Completed Project Case Study: Investigating Multimodal Knowledge Graphs as Digital Twin Infrastructures for Financial Operations Risks  

The Challenge 

Operational risk costs the financial sector billions each year. Fragmented systems, siloed data, cyber threats, and growing regulatory scrutiny make it increasingly difficult for institutions to gain a complete and reliable view of their operational risk landscape. 

Start-up IRMAI is developing an AI-driven “digital twin” of financial operations, a dynamic model designed to simulate and monitor risk in real time. However, a major bottleneck quickly became clear: AI systems are only as strong as the data infrastructure underpinning them. Without a structured, trustworthy, and interoperable data foundation, risk modelling remains limited and reactive. 

This project explored whether Multimodal Knowledge Graphs (MMKGs) could provide that missing infrastructure. 

The Project 

Led by Dr Anelia Kurteva at the University of Birmingham, the project investigated how ontologies and multimodal knowledge graphs could serve as a machine-readable foundation for modelling financial operational risks.  

In simple terms, the team asked: 
Can we organise complex financial risk information, from reports, datasets, images, and structured records, into a unified, interconnected digital system that AI can reliably reason over? 

To answer this, the team: 

• Designed a modular financial risk ontology extending the Financial Industry Business Ontology (FIBO), aligned with ISO 31000 risk standards 

• Collected and processed multimodal public datasets (text, tabular data, and synthetic examples) 

• Built and validated a proof-of-concept Multimodal Knowledge Graph using GraphDB 

• Developed a data processing pipeline integrating generative AI for multimodal feature extraction 

The resulting prototype MMKG enables structured querying of interconnected financial operations risk data via SPARQL endpoints. 

All source code has been made publicly available to support open science. 

Key Outcomes 

The project delivered: 

• A publicly documented financial risk ontology with over 150 classes and 200 relationships 

• A working prototype Multimodal Knowledge Graph modelling operational risks (with a focus on mortgage-related risks) 

• A multimodal data processing pipeline integrating text and image data 

• Public documentation and APIs enabling reproducibility and further development  

The findings demonstrate that MMKGs can effectively support the structured integration, contextualisation, and querying of complex financial risk data. 

While further work is required to build full predictive digital twins, the project provides strong evidence that MMKGs are a viable infrastructure layer for AI-driven operational risk systems. 

Collaboration 

Collaboration with IRMAI was central to grounding the research in real operational challenges. IRMAI provided domain expertise, practical use cases, and insight into the limitations of existing risk infrastructures  

The project also surfaced important lessons about academic–industry collaboration. Start-up environments move quickly and prioritise immediate functional outcomes, whereas academic research requires structured validation and methodological rigour. Regular bi-weekly meetings were essential for maintaining alignment and managing evolving expectations. 

A key takeaway was the importance of clearly defining mutual value at the outset of short-term collaborations, particularly when working with fast-scaling fintech companies. 

The Impact 

For IRMAI, the project delivered foundational digital infrastructure to support their long-term vision of an intelligent operational risk twin. The ontology and MMKG prototype provide a structured backbone for building future predictive algorithms. 

In academia, the project advanced research into multimodal knowledge graphs for financial risk management, an underexplored area with significant commercial relevance. It also strengthened capacity in data-centric AI, emphasising that high-quality, interoperable data architectures are just as important as model optimisation. 

More broadly, the work contributes to improving transparency and interoperability in financial risk data management. By promoting FAIR (Findable, Accessible, Interoperable, Reusable) data principles, the project supports more responsible AI deployment in financial services. 

Going Forward 

Future work will focus on scaling the multimodal knowledge graph (MMKG) beyond the current proof-of-concept stage and further developing risk prediction algorithms that leverage the patterns identified within the graph structure. The team will also explore opportunities for larger funding initiatives to support the development of an industry-ready solution. In parallel, efforts will be made to expand dissemination beyond academic audiences in order to attract additional collaborators from industry and the wider fintech ecosystem.  

Overall, the project has highlighted the need for sustained investment and broader collaboration to fully operationalise digital twin infrastructure for financial risk management. 

UKFin+ Role 

UKFin+ Agile funding provided critical early-stage support to test an ambitious and technically complex idea. The relatively short funding window enabled rapid prototyping while fostering meaningful collaboration between academia and a fintech start-up. 

By supporting exploratory, data-centric research at the intersection of AI, knowledge engineering, and financial risk, UKFin+ helped lay the groundwork for more advanced digital twin systems capable of transforming operational risk management in financial services.