Hierarchical Fallback Architecture for High Risk Online Machine Learning Inference

TL;DR

Open Banking–driven online ML incurs robustness challenges due to external data dependencies and potential outages. The authors propose a hierarchical fallback architecture with a main model and staged fallbacks to address data absence, data corruption, model downtime, and latency spikes, along with a client-side fallback for extreme outages; they also discuss selective redundancy guided by risk (e.g., higher-value transactions). The approach is validated conceptually and via a real-world case in Trustly’s payments risk evaluation, highlighting coverage trade-offs and performance gaps between main and fallback models. The work advances practical robustness for real-time financial ML systems and motivates simulation- and optimization-based evaluation of retry policies and fallback selection to balance reliability and efficiency.

Abstract

Open Banking powered machine learning applications require novel robustness approaches to deal with challenging stress and failure scenarios. In this paper we propose an hierarchical fallback architecture for improving robustness in high risk machine learning applications with a focus in the financial domain. We define generic failure scenarios often found in online inference that depend on external data providers and we describe in detail how to apply the hierarchical fallback architecture to address them. Finally, we offer a real world example of its applicability in the industry for near-real time transactional fraud risk evaluation using Open Banking data and under extreme stress scenarios.

Figures (2)

• Figure 1: Hierarchical Fallback Architecture.
• Figure 2: 99th percentile model latency and number of requests on a 5 minutes aggregation window under the effect of a NFL game weekend.