since cited "market distortions, excess liquidity, and difficulties for monetary management" (Central Bank of Nigeria, 2023; ThisDay, 2026).

The larger contribution central banks make is stability. Countries that adopted inflation targeting cut the foreign-currency share of their international debt by 3 to 6 percentage points (Ogrokhina and Rodriguez, 2018), and emerging-market governments raised the local-currency share of their external debt from 20% in 2003 to 60% in 2017. Companies didn't follow, and the local-currency share of corporate debt stayed at about 10% (Onen, Shin, and von Peter, BIS, 2025). Investors also need to be able to take their money out. Nigeria's central bank owed banks about $7 billion in overdue dollar obligations in late 2023, and after an audit found $2.4 billion of the claims invalid, it cleared the rest by March 2024 (IMF, 2024; VOA, 2024; Nairametrics, 2024). Egypt's backlog of unmet dollar demand fell to zero after it unified its exchange rate in March 2024, with a $35 billion investment from Abu Dhabi's ADQ behind it (IMF, 2024). Argentina freed individuals from its currency controls in April 2025, but firms can still send home only profits earned from 2025 onward (Infobae, 2026). Investors rank these things first. In a World Bank survey of foreign-owned firms in ten middle-income countries, 49.4% called political stability critically important and 49.0% said the same of macroeconomic stability, against 41.3% for low taxes (World Bank, 2020).

What only governments can build. A registry that lets lenders take equipment, inventory, or receivables as collateral raised firms' access to bank finance by about 8 percentage points in the seven countries studied that introduced one, cut interest rates by 3 points, and lengthened loans by six months. Private credit bureaus had similar effects, and both helped smaller firms most (Love, Martínez Pería, and Singh, World Bank, 2013; Martínez Pería and Singh, World Bank, 2014). Brazil switched its register of on-time repayments from opt-in to opt-out in 2019 and went from 5.5 million people enrolled to about 100 million, and new borrowers with a score paid spreads about 10% lower (Banco Central do Brasil, 2021). Public payment systems now produce the transaction records the underwriting innovators below rely on. In India, a 1% increase in UPI transactions went with a 0.73% increase in credit, fintech lenders were the only ones lending to first-time borrowers, and defaults didn't rise (Alok and others, NBER, 2024). In Brazil, each 1% increase in Pix users in a municipality brought a 0.45% increase in people taking on credit (Sampaio and Ornelas, Banco Central do Brasil, 2024), and Indonesia's QRIS payment standard now has 39.3 million merchants, 93% of them micro and small businesses, though I found no study yet of what it does for credit (Bank Indonesia, 2025). Governments can also act as buyers. Discounting on India's TReDS invoice platforms grew from ₹40,000 crore to ₹3.47 lakh crore (roughly $5 billion to $40 billion) between the 2022 and 2026 fiscal years, and in June 2026 the government required its own state-owned companies to settle small suppliers' invoices through them (PIB, 2026).

This kind of infrastructure is slow to pay off. The Philippines passed its secured-transactions law in 2018 and fully launched the registry in February 2025 (Lawphil; Cruz Marcelo, 2025). Nigeria opened its collateral registry in 2016, and commercial-bank loans to SMEs were still 0.38% of bank credit in 2025 (Central Bank of Nigeria; TechCabal, 2026). Resolved cases under India's insolvency code took an average of 597 days and recovered 32.8% of creditors' claims, and its fast-track process for MSMEs had admitted 14 applications in four years (IBBI, 2025). Pension rules show the same lag. Nigeria lets pension funds put up to 15% of some funds into private equity, and they hold about 1.7%, largely because few local fund managers qualify (PenCom, 2025; BusinessDay, 2026).

What they can't control. No host government controls the financial cycle outside its borders. One global factor explains about 20% of the variation in risky asset prices around the world, and countries with floating exchange rates still feel US tightening (Miranda-Agrippino and Rey, NBER, 2019). Emerging markets with large dollar debts move their policy rates about one for one with US rate surprises (Juselius and Xia, BIS, 2026). Borrowing in local currency helps, but it moves the currency mismatch onto the foreign investors who buy the bonds, so local yields still rise when the dollar strengthens (Hofmann, Shim, and Shin, BIS, 2019; Onen, Shin, and von Peter, BIS, 2025). Ratings are the other outside constraint. UNDP estimates that ratings closer to fundamentals would be worth $74.5 billion to 13 African countries in new financing and lower interest costs, a model estimate but a large one (UNDP, 2023).

In short, governments set the conditions every other player works within: a currency investors can take home, legal and data systems that make small firms visible to lenders, and guarantees that, when designed well, get banks to lend to firms they'd otherwise turn down. The cost of underwriting any single small loan is still left to the lenders and investors below.

Development finance institutions#

The DFIs and multilateral banks hold the largest pools of international patient capital aimed at these markets, and they are, by design, structurally unable to make small loans. IFC says that it "does not lend directly to micro, small, and medium enterprises or individual entrepreneurs" (IFC). Every investment passes through a Kafkaesque twelve-stage project cycle with Board approval (IFC), an environmental and social categorization (IFC), and the eight Performance Standards that apply "throughout the life of an investment" (IFC, 2012). At the U.S. DFC, the process runs "a minimum of six to nine months" from formal application to commitment, with background checks and site visits (DFC). The European Investment Bank's direct loans start at €25 million (EIB).

That process shows up in the ticket sizes. Between 2012 and 2016, the median commitment was $7.6 million at FMO, $12.6 million at OPIC, $22.8 million at CDC, and $25 million at IFC, and only 15% of IFC commitments were under $1 million (Kenny and others, CGD, 2018). The World Bank's own evaluators found in 2014 that direct investments in SMEs made up one percent of IFC commitments and that many targeted SME projects were "weakly justified, weakly focused on SMEs, and/or have limited potential for additionality" (IEG, 2014). Here additionality means whether an investment causes something that would not have happened otherwise. Low-income countries received 6.4% of DFI and MDB commitments in 2018; 72% went to countries within six rating notches of investment grade (Attridge and Gouett, ODI, 2021).

So DFIs reach small firms through intermediaries: they lend to and guarantee local banks and microfinance institutions, which lend onward, and the scale of that channel is real. IFC's client financial institutions held $285 billion of SME loans across 5.4 million loans in 2023 (IFC Annual Report 2024). The DFC's portfolio guarantees can cover up to 80% of a bank's losses, and its own worked example describes a $50 million "micro" portfolio of 20,000 borrowers averaging $2,500 each, because those are borrowers "it would be impossible for DFC to underwrite individually" (DFC; DFC). FMO's NASIRA program, backed by EU guarantees of up to €365 million, shares losses with more than forty banks on loans to young, female, and migrant entrepreneurs (European Commission). Proparco's new Impact+ facility will issue guarantees of €2 million to €50 million to banks and MFIs covering loans as small as €500, up to 80% (Proparco, 2025).

That intermediated channel also has a long record of money not reaching the borrowers it was meant for. An evaluation of $13.4 billion of World Bank lines of credit found outcomes satisfactory in 45% of commitments by value, cancellations above 40%, fewer than half the projects using clear eligibility criteria for intermediaries, and almost 40% with no information on repayment rates (IEG, 2006). IFC, per its evaluators, "typically neither demanded nor received any information from financial intermediaries about their clients," and for wholesale guarantees there was "no evidence that the long-term tenor was passed on to end-borrowers" (IEG, 2014). Blended finance, the hoped-for multiplier, mobilizes about $0.75 of private money per public dollar overall and $0.37 in low-income countries (Attridge and Engen, ODI, 2019). The market did $18 billion in 2024 with a median deal of $65 million; SME-focused deals in Sub-Saharan Africa were 27% of the count but 14% of the money, and only 18% of them were under $5 million (Convergence, 2025; Convergence, 2024). Even IDA's Private Sector Window, purpose-built for the hardest markets, had used only 53% of its first allocation and paid out about $1 million in six years against $1.2 billion set aside (IEG).

A DFI dollar reaches a $5,000 borrower only through two or three layers of institutions, each of which needs its own reason to bother, and most of the delay and loss happens in the handoffs between them.

Guarantors#

Partial credit guarantees are arguably the instrument best matched to the missing-middle financing problem, but their track record is surprisingly uneven. In principle, they allow public or development-finance institutions to absorb part of a lender's downside risk, making banks more willing to lend to firms they would otherwise reject. In practice, however, guarantees still support only a small share of SME lending, and the evidence that they meaningfully change bank behavior is mixed. Even low default rates can be misleading: a guarantee scheme may perform well precisely because participating lenders continue selecting relatively safe borrowers rather than extending credit to firms that would otherwise go unfunded.

This makes additionality the central measure of success. The World Bank's principles say public schemes should cover at least 50% of a loan, but determining whether that risk-sharing changes lending decisions is difficult. Those same principles acknowledge that evaluating additionality is "technically challenging because of the difficulties in establishing a counterfactual" (World Bank, 2015). The research bears that out: Colombia's Fondo Nacional de Garantías did increase SME lending, while Malaysian participants' default rates rose and German schemes were associated with more risk-taking (Abraham and Schmukler, World Bank, 2017). Across Central and South-Eastern Europe, schemes ran net loss ratios averaging 3%, which the reviewers suspected "may also reflect the fact that guarantees are not reaching the smaller SMEs" (Vienna Initiative, 2014). In OECD countries, guaranteed loans are about 8% of the SME loan stock (OECD, 2024).

Interestingly, the recurring complaint from lenders is actually operational. A survey of nineteen Latin American schemes found that "the more complex the procedures for claiming the guarantees, the less the interest from participating financial institutions." Ten schemes required proof of legal notification before paying, and the schemes' own most pressing difficulty was streamlining documentation and building IT to automate procedures (World Bank, 2022). Elsewhere, "banks often complain that PCG administrators require them to file cases in courts and carry out time-consuming verification of claims," with a median of 120 days from missed payment to trigger (World Bank). Sri Lanka's scheme saw so many claims rejected for late or incomplete paperwork that lenders ultimately just "shied away" (ADB, 2022). The IDB Group, over fifteen years, did fewer than 4% of its lending as guarantees (IDB Office of Evaluation and Oversight).

Where the paperwork has been engineered away, guarantees have proven that they actually do scale. India's CGTMSE runs entirely through a portal with an API that lets a bank branch initiate a guarantee, and settles 75% of claims on first submission (SBI, 2025). Colombia's FNG guaranteed 953,787 credits in 2025 through a portal the lender uses and the borrower never sees (El Tiempo, 2026; MinHacienda). The African Guarantee Fund has issued $2.75 billion of guarantees to 256 partner lenders in 44 countries (AGF). Nigeria's InfraCredit has about ₦310 billion of guarantees outstanding with zero calls across 26 guaranteed entities (InfraCredit). And the newest program is the largest: the World Bank's FINCLUDE project, approved in December 2025, routes $500 million through Nigeria's Development Bank to issue up to $800 million of partial guarantees to banks, microfinance banks, and fintech lenders for 250,000 MSMEs, with an "AI-enabled digital platform for loan appraisal" written into the design (World Bank, 2025). The Philippines' PhilGuarantee has issued ₱252 billion of guarantees cumulatively, though housing dominates and MSME programs account for ₱7.46 billion (PhilGuarantee, 2024). Guarantee capacity, in other words, is being created faster than the capacity to originate loans that qualify for it.

The currency specialists#

The most important DFI-owned fund most people have never heard of is TCX, The Currency Exchange Fund, set up in 2007 by a group of DFIs to sell the hedges that commercial banks will not. It hedged a record $2.84 billion in 2025 across 54 currencies and 572 transactions, has enabled almost $20 billion of local-currency lending since it started, writes fixed-rate hedges out to 25 years, and describes itself as "often the only option in frontier currencies" (TCX, 2026; TCX). Its shareholders are KfW, EBRD, FMO, EIB, IFC, and AFD among others, and its rating agency notes "very high exposure to market risk relating to frontier currencies," with three currencies driving more than half of its losses in each of its bad years (S&P via TCX, 2025). The microfinance and SME sectors are its main beneficiaries (TCX Annual Report 2025). MFX Solutions, created in 2009 for smaller microfinance lenders and backed by DFC guarantees, has hedged around $5 billion in more than 55 currencies, laying the risk off to TCX or banks (MFX). IDA's Local Currency Facility backstops IFC lending "in markets in which currency hedging options are absent or very limited" (IDA), and IFC has written more than $30 billion of local-currency debt in 67 currencies over a decade (IFC, 2025).

So where are the efforts to pool this risk at a larger scale? The short answer is that TCX is the pool, and nearly every proposal for a bigger one is a proposal to enlarge it. Its shareholders "diversify and mutualize currency risk" across more than 50 currencies (S&P via TCX, 2025), which is the logic of an insurer spreading risk across many policies. At the end of 2025, it carried $9.24 billion of local-currency exposure in 58 currencies on $2.05 billion of capital. Beneath the shareholders sits a ~$272 million first-loss layer (money that absorbs losses before the shareholders' capital does) funded by the Dutch, German, British, Swiss, and French governments, and the European Commission has put €561 million of guarantees and contingent capital behind the fund (TCX Annual Report 2025; TCX, 2026). TCX also doesn't keep most of what it takes on. It sold $1.76 billion of 2025's risk, 62% of the year's volume, to banks and bond investors, and it has sold $8.8 billion since it started (TCX, 2026).

The plans to substantively grow that pool are mostly still just that: plans. In 2023, TCX proposed growing its capital from $1.3 billion to $5 billion in the near term and to more than $10 billion by 2030, with a donor-backed guarantee trust covering up to $10 billion of swaps and total hedging capacity of $50 billion (TCX, 2023). The G20's expert group on development bank reform backed the idea, recommending that TCX be built out "to a scale commensurate with the challenge" so that it could aggregate hedges and sell the risk on to capital markets (G20 IEG, 2023). A UN working group proposed a publicly capitalized facility that would manage exposure "across currencies as a portfolio," while leaving open who would actually run it (UN DESA, 2025), and AIIB and EBRD have a shared onshore liquidity platform, called Delta, in design (OECD, 2025). In mid-2025, a BCG stocktake still rated the pooled facility a "Proposal" and Delta as "Design phase," and it set TCX's ~$5 billion of cover against estimated demand of $50 to $100 billion a year (BCG, 2025). TCX's capital has since grown to $2.05 billion, mostly through retained earnings and the EC facility, but I couldn't find a donor decision that funds the larger plan. In the meantime, somewhere between 80 and 90% of DFI and development bank financing is still denominated in foreign currency (OECD, 2025).

What has actually launched is narrower. The EU now subsidizes the price of TCX hedges: a pilot that discounted them by an average of 2.23% a year was eight times oversubscribed, and its successor, a €150 million guarantee signed in July 2025, is expected to support up to €2 billion of local-currency lending (TCX, 2025; TCX, 2025). In Brazil, the IDB and the central bank agreed in November 2025 to channel up to $3.4 billion of long-dated FX derivatives through local banks under the Eco Invest program (IDB, 2025). Both of these help, and both operate at the wholesale level.

Two limits define this niche corner of the market. The first is price. TCX's own analysis notes that "in many frontier currencies, there is no established price for the risk" and that the exchange-rate-adjusted yield on dollar debt for high-risk countries has exceeded 15% (TCX, 2026), and hedging premiums in the most volatile frontier markets can run 10 to 15% a year (BCG, 2025). For the naira, TCX offers fixed-rate hedges only. For the peso, shilling, and Philippine peso, it offers both fixed and floating (TCX). The second is size. The Mexican peso trades $147 billion a day in the global FX market; the naira and shilling are not even reported separately (BIS, 2025). The commercial market will price a $10 million naira forward, but nobody will price a $10,000 one. The development-finance market doesn't reach much lower. TCX's average transaction in 2024 was $4.6 million (TCX, 2025). MFX will hedge as little as $100,000, but a microfinance institution first has to post a 7% initial margin and pay $8,000 to $15,000 for a credit rating (MFX). I found nothing that aggregates hedges below that floor. That leaves a lender making $5,000 to $100,000 loans with three options: build a book large enough to hedge, borrow from a fund that hedges on its behalf, or raise its money onshore in local currency to begin with (the neobanks below do this with deposits), which is the route that the ALCB Fund ($462.5 million across 109 local-currency bond investments in 22 countries) and Frontclear ($830 million of guarantees in 19 countries in 2025) exist to open up (ALCB Fund; Frontclear, 2025). Hedging is a portfolio activity, and whoever wants it has to bring a portfolio.

Venture investors, studios, and accelerators#

DFIs do sometimes reach startups, but generally as limited partners rather than direct investors. IFC has committed about $3 billion to more than 120 venture funds, seed funds, and accelerators (IFC); DFC's recent fund commitments run from $13 million to $40 million per fund (DFC, 2024). Antler's East Africa fund was $13.5 million, writing checks up to $100,000 (TechCrunch, 2022); Flat6Labs made 64 investments worth $12.1 million in 2023, about $189,000 each on average (Flat6Labs, 2024); Village Capital's alumni have raised about $1 billion on $17 million of direct investment (Village Capital). Y Combinator took 24 African startups in its winter 2022 batch and three in winter 2024 (WeeTracker, 2024). Flat6Labs is worth a closer look, because it's one of the few investors whose standard checks sit inside the missing-middle band, though only for tech startups and only as equity. Its 2023 numbers are also the most recent I can find. In August 2025, it split its accelerator programs from its investment arm, which became a separate firm, F6 Ventures (MENAbytes, 2025). I couldn't find an annual report from either company since, and F6 Ventures' website gives only running totals: more than $90 million across six funds and more than 300 companies (F6 Ventures). So the public record for one of the few investors working at these ticket sizes stops at the end of 2023. This layer selects well and signals well, but it doesn't move the missing middle, because most of the missing middle isn't venture-shaped: a viable trucking company or rice processor needs debt with a repayment schedule, and the venture industry, correctly, doesn't lend.

The neobanks#

Neobanks, or digital banks, are deposit-taking institutions that "deliver banking services primarily through electronic channels instead of physical branches," which in these markets mostly means a phone app (Ehrentraud, Garcia Ocampo, and Quevedo Vega, BIS, 2020). The label also covers app-first lenders with narrower licenses than a full bank. Moniepoint, for instance, is licensed as a Nigerian microfinance bank, and Nubank's Mexican arm was a SOFIPO, a regulated savings and loan company, when it won approval to become a bank in April 2025 (Moniepoint; Nubank, 2025). They deserve their own entry, because they've made real progress on three of the five problems above (cost, information, and currency) at a scale that nobody else in this note has reached. In mid-2026, Nubank had 139 million customers across Brazil, Mexico, and Colombia, with $45.3 billion of deposits funding a $39.4 billion credit portfolio, and it paid 88% of the interbank rate for those deposits (Nu Holdings, 2026). Brazil's Inter paid 66% (Inter&Co, 2026). Because the deposits and the loans are in the same currency, the hedging problem from the currency section simply doesn't arise. Nu's annual report says that none of its entities "had significant financial exposures in a currency other than their respective functional currencies" (Nu Holdings 20-F, 2026). Transaction data stands in for collateral, too. A BIS study of more than two million Chinese firms found that big-tech credit "reacts strongly to changes in firm characteristics, such as transaction volumes and network scores," while bank credit tracks house prices (Gambacorta and others, Review of Finance, 2023). China is the precedent at scale: MYbank has lent to more than 53 million small and micro businesses at an average loan of about $9,900, and WeBank's unsecured business loan has reached 1.8 million firms (Asian Banker, 2024; WeBank, 2026).

What they mostly lend for, though, is consumption. Nubank's book is $26 billion of credit cards, $10.3 billion of unsecured personal loans, and $3.1 billion of secured loans. It doesn't report a small-business line at all, even though its business arm has 6 million customers and is the largest in Brazil by number of business accounts (Nu Holdings, 2026; Nubank, 2026). At Mercado Pago, merchant credit is $2.5 billion of a $16.4 billion book, or about 15%, and the allowance against those merchant loans is roughly 40% of their value. The company carries loans until they are 360 days overdue, which inflates that figure, but it still says something about what unsecured merchant credit costs (MercadoLibre 10-Q, 2026). The business lending that does exist sits at the very bottom of the missing middle. Moniepoint lent more than ₦1 trillion (over $700 million, by the company's count) to about 70,000 Nigerian businesses in 2025, or roughly $10,000 per business over the year, and says that three out of four of them were borrowing formally for the first time (Techpoint, 2026; Vanguard, 2026). South Africa's GoTyme, formerly TymeBank, reports R17 billion advanced to 90,000 businesses (Cape Business News).

The BIS's summary of the evidence is that these loans "typically remain small and of short maturity," often six months to a year, and that "digital innovation by itself may not be enough to substantially improve SME lending" (BIS, 2025). Three things hold them there. The first is that the loans are unsecured working capital, underwritten from the borrower's own activity on the platform, so a firm that doesn't run its sales through the app is invisible. Moniepoint's working-capital loan requires a Moniepoint account, and the average loan at Kazakhstan's Kaspi lasts nine months (Moniepoint; Kaspi, 2026). The second is that the licenses are narrow. Nigeria's rules for microfinance banks, the license behind Moniepoint's bank, expect 80% of the book to be micro loans of up to ₦1 million with tenors "usually within 180 days," and they cap any single SME loan at 1% of shareholders' funds (Central Bank of Nigeria, 2020). South Asia has been more cautious still. India has no digital bank license (NITI Aayog proposed one in 2022), and its closest equivalent, the payments bank, "cannot undertake lending activities" (RBI, 2014; NITI Aayog, 2022). Pakistan licensed its first two digital banks for commercial operations in 2025 and 2026, with lending capped as a share of deposits while they ramp up (State Bank of Pakistan, 2022; State Bank of Pakistan, 2025; Raqami, 2026). Bangladesh's rules bar its digital banks from lending to medium and large industries, and none has opened yet (Bangladesh Bank, 2025). The third is that, outside the largest players, the model isn't paying for itself yet. The Philippines' digital banks lost a combined ₱3.97 billion in the first nine months of 2025, only two of the six were profitable in 2024, and the central bank has closed licensing again (BusinessWorld, 2026).

Where neobanks have moved up-market, something from the other sections of this note was usually involved. In May 2026, Nubank launched a working-capital line with terms of 6 to 82 months and a grace period of up to two years, and what makes those tenors possible is a guarantee from BNDES's FGI-PEAC program (Nubank, 2026). GoTyme's advances of up to R5 million came from buying a specialist lender, Retail Capital (GoTyme). The BIS also finds that fintech credit works as "a stepping stone for bank credit," and that almost a third of Mercado Libre's borrowers wouldn't have qualified at a traditional bank (BIS, 2025). I couldn't find a DFI credit line to a neobank that was earmarked for small-business lending. IFC's $265 million loan to Nu Colombia was aimed at consumers, and its stake in Moniepoint's $200 million Series C, alongside Proparco and Swedfund, is equity (IFC, 2023; Moniepoint, 2025). So the neobanks fit into this picture as the cheapest local-currency balance sheets in these markets, and they already have the distribution and the data, but so far they lend small and short. Nubank's guaranteed line suggests that pairing those balance sheets with guarantees is one of the more practical ways to push loan sizes and tenors up into the missing middle.

The non-bank innovators#

This is where some interesting things are happening. It's mostly at loan sizes that the heavy-hitting DFIs have not historically reached directly, and in lending, which venture capital funds generally don't do (encouragingly, however, many of these lenders are partly funded or backed by one or the other).

Reverse factoring anchors the credit on the large buyer instead of the small supplier. Mexico's NAFIN pioneered it at scale: by 2004 its Cadenas Productivas platform had 190 large buyers, more than 70,000 SME suppliers, about twenty lenders, and eleven million transactions, without recourse to the supplier and priced well below commercial rates (Klapper, World Bank, 2006). Its volumes later fell 40% in real terms as private banks copied the model (IPD Columbia), which, for a development bank, counts as success. India's TReDS exchanges, on which only MSMEs may sell invoices and buyers above ₹250 crore of turnover must register, have crossed ₹2 lakh crore of cumulative financing on RXIL alone (IBEF, 2025; RBI). In Mexico, Xepelin, Konfío, and Mundi now underwrite from the government's electronic invoice data and fund themselves with Goldman Sachs and Column facilities, in Konfío's case with a proposed 80% DFC guarantee on a $340 million line (DFC; Bloomberg Línea, 2022).

Asset financing ties repayment to a productive asset the lender can see and track. M-KOPA has extended more than $2 billion of credit to seven million customers for phones and solar, with default rates it has described as "a little above 10%" (M-KOPA, 2025; TechCrunch, 2023). Untapped Global finances motorbikes, water pumps, and similar assets that report their own usage and repays investors from a share of each asset's revenue, across 55,000 assets (Untapped Global).

Private credit funds and venture debt aggregate the small tickets that DFIs can't touch. Lendable had invested $576 million in eighteen countries with a 13.15% net IRR and 2.3% principal write-offs by the end of 2024, and 60% of its capital sits in subordinated tranches that make the senior layer safe enough for institutions (IFC; Convergence, 2026). Partech notes that most of Africa's record $1.64 billion of venture debt in 2025 was "labelled in USD with high interest rates" (Partech, 2026), which sends us back to the currency section.

Retail and crowd platforms prove the demand exists, and they also show its ceiling. Kiva has moved $2.4 billion through 2.9 million loans with a 96.4% repayment rate, and its lenders earn nothing (Kiva; Kiva). Lendahand, licensed under the EU's crowdfunding regulation, has placed €232 million since 2013 from about 20,000 investors at 2% to 8% interest, with write-offs of 0.37% on loans to financial institutions and about 5% on direct loans to companies (Lendahand; Lendahand). Colombia's A2censo bolts a 50% to 70% FNG guarantee onto crowd-funded SME debt and had five defaults in its first hundred campaigns (Valora Analitik, 2023). The ceilings are regulatory and arithmetic: the EU regime caps offers at €5 million per issuer per year (EU Regulation 2020/1503), U.S. Regulation Crowdfunding is closed to issuers not organized in the United States (17 CFR 227.100), and global equity crowdfunding was about $2.2 billion in 2020 against venture capital in the hundreds of billions (CCAF).

The underwriting innovators#

A mainstay of fintech optimism has long been that data can substitute for collateral. We're now starting to have enough data about the fintech ventures themselves to assess the validity of this assertion.

Tala, founded in 2011, underwrites from the data on a borrower's phone and has lent nearly $6 billion to ten million people in Kenya, the Philippines, Mexico, and India, reporting 90% repayment on its flagship product (Tala, 2024). JUMO has disbursed more than $10 billion in nine African markets (GlobeNewswire, 2026). Pula insures more than 21 million smallholders using satellite and weather data, which is underwriting by another name (Pula).

The newest layer reads the documents themselves. Kita, a 2025 startup in Y Combinator's winter 2026 batch, uses vision-language models to turn bank statements, invoices, and payslips into "fraud-checked, decision-ready signals," raised a $4.5 million seed led by BoxGroup with Tala's founder among the angels, and reports $130 million of loan volume processed for lenders in the Philippines, Indonesia, Mexico, and the United States in its first five months (Y Combinator; Menlo Times, 2026). In Nigeria, Indicina sells statement analysis and decisioning to about 120 lenders (TechCabal, 2022) and Lendsqr sells the loan-management rails, and as its founder has said, "Technology alone cannot scale a loan business without adequate capital" (TechCabal, 2024).

That said, the academic literature is a bit more careful than the vendors are in this area. The strongest recent evidence says data really can stand in for a missing credit file. In Mexico, a model built on the phone, email, and app activity of 181,488 RappiCard applicants who had no credit history ranked them with an AUC of 0.796, on a scale where 0.5 is a coin flip and 1 is perfect, and it held up on borrowers from a later period (Chioda and others, NBER, 2025). For small businesses, cash flow that a lender can verify matters too. At the Indian lender Indifi, a one-standard-deviation increase in a firm's use of cashless payments raised its chance of approval by 2 percentage points, lowered its interest rate by half a percentage point, and cut its probability of default by 13%, on top of what the owner's credit score already showed (Ghosh, Vallee, and Zeng, Journal of Finance, 2026). In China, Ant Group's automated credit lines raised online vendors' sales by 22% in the month after approval, with larger gains for vendors with less collateral (Hau and others, Journal of Finance, 2024).

What data hasn't yet replaced is the cushion collateral provides when loans go bad. Kenya's digital credit providers had a non-performing ratio of about 16% in early 2025, against 17% for banks, but 83% on loans under KSh 1,000 and 69% on loans of KSh 1,000 to 5,000, and the central bank noted that "the behavioural change after obtaining the loan may not be captured or predicted accurately by the scoring algorithms" (Central Bank of Kenya, 2025). In Indonesia, the share of peer-to-peer loans more than 90 days overdue rose from 2.5% in November 2024 to 4.4% in May 2026 (OJK, 2024; OJK, 2026), and a World Bank study of nearly 140,000 borrowers found that people whose first loan came from a fintech lender stayed 5 to 7 percentage points more likely to fall behind, even five years later and after moving on to bank loans (Alibhai and others, World Bank, 2026). The randomized evidence is similarly mixed. Small digital loans helped Kenyan households cope with shocks (Bharadwaj, Jack, and Suri, 2021), didn't measurably hurt borrowers in Malawi, and raised reported well-being in Nigeria, but repayment was weaker than headline figures suggest: in Malawi, only 38% of loans were repaid in full and on time, and in Nigeria, 23% of borrowers defaulted at least once (Brailovskaya, Dupas, and Robinson, 2024; Björkegren and others, 2022). Some of that default comes from the product design. One Mexican lender cut default by 21% just by taking twenty hours to deliver the money instead of ten (Burlando, Kuhn, and Prina, 2025).

I couldn't find an independent evaluation of document-reading or large-language-model underwriting in these markets yet. The closest evidence is from U.S. mortgage data, where large language models recommended more denials and higher rates for Black applicants than for otherwise identical white ones, a gap that simple prompt changes reduced (Bowen and others, 2024). The BIS's Financial Stability Institute warns that "the use of historically limited data sets in some regions has resulted in inaccurate credit self-assessments and contributed to elevated credit risk" (BIS FSI, 2026). In the Mexican study, most of the model's power came from the applicant's digital footprint, meaning the phone and email used to apply, rather than from what they bought. And across the industry, 65% of financial providers don't monitor their AI for discrimination, exclusion, or bias (CGAP, 2026). The open-finance rails that would feed these models are uneven: the Philippines launched a central-bank pilot in 2023 (IFC, 2023), Mexico's rules cover open data but not transactions (LatamFintech), and Nigeria's open banking had not gone live as of late 2025 (TechCabal, 2025).

The unavoidable risks#

Anyone proposing to lend small in these markets should be able to recite the losses.

Default. Kenya's banks ended 2024 with a gross non-performing ratio of 17.1%; on MSME loans specifically, 19.1% of value and 28.4% of accounts were non-performing, and banks wrote off 95,179 MSME loans that year. The same report shows why banks stay in the segment anyway: MSME lending was 21.4% of loans but 35.3% of lending income (Central Bank of Kenya, 2025). Nigeria's system-wide ratio rose to 8.1% in 2025 and Ghana's stood at 21.8% (World Bank WDI). Global microfinance portfolio at risk over 30 days sat around 7% before the pandemic (Microfinance Barometer, 2019). The gap between small and large borrowers shows up outside Africa too, even where the headline numbers are lower. In India, the gross non-performing ratio on MSME loans was 3.6% in March 2025, against 1.9% for large borrowers, and in Brazil, 5.7% of bank credit to micro, small, and medium firms was more than 90 days overdue in June 2026, against under 1% for large firms (RBI, 2025; RBI, 2025; Banco Central do Brasil, 2026). Kenya's government-run Hustler Fund reported defaults of 78% before recovery efforts brought them to 15%, and its auditor found 386,735 defaulter accounts closed without supporting evidence (The Star, 2026; Nation, 2026).

Fraud. Across 110 million identity checks in Africa in 2024, one provider recorded an overall fraud rate of 25%, biometric fraud averaging 16% a quarter, and microfinance institutions and digital banks as the most targeted lenders (Smile ID, 2025). Nigeria's interbank settlement system logged ₦17.67 billion of payment fraud losses in 2023 across 95,620 incidents (NIBSS via Vanguard, 2024). Loan stacking is structural: in Kenya's first digital-credit wave, 14% of borrowers held loans from multiple providers at once (CGAP, 2018). In India, an RBI working group found in 2021 that 600 of the roughly 1,100 lending apps it surveyed were illegal (RBI, 2021). Document fraud is precisely the thing the new document-intelligence tools exist to catch, which tells you how common it is.

Concentration. Supply-chain finance concentrates on the anchor by design, and Greensill showed what happens when nobody watches that concentration: Credit Suisse froze $10 billion of Greensill-backed funds in 2021, and the U.K.'s auditors found Greensill had made six loans worth £240 million to parts of one group on a single day under a scheme with a £50 million per-borrower cap (The Guardian, 2025; The Guardian reporting the NAO, 2021).

Currency and politics. The devaluations above are the ordinary case. For the extraordinary ones, MIGA insures against expropriation, currency inconvertibility, war, and breach of contract, issued a record $9.5 billion of guarantees in fiscal 2025, and has paid twelve claims in its history (MIGA; World Bank, 2025).

Model drift and regulatory whiplash. Every scoring model above was trained on a particular macro regime, and the IMF's warning about structural change is a warning about exactly the devaluations and rate caps this note began with. Tax rules move too. India extended its "angel tax" on startup share sales to foreign investors in 2023 and abolished it for all investors fifteen months later (Ministry of Finance, 2023; Ministry of Finance, 2024).

How other industries solved the same problem#

This is the part I find most encouraging: the problem of letting thousands of small originators safely draw on large pools of capital has already been solved, repeatedly, in other industries, and the solutions share a shape.

The U.S. Small Business Administration's 7(a) program guarantees 85% of loans under $150,000 and 75% above, and its Preferred Lender Program gives approved lenders "delegated authority to process, close, service, and liquidate" loans without prior SBA review. About 80% of 7(a) approvals now come through delegated lenders; the program did 77,600 loans for $37 billion in fiscal 2025, with charge-offs around half a percent (SBA; SBA, 2026; SBA, 2025). The mechanism is a rulebook, the Standard Operating Procedure, that defines an eligible loan precisely enough that the government can trust a stranger to make it.

The U.S. mortgage market does the same at a scale that is hard to picture. Fannie Mae's Selling Guide defines what a conforming loan is; about 1,200 lenders delivered loans to Fannie Mae in 2025, the top five accounting for only 36%, and non-bank originators now write 85% of agency mortgages against $9.3 trillion of agency securities outstanding (Fannie Mae 10-K, 2026; Urban Institute, 2026). Europe wrote a version of this into law with the 2019 "simple, transparent and standardised" securitization label, which now covers 41.8% of new issuance (AFME, 2026), and the European Investment Fund deploys close to €35 billion of InvestEU guarantees through more than 300 partner institutions (EIF; EIF, 2026).

Trade finance has run on shared rules for longer than any of them. The ICC's UCP 600 has governed letters of credit since 2007 and URDG 758 has governed demand guarantees since 2010; on 32 million transactions worth $16 trillion, the ICC Trade Register measured default rates of 0.04% on export letters of credit and 0.36% on import ones (ICC). Because the documents are standard, IFC's Global Trade Finance Program can guarantee local banks it has never underwritten in detail, and has supported more than $141 billion of trade over two decades (IFC).

Microfinance built its own version. Microfinance investment vehicles such as Symbiotics and BlueOrchard pool hundreds of small institutions for institutional money; Symbiotics alone has made more than 9,400 investments totaling over $12 billion in 669 institutions across 99 countries (Symbiotics), and private impact funds now hold $103.7 billion (Tameo, 2025). What made a Cambodian MFI investable from Geneva was a reporting standard (MIX Market, founded 2002) and a certification (the Smart Campaign's client-protection standards, now run by Cerise+SPTF) that reduced the diligence a buyer had to do from scratch (Cerise+SPTF). Insurance has done it since the eighteenth century: Lloyd's writes £55.5 billion of premium through more than a hundred syndicates and 3,000 coverholders who bind risk on the market's behalf under its rules (Lloyd's, 2025). Warehouse receipt laws in Kenya, Uganda, and Tanzania turned stored grain into collateral a bank will accept (WRSC Kenya). Visa's core rules let about 14,500 financial institutions interoperate on 329 billion transactions a year (Visa); India's account-aggregator framework has processed 538 million data-sharing consents (Sahamati). The American fintech lenders that could not get bank charters solved it with partnership: Upstart originates through more than a hundred lending partners with 91% of loans fully automated (Upstart 10-K, 2026), though the legal fights over who the "true lender" is have never fully ended.

Development finance has pieces of this stack, but nobody has assembled the whole thing. The World Bank published sixteen principles for credit guarantee schemes in 2015 (World Bank); the Operating Principles for Impact Management have 183 signatories (Impact Principles); IRIS+ has 781 metrics (GIIN); the 2X Challenge has moved $33.6 billion under a shared gender-lens definition (2X Global). CGTMSE and FNG have the portals. NASIRA and the DFC's pooled guarantees have the risk-sharing. GTFP has the standard documents. What none of them has, for a $5,000 to $100,000 business loan in Lagos, Lahore, Cebu, or Lima, is the combination those other industries take for granted: a written definition of an eligible loan, delegated authority to originate against it with audit rather than pre-approval, a shared backstop that sits behind many small originators, reporting rails that make the portfolio legible without a site visit, and a certification that lets a twelve-person lender be trusted by a $100 billion balance sheet. Each industry above built that combination once, usually after a crisis, and then grew for decades on it.

Where the economics might actually be moving#

I want to be careful here, because this is where most writing about emerging-market finance starts to overpromise. The deep constraints, informality, thin markets, weak enforcement, currency, won't be automated away. What can change is the cost of the steps that turn a willing borrower into a loan a guarantor will stand behind and a lender will book. Five of those steps look different than they did five years ago.

The credit decision is getting cheap for the first time. Document reading, statement analysis, and alternative-data scoring have moved from research to production, and the evidence above says they work best exactly where bureau data is thin. If the fixed cost of a decision falls from a credit officer's afternoon to a few cents of compute plus a review, the arithmetic in the Aceli report changes and the loan size at which a lender breaks even falls with it. That is the single most important shift, and it is already visible in the Mexican and Indian studies above and in the speed at which a three-person company like Kita has reached lenders on three continents. It is also incomplete, as Lendsqr's founder said, because a cheaper decision does nothing without capital behind it.

Guarantee eligibility can be checked before the loan is made, and claimed without a lawsuit. Every failure mode in the guarantee literature is a paperwork failure: eligibility unclear, reporting late, claims rejected for missing evidence, court filings required before payout. CGTMSE's portal and FNG's show that when the guarantor exposes its rules as software, uptake follows. Both guarantors are government-owned, so this is a choice a finance ministry can make. FINCLUDE's $800 million of new Nigerian guarantees will reach 250,000 businesses only if the long tail of microfinance banks and fintech lenders can originate qualifying loans and report on them, and that is a software problem sitting on top of a policy achievement.

Shorter loans, backed by what a large buyer owes, turn currency risk into something a fund can hedge. Nobody will hedge one $8,000 naira loan, but TCX and MFX will hedge a portfolio of them. The length of each loan matters too. An invoice that a large, creditworthy buyer has agreed to pay in 60 days (an account receivable) carries far less currency risk than an 18-month loan to the small supplier who sent it, because the exchange rate has less time to move. That is the logic of reverse factoring, described among the non-bank innovators above: the lender advances cash against invoices the large buyer, or "anchor," has already approved, so the credit risk sits with the party best able to carry it. It has precedents at scale (NAFIN, TReDS) and a cautionary tale (Greensill), and TReDS shows how much faster it grows when a government tells its own companies to pay through it, and together they describe what a design has to include: limits on exposure to any one anchor, checks that each invoice is real, no recourse to the small supplier if the buyer fails to pay, and someone watching the concentration.

Pooling to the thresholds that DFIs can actually fund. The DFC's own model of a $50 million portfolio of 20,000 borrowers at $2,500 each, Proparco's guarantees covering loans from €500, Lendable's subordinated structure, A2censo's guaranteed campaigns: these are all the same idea. A DFI will never appraise the borrower, but it will appraise the pool if the pool is built to a standard it recognizes. The originator's job is to build the pool to that standard from the first loan, and to make the reporting so legible that the layers of intermediaries stop losing money and information along the way.

The diaspora is a capital source that already prices the currency risk correctly. Remittances to low- and middle-income countries reached $685 billion in 2024, more than foreign direct investment and aid combined; India received an estimated $129 billion, Mexico $68 billion, the Philippines $40 billion, and Pakistan $33 billion, and Nigeria received about $19.5 billion the year before (World Bank, 2024; World Bank/KNOMAD, 2024). Someone who sends money home already holds naira or peso obligations and reads a Lagos business more accurately than a fund manager in Geneva. The record of diaspora bonds is mixed, from Nigeria's oversubscribed $300 million in 2017 to Kenya's 24% of target on M-Akiba (Vanguard, 2017; Diaspora for Development), which suggests the product has to be closer to the businesses than a sovereign bond and more structured than a family loan.

None of these five is a business on its own, and none is new as an idea. What is new is that the first one, the cost of the decision, has moved enough that the other four might finally be worth assembling around it. That is the hypothesis I'm spending this year testing, in conversations with lenders, guarantors, founders, policymakers, and the people who send money home. I don't have the design yet. What I can say is that its shape is legible now in a way it wasn't before, that the pieces exist in other industries and in fragments in this one, and that the real work is in the unglamorous operational details.