Home / Insights / AI and Machine Learning
AI and Machine Learning
AI in Banking Software: Use Cases, Risks and How to Implement Responsibly
Banks were early adopters of machine learning and are now adding generative AI across service and operations. This guide maps the main use cases, the risks specific to financial services and how to implement AI within model risk and regulatory expectations.
TopicsAI and Machine LearningGenerative AI and LLMsBusiness Operations and Integration
Key takeaways
- Banking AI spans customer onboarding, fraud and AML, credit decisions, customer service, operations, and compliance and reporting.
- Generative AI is most useful today in service, operations and knowledge work; predictive models remain central in fraud and credit.
- Financial services require model risk management, explainability for decisions affecting customers, fairness testing and strong data protection.
- In the US, SR 26-2 replaced SR 11-7 in April 2026 as the main model risk guidance for banking organisations.
- Start with internal, assistive uses and expand to customer-facing and decision-making uses as controls mature.
On this page
Banks have used statistical models for decades in credit scoring and fraud detection. Machine learning and, more recently, generative AI have expanded what is possible: faster onboarding, better fraud detection, more responsive service and less manual work in operations. Because banking is heavily regulated and trust-dependent, implementation must be careful.
Use cases
| Area | AI applications | Typical approach |
|---|---|---|
| — | — | — |
| Onboarding and KYC | Document extraction, identity verification, risk scoring | Document AI, biometrics, ML scoring |
| Fraud and AML | Transaction scoring, network analysis, alert investigation | ML models, graph analytics, agents for investigation |
| Credit | Underwriting, early warning of default, collections prioritisation | Interpretable ML with explanations |
| Customer service | Assistants for customers and staff, call summarisation | Generative AI grounded in policies |
| Operations | Reconciliation, exception handling, document processing | ML and generative AI with human review |
| Compliance | Regulatory change analysis, policy search, report drafting | Generative AI with expert review |
| Risk and treasury | Forecasting, stress-testing support | Statistical and ML models |
Our guide to AI agents for fraud detection covers one area in depth.
Banking-specific risks
- Model risk: errors or misuse of models can cause financial loss and customer harm.
- Fair lending and discrimination: models must not unjustifiably disadvantage protected groups.
- Explainability: customers and regulators may require reasons for adverse decisions — see our guide to explainable AI.
- Data protection and secrecy: customer financial data is highly sensitive.
- Third-party risk: reliance on AI vendors and model providers must be managed like other outsourcing.
- Generative AI risks: hallucinated answers, prompt injection and inappropriate advice in customer channels.
Governance and model risk management
In the US, the Federal Reserve, OCC and FDIC issued SR 26-2, Revised Guidance on Model Risk Management, in April 2026, replacing SR 11-7. It emphasises a risk-based approach tailored to each organisation’s model risk profile. Practices typically include:
- an inventory of models and AI systems with owners and risk tiers;
- documentation of purpose, data, methodology and limitations;
- independent validation before use and periodically after;
- ongoing monitoring of performance, drift and fairness;
- change management for retraining and prompt or model updates;
- clear accountability and escalation.
Other jurisdictions have their own supervisory expectations, and AI-specific rules such as the EU AI Act classify some banking uses, like creditworthiness assessment, as high-risk.
Implementation approach
- Start internal and assistive: knowledge assistants for staff, document processing with review, call summarisation.
- Build the controls: inventory, evaluation sets, monitoring, documentation templates and approval workflows.
- Move to decision support: recommendations for fraud analysts, credit officers and operations staff, with humans deciding.
- Automate selectively: low-risk decisions within limits, with monitoring and audit.
- Customer-facing AI last, with grounding, guardrails and escalation.
Frequently asked questions
Can banks use large language models with customer data?
Yes, with appropriate contracts, data protection, access control, model risk governance and, often, private deployment options.
Should credit decisions use deep learning?
Many banks prefer interpretable models for credit because of explanation and fairness requirements. Complex models need robust explanation and validation.
What is the quickest win?
Internal knowledge assistants and document processing with human review often deliver value quickly with manageable risk.
How do AI vendors fit into model risk management?
Vendor models should be inventoried, validated as far as possible and monitored, with contractual rights to information.
Sources and methodology
Facts in this article were checked against the sources below on 6 November 2026. Directory data (ratings, review counts, rate bands and team sizes) changes over time and is shown as published on the date accessed. Illustrative examples are labelled as such and are not quotes.
- Federal Reserve: SR 26-2, Revised Guidance on Model Risk Management (17 April 2026) — accessed 6 November 2026. Joint guidance replacing SR 11-7; risk-based approach
- Gibson Dunn: EU AI Act Omnibus Agreement — accessed 6 November 2026. Timeline of high-risk obligations under the EU AI Act
Related reading
Continue your research
Exploring AI in banking or fintech?
Share the use case and constraints. We will help you design models, controls and documentation that satisfy risk and compliance teams.