Understanding an SME’s credit risk better requires more information than the file presented at the counter. There are several ways to get there. They are not mutually exclusive, but they do not offer the same guarantees.
The central database
All data is gathered in one place to train a model. It is simple to design, but it concentrates risk: a breach or misuse exposes every institution’s clients. It is also hard to accept for an institution that rightly considers its portfolio a strategic asset.
The credit information bureau
Credit information bureaux, regulated in the UMOA and licensed by the BCEAO, such as Creditinfo West Africa, share borrowers’ credit history. They are essential, but they mostly describe credit history. Many SMEs have little or none: their business is real, but it does not appear in a loan history.
Federated learning
The model is trained where the data lives, then only encrypted parameters are combined. The result is a model that learns from the whole network, without any central database of client data ever existing. It can also draw on alternative data from SMEs’ real commercial activity.
What PI-SPI changes
For a long time, many SMEs’ activity escaped any measurement: cash payments, light bookkeeping, no usable trail. With the BCEAO’s PI-SPI, instant collections and payments become dated, structured flows. With the SME’s consent, they measure its real revenue, regularity and seasonality: exactly what is missing to assess a business with no credit history.
SME credit risk: quick comparison
- Confidentiality: a central database concentrates data; federated learning leaves it with each institution.
- Coverage: the credit bureau covers borrowers with a history; federated learning also helps assess poorly documented SMEs.
- Data: federated learning can draw on PI-SPI flows, which reflect real activity day by day.
- Freshness: a federated model is retrained round after round and tracks changes in activity.
- Governance: each institution keeps control of its data and its decision.
Key takeaway
For a UEMOA institution, the right strategy combines tools: the credit information bureau for history, guarantee schemes such as the GARI Fund or the African Guarantee Fund to share risk, and federated scoring, fed by PI-SPI flows, to assess SMEs’ real activity without ever exposing client data. That is how credit can reach the viable businesses that are shut out today.
FAQ
How do you assess the credit risk of an SME with no credit history?
By looking at its real activity. With the SME’s consent, its PI-SPI instant payment flows measure its revenue, regularity and seasonality. A federated scoring model uses this alternative data inside each institution’s own environment, without ever centralising it. This makes viable businesses with no loan history assessable.
Does Federated Learning replace the credit information bureau?
No, it complements it. The credit information bureau, such as Creditinfo West Africa, shares borrowers’ credit history. Federated scoring additionally helps assess viable SMEs that do not yet have a credit history, based on their real commercial activity. The two sources work together rather than compete.
Why not simply pool data in a central database?
A central database concentrates risk: a breach or misuse would expose the clients of every institution. It is also hard to accept for an institution that considers its portfolio a strategic asset. Federated learning builds a shared model without any central database of client data ever existing.
What strategy helps control SME credit risk?
Combine the tools: the credit information bureau for history, guarantee funds such as the GARI Fund or the African Guarantee Fund to share risk, and federated scoring fed by PI-SPI flows to assess SMEs’ real activity without exposing client data.