Free 30-minute AI & automation audit with a senior engineer. You keep the notes. Book yours

Use case · FinTech · Europe

Turning open-banking transactions into credit scores, categories and affordability checks

Analytics services that categorise bank transactions and score credit risk for lenders, with country-specific logic, explainable models and queue-based delivery.

Client details are kept confidential. This use case describes the problem and what we built.

The brief

An open-banking data provider serves lenders across several European countries. Lenders use bank-account data, shared by applicants with their consent, to assess creditworthiness and affordability, and they wanted analytics delivered as APIs rather than raw transactions.

We worked with the provider’s analytics team to build and extend its analytics services: transaction categorisation, credit-scoring models, policy rules and customer-specific modules, all delivered through a common API and queue architecture.

The challenge

What needed solving

Raw bank data is noisy

Account-information data arrives as free-text transaction descriptions that differ by bank and country. Before any decision can be made, income, expenses, transfers and risk signals have to be recognised.

Each lender wants its own decisioning

Dozens of lenders, from consumer-credit firms to business lenders, each need their own scorecards, policy rules and response formats on top of shared analytics.

Decisions must be explainable

Credit decisions in regulated markets need reasons, not just a score.

The solution

What we built

  1. Transaction categorisation engine

    A FastAPI service with preprocessing, rule-based and machine-learning categorisation for consumer and corporate accounts. Post-processing applies country-specific logic across several European markets.

  2. Credit and risk models

    XGBoost models, weight-of-evidence binned scorecards, clustering models and fraud signals, with SHAP explanations returned alongside each score.

  3. Policy and affordability rules

    A rules layer runs alongside the models, covering credit policies, affordability checks, income and salary verification, main-account detection, internal-transfer detection and gambling-spend reports.

  4. Per-lender modules on a shared core

    A Flask gateway routes each partner to its own module, which reuses common preprocessing and feature engineering. Requests and responses are mirrored to AWS SQS, with a Lambda consumer archiving them to S3.

  5. Testing and delivery

    An end-to-end test suite that any service pipeline can trigger, plus CodeDeploy-based releases and OpenTelemetry instrumentation.

The outcome

What changed

Faster lender onboarding

New lenders get custom scorecards and rules as modules on a shared analytics core, instead of separate systems.

Explainable outputs

Scores come with feature-level explanations that support credit decisions and review.

Consistent categories across markets

A single categorisation engine handles bank-specific and country-specific quirks in one place.

Technology used

  • Python
  • FastAPI
  • Flask
  • pandas
  • scikit-learn
  • XGBoost
  • SHAP
  • SQLite
  • AWS SQS
  • AWS Lambda
  • AWS S3
  • AWS CodeDeploy
  • Docker
  • OpenTelemetry
  • RSpec

Have a similar problem to solve?

Share a few lines about your project. A senior engineer — not a salesperson — replies within one business day.

Free AI audit Let’s talk