Back to PortfolioFinTech

Open Banking Payments Intelligence Platform

Custom anomaly detection model with 98.7% precision — far beyond rules-based fraud flags

🇬🇧 UK£2M+ processed monthly · 98.7% anomaly precision

Overview

PSD2-compliant multi-bank aggregation and payments intelligence platform built for a UK fintech. Processes transactions across 12 UK banks through a single API, with a real-time AI anomaly detection layer that flags suspicious activity before it becomes fraud.

منصة تجميع متعددة البنوك متوافقة مع PSD2 ومنصة ذكاء للمدفوعات بُنيت لشركة تقنية مالية بريطانية. تعالج المعاملات عبر 12 بنكاً بريطانياً من خلال واجهة API واحدة، مع طبقة ذكاء اصطناعي لاكتشاف الشذوذ في الوقت الفعلي.

The Challenge

The client was losing clients to competitors offering unified bank data and smarter fraud detection. Their existing rules-based fraud system generated 40% false positives — flagging legitimate transactions and degrading customer experience. They needed a single API across all major UK banks plus an intelligence layer that actually learned.

كان العميل يخسر عملاءه لصالح منافسين يقدمون بيانات بنكية موحدة وكشفاً أذكى للاحتيال. كان نظام الكشف القائم على القواعد يولد 40% من النتائج الإيجابية الخاطئة.

What We Built

A unified open banking aggregation layer connecting to all 12 major UK banks via their PSD2 APIs, normalising transaction data into a consistent schema, and exposing it through a single authenticated API. On top of this, a real-time intelligence dashboard for the client's finance teams showing spending patterns, anomaly alerts, and merchant category analysis.

طبقة تجميع مصرفي مفتوح موحدة تتصل بجميع البنوك البريطانية الرئيسية الـ12 عبر واجهات PSD2 الخاصة بها، وتوحيد بيانات المعاملات في مخطط متسق.

Tech Stack

PythonFastAPIPostgreSQLAzureTensorFlowReact

Key Outcome

£2M+ processed monthly · 98.7% anomaly precision

Start your project →
Here's what shipped
The Intelligence Dashboard
01

The Intelligence Dashboard

Real-time monitoring across all 12 connected UK banks — volume, anomalies, and live alerts in one view.

  • £2.4M monthly volume, 14,832 transactions tracked
  • 23 anomalies flagged at 98.7% detection accuracy
  • 30-day volume chart with anomaly spike annotations
  • Live alerts feed with severity badges and confidence scores
Explainable Anomaly Detection
02

Explainable Anomaly Detection

Every flagged transaction comes with a full explanation, not a black-box score — a £47,200 transfer gets a 6-axis risk breakdown and four plain-language reasons.

  • 6-axis radar chart comparing flagged vs. normal behavior
  • Four plain-language reasons for every flag
  • Color-coded risk factor breakdown per axis
  • One-click Block, Approve & Learn, or Investigate actions
Unified Transaction Feed
03

Unified Transaction Feed

Every transaction across all 12 banks in one filterable table, with AI categorization and expandable risk detail per row.

  • Search, bank, and category filters plus an anomalies-only toggle
  • Color-coded category badges and risk scores per row
  • Anomalous rows highlighted with expandable detail
  • AI classification confidence and merchant history on expand
Merchant & Subscription Intelligence
04

Merchant & Subscription Intelligence

Spending broken down by category and merchant, with automatic subscription detection most competitors don't offer.

  • Category breakdown across £8,420 in tracked spend
  • Automatic subscription detection — 8 active, £142/month
  • Per-category anomaly counts
  • Trend arrows and top merchants per category
Bank Connections & Compliance
05

Bank Connections & Compliance

All 12 bank connections in one place, with PSD2 and GDPR compliance front and center instead of buried in a settings page.

  • Live sync status across all 12 connected banks
  • One-click reconnect for expired consents
  • AES-256, TLS 1.3, PSD2, and GDPR badges up front
  • Per-bank account count and last-sync time

The AI Layer

The anomaly detection model was trained on 14 months of transaction history per merchant category. Rather than rules ("flag anything over £500"), it learns what normal looks like for each merchant, user, and time pattern — then flags deviations from that baseline. False positive rate dropped from 40% to 1.3%, while true positive detection improved from 67% to 98.7%.

تم تدريب نموذج كشف الشذوذ على 14 شهراً من تاريخ المعاملات لكل فئة تجارية. بدلاً من القواعد الثابتة، يتعلم النموذج ما يبدو طبيعياً لكل تاجر ومستخدم ونمط زمني.

Results

  • £2M+ processed monthly across 12 UK banks
  • Anomaly detection precision: 98.7% (up from 67%)
  • False positive rate reduced from 40% to 1.3%
  • GDPR-first data architecture, SOC 2 certified
  • Delivered in 34 days
- معالجة أكثر من **2 مليون جنيه** شهرياً عبر 12 بنكاً - دقة كشف الشذوذ: **98.7%** - انخفاض معدل الإيجابيات الخاطئة من **40%** إلى **1.3%**