Federated learning, PI-SPI, portfolio at risk, probability of default: the key terms of SME lending and credit scoring in the UEMOA, explained in plain language for teams at banks, MFIs, SFDs and fintechs.
A
Information used to assess a borrower beyond traditional sources such as financial statements or credit history: mobile payment flows, QR code collections, utility bills, online activity. It is especially useful for SMEs with little or no banking history. Its use requires the customer’s consent and compliance with data protection rules.
Our platform →B
The Central Bank of West African States is the common issuing institution of the eight UMOA member states. An international public institution headquartered in Dakar, it holds the exclusive right to issue currency across the Union. Among other things, it sets the rules for credit institutions, decentralised financial systems, payment services and credit information bureaus.
SCORE360 in the UEMOA →Source: BCEAO, présentation de la Banque
C
A licensed body that collects borrowers’ credit and payment history from financial institutions and some large billers (water, electricity, telecoms), then provides lenders with creditworthiness reports. In the UMOA, credit bureaus are governed by a uniform law adopted in 2013 and by BCEAO instructions. They reduce information asymmetry between borrowers and lenders.
A summary rating, calculated by a statistical model, that captures a borrower’s level of risk based on its characteristics and behaviour. The better the score, the lower the estimated probability of default. It helps the institution decide faster and more consistently, without replacing the credit analyst’s judgement.
Our scoring platform →Prior, free and informed agreement by which a customer authorises the use or sharing of their data for a specific purpose, such as assessing their creditworthiness. It must state who processes the data, why, for how long and what the customer’s rights are. In the UMOA, sharing data with credit bureaus is subject to consent rules set by the BCEAO.
D
An approach in which a borrower’s score is recalculated regularly as new data comes in, such as payment flows, repayments and activity, instead of being fixed at the time of the loan application. It makes it possible to track risk throughout the life of the loan and to trigger early warnings.
A living score →E
A signal showing that a borrower may struggle to repay before the first missed payment even occurs: falling collections, repeated overdrafts, late supplier payments, a declining score. An early warning system lets the relationship manager act early (contact, rescheduling, support) and limits the growth of non-performing loans.
Solutions for banks →Elements showing why a model gave a borrower a particular score, and in which direction each one weighed: regularity of collections, years in business, debt level, payment incidents. They make the decision understandable for the credit analyst, the customer and the supervisor. A score with explanatory factors is easier to check, challenge and improve.
An explainable score →The amount an institution stands to lose at the moment a borrower defaults, before any recoveries. It includes the drawn balance and, where relevant, part of undrawn facilities such as an authorised overdraft. Combined with probability of default and loss given default, it is used to calculate expected loss.
F
An artificial intelligence method that trains a shared model on data spread across several organisations, without ever pooling that data. Each participant trains the model locally and shares only parameters, often encrypted. SCORE360 uses federated learning: data stays with the institution, which still benefits from a model enriched by collective experience.
How our engine works →Effective access for individuals and businesses to useful, affordable and suitable financial services: accounts, payments, savings, credit and insurance. For SMEs, the main barrier is often credit, due to a lack of history or collateral. Interoperable digital payments and scoring based on alternative data are two ways to broaden it.
A company that uses technology to offer or improve financial services: payments, transfers, digital lending, savings, insurance or tools for institutions. Depending on its activity, a fintech may need a licence or operate in partnership with a bank or licensed institution. In the UEMOA, fintechs play a growing role in SME finance.
Solutions for fintechs →H
An encryption technique that allows certain computations to be performed directly on encrypted data, without decrypting it. Once decrypted, the result is identical to the one obtained on the plain data. In federated learning, it allows participants’ contributions to be added up while keeping them unreadable to the server performing the operation.
Security and encryption →A form of federated learning in which participants hold the same kinds of information about different customers, for example several microfinance institutions tracking comparable variables for their own borrowers. By pooling the learning without exchanging files, each one gets a more robust model, especially for segments where it has little history.
Horizontal or vertical: our modes →I
A payment QR code built on a common format, so that it can be read by the apps of different banks, microfinance institutions or electronic money issuers. The merchant displays a single code, whatever its customer’s provider. In the UEMOA, this payment method is associated with PI-SPI and makes collecting payments easier for small merchants.
Payments and partner aggregators →K
The set of procedures through which a financial institution verifies its customers’ identity, understands their activity and assesses their risk profile, at onboarding and throughout the relationship. KYC meets anti-money laundering and counter-terrorist financing obligations. For an SME, it also covers the business and its managers.
L
The share of the exposure an institution does not recover when a borrower defaults, expressed as a percentage. It depends on guarantees, collateral, recovery costs and timeframes. An LGD of 60% means that 40% of the amount exposed is eventually recovered. Together with PD and EAD, it is used to calculate expected loss.
M
An institution offering local financial services, such as loans, savings and sometimes insurance or transfers, to individuals, artisans and small businesses often excluded from mainstream banking. In the UMOA, most MFIs legally fall under the category of decentralised financial systems (SFD). They are key lenders to micro and small enterprises.
Solutions for MFIs and SFDs →N
A loan whose repayment is at risk, usually because instalments have been unpaid for more than 90 days or because the borrower is deemed unable to pay. Such loans must be classified and provisioned under prudential rules. Their share of the portfolio, the NPL ratio, is one of the main indicators of a bank’s asset quality.
Reducing NPLs: bank solutions →P
A provider that lets a merchant or SME accept several payment methods (cards, mobile money, transfers, QR code) through a single integration. It centralises collections, reconciles them and pays them out to the merchant. For a lender, the collection history structured by an aggregator, shared with the business’s consent, is a valuable source of alternative data on actual activity.
Payments and partner aggregators →Rules governing the collection, use, storage and sharing of information about an identifiable person. In Côte d’Ivoire, the law is enforced by ARTCI. Each UEMOA country has its own legal framework and competent authority. For a lender, this means a clear purpose, data minimisation, security and respect for customers’ rights.
The Interoperable Instant Payment System Platform is the infrastructure developed by the BCEAO to enable instant transfers and payments between customers of different UEMOA institutions: banks, microfinance institutions, electronic money issuers and payment institutions. It notably supports merchant payments via interoperable QR code. For SMEs, its flows form a usable activity history, with their consent.
PI-SPI and payments on the platform →A key microfinance indicator: the share of the outstanding loan portfolio with at least one instalment overdue beyond a given threshold. PAR 30, for example, divides the outstanding balance of loans more than 30 days late by the total gross portfolio. It reveals portfolio quality earlier than the non-performing loan ratio.
Managing PAR: MFI and SFD solutions →An estimate of the likelihood that a borrower will fail to meet its obligations over a given horizon, usually one year. It is derived from a rating or scoring model and expressed as a percentage. PD is used to decide whether to lend, to price the loan and, together with LGD and EAD, to estimate expected losses and provisions.
S
A cryptographic protocol used in federated learning to combine the model updates sent by several participants without the server being able to read any individual contribution. Only the aggregated result is revealed. The technique protects each institution’s own information, even from the operator coordinating the training.
Our technology →A UMOA legal category covering licensed microfinance institutions: savings and credit mutuals and cooperatives, associations and microfinance companies. SFDs are governed by a dedicated law and by BCEAO regulations, with prudential rules suited to their size. They serve a wide customer base of households and small businesses.
Solutions for MFIs and SFDs →U
The West African Monetary Union (UMOA) brings together eight states that share the CFA franc and a common central bank, the BCEAO: Benin, Burkina Faso, Côte d’Ivoire, Guinea-Bissau, Mali, Niger, Senegal and Togo. The West African Economic and Monetary Union (UEMOA), created by the 1994 treaty, groups the same states and adds economic integration.
SCORE360 across the 8 UEMOA countries →V
A form of federated learning in which several organisations hold different information about the same customers. For example, a bank knows an SME’s credit history while a payment aggregator knows its collections. The model combines these complementary views without either party passing its raw data to the other.
Horizontal or vertical: our modes →