Kaspi Bank. Anti-Fraud Platform Audit and Optimization

Kaspi Bank

Working with a major retail bank, I conducted an independent audit of its production anti-fraud platform built on IBM Safer Payments, which processed high volumes of transactions across multiple channels.

The goal was not to "patch individual bugs" but to deliver a comprehensive assessment of the anti-fraud architecture's maturity, its resilience under load, and its alignment with the bank's business objectives.

Context and Challenges

The bank's anti-fraud environment processed transactions across several channels: cards, digital banking, transfers, ATMs, and acquiring. As the customer base and digital services grew, the system began to face a number of problems typical of large-scale fintech platforms:

  • performance degradation during peak periods;
  • inefficient use of computing resources;
  • difficulties scaling and maintaining the model;
  • growing operational risks when updating and evolving the system.

The task was to understand the root causes rather than fight the symptoms.

What Was Done

I performed a full review of the anti-fraud environment, from the logical architecture to the principles of transaction processing and decision-making. The analysis focused on:

  • the overall structure of the anti-fraud model and its scalability;
  • how data flows were organized between system components;
  • the principles of transaction and customer segmentation;
  • the system's resilience to load and failures;
  • how well the architecture matched real business scenarios.

Special attention was paid to how architectural decisions affected the speed of anti-fraud response and detection quality, as well as the bank's operating costs.

Key Findings

The analysis made it clear that most of the problems were not tied to individual rules or models but lay at the level of the system's architecture and configuration:

  • a single "monolithic" processing logic led to excessive load and inefficient use of resources;
  • there was no clear separation by business entities (customers, terminals, channels);
  • the system performed a large number of non-essential computations;
  • model updates and changes came with elevated operational risks.

Proposed Solution

I proposed an architectural transformation of the anti-fraud model, moving to a more flexible and scalable structure:

  • logical separation of processing by key business entities;
  • optimization of data flows and processing priorities;
  • reduced load by eliminating redundant operations;
  • improved fault tolerance and manageability of the system.

This approach would not only stabilize the platform's current operation but also lay the groundwork for future growth: onboarding new channels, developing behavioral models, and introducing advanced analytics.

Business Impact

Implementing the recommendations would enable the bank to:

  • make the anti-fraud environment more stable during peak loads;
  • cut infrastructure and operating costs;
  • deploy new fraud detection rules and models faster;
  • reduce risks when updating and scaling the system;
  • improve the quality of cross-channel transaction monitoring.

Summary

This project showed how an architectural, systems-level approach to anti-fraud delivers greater impact than local optimization of rules or models.

My role was to bring together expertise in anti-fraud, high-load systems, and processing to develop solutions that work not only at the technology level but also at the business level.

← Back to projects