A research company,
building production infrastructure.
Mansa Research Technologies is a Paris- and Dakar-based credit intelligence company. We build the underwriting infrastructure that African banks and microfinance institutions use to decide on SMB and consumer credit.
We founded Mansa to help lending institutions transform their decision-making processes, with the specific realities of the African context in mind. We bring artificial intelligence and machine learning algorithms together to build innovative technologies, adapted to the markets we serve, that strengthen our partners' decisions.
Our goal is twofold: to drastically reduce manual work for greater efficiency, and to lower the default rate through scoring models calibrated with rigour on the data and dynamics specific to the continent.
Two products. One pipeline.
Castel handles the loan from application to disbursement. Almami takes over once the loan is on the books, monitoring repayment, surfacing early signals, and managing the portfolio through to maturity.
From application intake to the analyst's final decision, on a single platform. Instant pre-scoring, document collection, AI-led instruction, scoring, and committee review, with every step traceable and explainable.
- Multi-signal scoring built for African data
- Custom models trained on your customer data
- Branded public application portal in EN, FR, AR, Wolof
- Full audit trail and feature attribution per decision
- Configurable loan products and decision rules
The watchful second half. Once the loan is booked, Almami monitors repayment behaviour, surfaces early warning signals, and triggers interventions before a loan turns non-performing. Currently in development, available for design partners.
- Systematic credit portfolio management
- Core banking integration with external data sources
- Behavioural signal monitoring after disbursement
- Automated reporting aligned with international standards
How an underwriting decision is made.
Every Mansa decision passes through four steps. Each one is logged. Each one is auditable. The model never decides alone. It produces a probability, and a risk officer reads it.
Application
Submitted through a branded public form your bank operates under its own identity. Available in EN, FR, AR, and Wolof. KYC, document upload and consent capture happen here, and the applicant never sees Mansa.
Ingestion
Castel ingests data from external providers that describe the applicant: bank statements, mobile money operators, supplier records, BCEAO RCCM. Inputs are normalised, deduplicated, and timestamped against a single ledger.
Scoring
A model tailored to your portfolio produces a calibrated probability of default. Each output is accompanied by feature attribution, showing which signals contributed how much, in plain language and as numerical weights. The score is not a verdict.
Decision
A risk officer reviews. They approve, decline, or override. Every override is logged with a reason, and feeds back into model evaluation. Regulators receive a complete audit trail per decision, on request.
At the intersection of three disciplines.
Mansa is built by a team trained in quantitative finance, machine learning engineering and the African credit market: backgrounds from ENSTA Paris, École Polytechnique and Sorbonne Université, alongside experienced profiles from major banks and MFIs, where they designed and industrialised artificial intelligence models within the most demanding prudential frameworks.
If you underwrite SMB and consumer credit on the continent, we'd like to meet you.
A focused walkthrough with your risk team, on a few representative dossiers from your loan book. We'd like to understand how you decide today, and show how Castel could fit into that.