Financial services Usage scenario

Credit scoring from alternative data

Assessing the creditworthiness of people with no banking history from alternative data, with explainable decisions.

Describe your project See the solution

Project overview

This scenario illustrates what we build; it is not a client reference.

This scenario shows how a digital microcredit organisation in West Africa can assess the creditworthiness of people with no banking history. A scoring model relies on alternative data (mobile usage, transactions), with the applicants' consent. The goal: lend to more people while keeping default risk under control.

Client challenges & requirements

A microcredit organisation wants to lend to an underbanked population with no credit history.

Relying only on self-declared information is risky and severely limits how much it can lend.

Main objectives

Develop a scoring model capable of evaluating creditworthiness of applicants without banking history

Use alternative data sources (smartphone usage data, behavioral questionnaires)

Increase the loan acceptance rate without increasing risk

Ensure a fair and interpretable decision model

Project solution

Prisma Soft builds an end-to-end credit scoring engine. With the user's consent, alternative data is collected and turned into predictive variables. A LightGBM model is trained on these variables.
SHAP explains every decision. The model is deployed as a service on AWS and answers in real time.

Indicators to track

Acceptance rate for first-time applicants
Default rate per segment
Decision time
Share of applications handled without manual intervention

Tech stack

Infrastructure: AWS (Lambda, S3, API gateway)
Data Science: Python Pandas LightGBM Scikit-learn
Interpretability: SHAP
Database: PostgreSQL
Deployment: Docker Serverless framework

Frequently asked questions

With explicit user consent, the system collects data such as mobile top-up frequency, data usage, social network stability or transaction habits. This data is anonymised and turned into predictive variables through feature engineering.

Yes. The SHAP library makes every decision transparent. For each credit application, the system shows the main factors that influenced the score, which provides the transparency owed to clients.

Privacy is at the heart of the system. All data is collected with explicit consent, encrypted in transit and at rest, and anonymised before use. Compliance rules are set with the client and its legal counsel.