By William Song-Aabo
On August 3, 2026, the Bank of Ghana published a notice naming 20 mobile lending apps, including Adamfo Loan, CediGo, and Sika Tap, and warning the public against using them. The apps, the central bank said, had been offering digital credit without a licence, in violation of a directive introduced less than a year earlier to bring Ghana’s fast-growing digital lending industry under formal oversight.
The notice was a rare, concrete assertion of authority over a sector that has expanded largely unchecked: dozens of apps now approve or deny credit in seconds, using artificial intelligence trained not on pay slips or bank statements but on mobile money transaction histories and smartphone usage patterns. For banked, urban Ghanaians, that shift has made credit faster than ever. For the roughly 80 per cent of the workforce that operates informally, it risks doing the opposite: quietly, and with far less scrutiny than an illegal loan app gets.
How the Algorithm Decides
Fido, a Ghanaian lender operating since 2014, calls its underwriting engine the “Fido Score.” Instead of pulling a traditional credit history, which most of its customers don’t have, the model draws on alternative data: mobile money transaction patterns, phone usage behaviour, and other digital signals, assembled into a real-time risk profile. Carbon, a lender also active in Ghana, describes its own approach even more directly on its website: “We’re not just a lender, we’re a data company.” Both promise what the industry calls instant credit: approval or denial within minutes, no collateral, no paper application, no loan officer.
The pitch is genuine. For people the traditional banking system has never served, this can be the first credit they’ve ever accessed, and demand is real in a country where formal bank branches thin out fast outside major cities.
But alternative data is not neutral data. A human loan officer might simply have no way to evaluate someone without a credit history. An AI model trained on mobile money and phone data does something more specific: it learns which data patterns count as low-risk based on the loan performance records used to build it in the first place, a certain volume and rhythm of mobile money activity, a certain kind of device, a certain digital footprint. Applicants who match that pattern get fast and cheap credit, while applicants who transact in cash or share a phone do not. This is because these individuals are statistically invisible to a model that was never trained with their data, so it has nothing to judge them by.
The Shape of Exclusion
The scale of who this could affect is large. Ghana’s informal economy (traders, artisans, drivers, small-scale farmers, day labourers) accounts for roughly 80 per cent of the country’s workforce, according to the Ghana Statistical Service. Informal work is, by definition, cash-heavy and undocumented: no payslips, no employer records, and often thin, irregular mobile money activity that looks nothing like the steady digital trail an algorithm is built to recognise as low-risk.
Gender compounds the gap. The World Bank’s Global Findex survey found that 63 per cent of Ghanaian women held a formal financial account in 2021. That’s real progress, but the gender gap in account ownership actually widened over the preceding five years, from 8 percentage points to 11. Women in Ghana’s informal economy are disproportionately represented among traders and small-business owners who transact in cash and shared phones, patterns that generate exactly the kind of sparse, inconsistent digital footprint that alternative-data models are least equipped to read.
None of this means any specific lender’s algorithm has been proven to discriminate: there is no public audit of Fido’s or Carbon’s models, and none exists for most of Ghana’s digital lenders. What the data does show is the shape of the risk. A scoring system built on mobile money and smartphone signals will, by construction, work best for the minority of Ghanaians who are already formally banked, urban, and digitally active, and worst for the informal majority the fintech sector says it’s trying to reach. The absence of an audit isn’t reassurance. It’s the problem: nobody, including the regulator, currently knows how large this gap actually is.
Not Ghana’s Problem Alone
This dynamic isn’t unique to Ghana, and it isn’t hypothetical. CGAP, the financial-inclusion research group housed at the World Bank, has warned since at least 2019 that alternative-data credit models risk encoding the same exclusion they’re meant to fix. Researchers Maria Fernandez Vidal and Jacobo Menajovsky note that if a model is trained on data reflecting who has historically been excluded, it can reproduce that exclusion under the appearance of neutral maths. The algorithm doesn’t need to be told someone is poor or informal to penalise the data signature that comes with being poor or informal.
The clearest empirical evidence comes from outside Africa. A peer-reviewed study published in the Journal of Finance in 2022, examining machine-learning underwriting in the US mortgage market, found that algorithmic credit models increased disparities in loan pricing between demographic groups, not despite their statistical sophistication but because of it. No comparable empirical study yet exists for Ghana’s digital lending market. That gap is itself worth noting: a sector processing credit decisions for millions of people is operating with no independent research measuring its distributional effects at all.
What the Regulator Should Do Next
Closing that gap doesn’t require Ghana’s regulator to start from scratch. The Bank of Ghana has already shown it’s willing to act. Its September 2025 Directive for Digital Credit Services Providers, effective from November 2025, brought digital lending under formal licensing for the first time, requiring providers to meet capital thresholds, maintain at least 30 per cent Ghanaian equity, and register through the central bank’s online licensing system. The August 3 crackdown on 20 unlicensed apps shows the directive isn’t just paperwork: the regulator is willing to name names and enforce it.
That’s real progress, but it targets the wrong layer. Licensing controls who is allowed to lend. It says nothing about how the licensed lenders decide who qualifies. A fully compliant, properly capitalised, majority-Ghanaian-owned digital lender can still run a model that systematically underserves the informal majority, and under the current directive, no one would know, because nothing requires it to be checked.
Three additions would close that gap without requiring the regulator to build new institutions from scratch.
First, mandate periodic algorithmic fairness audits for licensed digital credit providers: testing whether approval rates or pricing differ systematically across demographic or geographic groups. The results don’t need to be public to be useful; even a supervisory-only requirement would surface problems currently invisible to everyone, including the lenders themselves.
Second, require disclosure of the categories of data used to build and run credit models, not the proprietary algorithms. Regulators don’t need a lender’s source code; they need to know whether “alternative data” quietly means “urban and digitally active” as a proxy.
Third, extend the directive’s existing transparency language to require a real reason for denial: not a generic rejection, but enough specificity for a rejected applicant, or a researcher, to spot a pattern if one exists.
None of this is radical; versions of all three already exist in fair-lending frameworks elsewhere. What’s missing in Ghana isn’t regulatory capacity. The directive and the enforcement action both prove that exists. What’s missing is scope: the rules written for who can lend need to be extended to cover how lending decisions get made.
Beyond the Crackdown
The 20 apps named on August 3 will likely disappear from app stores within weeks, replaced by others just as quickly. That’s the nature of enforcement aimed at unlicensed operators: it’s a game of whack-a-mole that the licensing regime is designed to eventually close. But the licensed lenders that remain, the Fidos and Carbons of Ghana’s fintech sector, aren’t going anywhere. They’re growing and raising capital. For a generation of Ghanaians with no credit history at all, these apps are becoming the default door into the financial system.
That’s precisely why the algorithms deserve the same scrutiny the apps just got. Ghana’s regulator has proven, twice now, that it’s willing to act. The harder, less visible work, auditing not who lends but how lending decisions get made, is the one that will actually determine who gets left out of Ghana’s financial system for the next decade, not just the next news cycle.
Sources
1. Bank of Ghana, notice naming 20 unlicensed digital credit apps, August 3, 2026: GBC Ghana Online (news coverage of Notice No. BG/GOV/SEC/2026/25; also corroborated by abcnewsgh, Pulse Ghana, and Adomonline; no direct bog.gov.gh URL for this specific notice was locatable at time of writing)
2. Bank of Ghana, Directive for Digital Credit Services Providers in Ghana, September 2025 (primary)
3. Ghana Statistical Service, informal-sector employment share (~80%), via Ghana News Agency reporting; corroborated by WIEGO statistical brief
4. World Bank, Global Findex Database 2021, Chapter 1: Ownership of Accounts (Ghana account-ownership and gender-gap data)
5. Fido (fintech press corroboration: FinTech Futures, Africa Fintech Network, Launch Base Africa, Dabafinance, Lucidity Insights); Carbon (company’s own site)
6. Fernandez Vidal, M. & Menajovsky, J., “Algorithm Bias in Credit Scoring: What’s Inside the Black Box?” CGAP, September 5, 20197. Fuster, A., Goldsmith-Pinkham, P., Ramadorai, T., & Walther, A., “Predictably Unequal? The Effects of Machine Learning on Credit Markets,” Journal of Finance, 2022. Published version: DOI 10.1111/jofi.13090; open-access working paper: SSRN
