The adoption of AI within banks reveals a complex relationship between technology and risk. In banking, operational risk refers to financial losses resulting from inadequate or failed internal processes, poor employee performance, or a bank's inability to manage disruption caused by external actors.
While AI integration can deliver efficiency gains in areas such as customer service, fraud detection, and trading risk management, it also introduces new sources of operational risk. A study published by the Federal Reserve Bank of Richmond¹ quantifies the impact of AI investments on operational losses at large U.S. banks.
How AI Can Introduce Operational Risk
The Fed study identified four main channels through which AI can generate operational risk:
- Model performance issues — stemming from biased training data or models that fail to adapt to changing market conditions
- Increased system dependency — creating points of failure through greater cybersecurity vulnerability
- Reduced human oversight — which can delay detection of novel types of fraud
- Short-term transition losses — as organizations adapt to new AI systems and processes
Study Methodology
The Fed study drew on data from 36 U.S. banks each holding assets exceeding $100 billion. On average, these banks incurred operational losses of $181 million per quarter — equivalent to approximately 0.04% of total assets.
An econometric model was developed to explore the relationship between a bank's AI investment and its operational risk exposure, after controlling for key environmental factors. AI investment was measured by the share of AI-skilled employees, identified through individual worker résumés and job postings relative to total headcount.²
The model controlled for the following bank-level characteristics:
- Size — larger banks face higher operational risk due to greater transaction volumes and organizational complexity
- Asset growth — faster-growing banks tend to experience higher operational losses
- Business model — banks focused on deposit-taking and lending carry different risk profiles than those deriving income from trading and investment banking
- Risk management discipline — the strength and independence of a bank's internal risk management function
Key Findings
Using supervisory data on operational losses combined with company-level data on AI-skilled human capital, the study found that banks with higher AI investment are exposed to greater operational risk and suffer larger losses per dollar of total assets.
The three categories of operational risk most affected by AI adoption were:
- Clients, Products and Business Practices — challenges in ensuring AI-driven services meet regulatory requirements and customer expectations
- Business Disruption and Systems Failure — increased vulnerability to technical failures and outages
- External Fraud — new attack vectors created by AI systems and digital infrastructure
These losses arise not only from frequent, smaller incidents but also from severe tail-risk events. Importantly, the risk-amplifying effect of AI is significantly more pronounced at banks with weaker risk management practices — those with less robust internal controls suffered disproportionately higher AI-related operational losses.
How Teradata Can Help
With Teradata AI Studio, you can accelerate AI/ML innovation while helping to reduce operational risk.
Teradata AI Studio is a unified environment for building, deploying, and scaling high-value AI across the enterprise — covering predictive, generative, and agent-driven use cases.
A key component is ModelOps: a streamlined approach to delivering AI/ML models into production. ModelOps features built-in model lifecycle management and supports responsible AI governance through compliance tracking, end-to-end traceability, and full visibility into model development and deployment.
Explore AI Studio to discover how a unified approach to AI can help mitigate operational risk for your bank.
1. Ping McLemore and Atanas Mihov, “AI and Operational Losses: Evidence from U.S. Bank Holding Companies,” 2024, https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5544858.
2. Ping McLemore and Atanas Mihov, “AI and Operational Losses: Evidence from U.S. Bank Holding Companies,” 2024, https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5544858.