WeBank, China’s first fully digital bank, founded in 2014 with Tencent as its lead shareholder, has become a global reference for unsecured SME lending. Its online working-capital loan Weiyedai, launched in 2017, has received applications from more than five million micro, small and medium-sized enterprises (PR Newswire). Its secret is not a better gut feel for risk: it is a better way to use other parties’ data, without ever collecting it.
Lending on data, not collateral
WeBank has no branches and asks for no property collateral. It assesses businesses using third-party data authenticated and authorised by the client: tax authorities, company registry, credit, courts, electricity and social security (WeBank). These signals describe an SME’s real activity far better than a balance sheet it often does not have.
Vertical Federated Learning, the core of the model
Most presentations of Federated Learning describe similar institutions training one model on different clients: that is horizontal Federated Learning. WeBank mainly developed the other form, vertical Federated Learning: players holding different information about the same businesses build a risk model together.
Its founding case, in 2019, brought together WeBank, which knows loan repayments, and the national electronic invoice centre, which knows the same businesses’ invoices (Intel and WeBank). The process has three steps:
- Matching common businesses, without anyone revealing their client list.
- Training on encrypted computations: each party computes on its side, and only encrypted results travel, unreadable even to whoever combines them.
- A shared model that is more accurate than the bank’s alone, with no raw data ever exchanged (scientific review).
An open, standardised technology
WeBank open-sourced its Federated Learning platform and handed it to the Linux Foundation. It also led IEEE 3652.1, the international architecture guide for Federated Learning (IEEE). A trust strategy: banks and regulators adopt a technology they can audit more readily.
What it changes for the UEMOA
Equivalents of WeBank’s data exist in our region, and some are finally becoming usable:
- Payments: BCEAO PI-SPI flows, received through partner fintechs and payment aggregators, with the SME’s consent.
- Tax: the tax administration and its invoices.
- Energy and social security: electricity companies and social security funds.
- Credit: credit information bureaux such as Creditinfo West Africa.
For a bank or an MFI, the benefit is direct: assessing an SME with no credit history from its real activity, without that data ever leaving its holders. It is the same logic described in our article on financing African SMEs.
Adapt, don’t copy
WeBank relies on Tencent’s ecosystem, which nobody has in the UEMOA. An African model’s strength will therefore come from data partnerships, built within the UMOA credit information bureau framework and personal data protection rules. And where WeBank lends itself, SCORE360 remains a Tech-Enabler: the lending decision always belongs to the institution.
Our approach for the pilot
SCORE360 combines both forms of Federated Learning. For the pilot phase, planned for November 2026 with one bank and two microfinance institutions:
- Horizontal: the bank and the two MFIs train the same model, each on its own files.
- Vertical: each institution is paired with e-SIGNAL, our first data partner, which brings SMEs’ digital signals: orders and messages, customer interactions, online reputation.
Other sources, such as tax, energy or social security data, will follow. Explore the architecture on our Technology page, or request a demo to see what this approach can bring to your institution.
FAQ
What is vertical Federated Learning?
It is a form of federated learning where parties holding different information about the same clients build a model together. For example, a bank that knows repayments and a player that knows payments or invoices, without ever exchanging their raw data.
How does WeBank lend to SMEs without collateral?
WeBank assesses businesses using data authenticated and authorised by the client: tax, company registry, credit, courts, electricity and social security. Federated Learning lets it use some of these sources without centralising them.
How does SCORE360 apply vertical Federated Learning?
For the pilot phase, each partner institution is paired with e-SIGNAL, SCORE360’s first data partner, which brings SMEs’ digital signals. Common businesses are matched without revealing lists, and only encrypted computations travel, with the SME’s consent.
Can the WeBank model work in the UEMOA?
Yes, if it is adapted. PI-SPI flows, tax, energy and social security data offer equivalents to WeBank’s sources. The difference lies in the data partnerships to build and in the BCEAO and ARTCI regulatory framework.