Artificial Intelligence from a Depositary Perspective

Opportunities for AI within the Depositary Control Function

AI can support the provision of depositary services in several meaningful ways, particularly in areas that are data-heavy, repetitive, document-intensive or pattern-based and where large volumes of information must be reviewed, exceptions assessed and oversight evidence documented. Some of the more relevant opportunities can be considered through the lens of Predictive and Agentic AI.

Predictive AI is designed to estimate likely outcomes based on historical data. Predictive AI can improve the effectiveness of controls by helping teams’ direct attention to the areas of greatest risk. Predictive models can support identifying unusual cash movements, recurring settlement failures and anomalies in subscription and redemption activity.

This supports a more risk-led surveillance model. Predictive AI does not replace the need for full control accountability, but it can help allocate time intelligently by highlighting which exceptions or patterns are most likely to indicate operational error, control weakness or potential investor harm. It may also help identify thematic issues over time, such as repeated breaks/issues linked to specific counterparties, instruments, markets or service providers.

Agentic AI offers significant potential to improve workflow efficiency. Depositary teams routinely manage reconciliations, transaction feeds, cash and asset records, exception reports and supporting documentation from multiple internal and external sources. Agentic AI can be used to collect required data, test completeness, initiate reconciliations, route exceptions to the appropriate reviewer, trigger follow-up actions and maintain an auditable record of each step performed.

In practice, this could reduce turnaround times and allow depositary teams to focus more attention on higher-risk issues and judgment-based review.

Governance Requirements for AI Implementation

AI should be deployed only within clearly defined parameters, subject to robust controls, clear limitations and effective human oversight. Human expertise, judgment, experience and professional challenge must therefore remain central to the discharge of depositary responsibilities, with AI serving only as an aid to more efficient process execution rather than a substitute for substantive oversight.

Clear Ownership and Defined Purpose

Every AI use case should have a defined business objective, clear scope, accountable senior ownership and a classification of whether the use is advisory, semi-automated or embedded within a control workflow. Without this, the function risks introducing tools into core controls without a clear understanding of what they are intended to achieve or how they affect the wider control framework.

The expertise required includes control ownership, operational understanding of depositary activities and sufficient technical literacy to challenge proposed use cases and implementation decisions.

Transparency, Validation and Explainability

Model documentation should set out intended use, methodology, data lineage, assumptions, limitations, fallback procedures and approval requirements. Independent review should test whether the model performs as intended and whether outputs are sufficiently stable, interpretable and reproducible.

If this governance is not applied, it may create risks, including over-reliance on outputs, outputs that cannot be independently verified, explained or challenged. The expertise required includes model validation capability, risk and control knowledge and individuals able to interpret technical outputs in the context of depositary obligations.

Strong Data Governance

Depositary controls depend on a significant amount of data, which needs to be complete, accurate, timely and well-governed. AI models trained or operated on incomplete, stale, poorly mapped or weakly controlled data will generate unreliable results at scale.

Where data governance is weak, automation can amplify rather than reduce risk. The expertise required includes data management, reconciliation discipline, source system knowledge and information security capability.

Ongoing Monitoring and Escalation

AI performance must be monitored for drift, false positives, false negatives, unexplained output changes and increasing override frequency. Human escalation thresholds must be explicit, and material exceptions must trigger investigation rather than silent acceptance.

Without this discipline, teams may develop false confidence in automated outputs and fail to identify when models are degrading or missing emerging risks. This requires operational control expertise, monitoring capability and managers prepared to intervene where model behavior changes.

Harnessing the Potential of AI for Depositary Services

As investment structures, operating models and data volumes continue to evolve, depositaries will need to oversee increasingly complex environments while maintaining the same standards of investor protection, governance and control. AI has the potential to support this transformation by helping depositaries manage scale more effectively, identify emerging risks more quickly and give attention to issues that matter most.

The opportunity is not to redefine the role of the depositary but to strengthen how its role is performed. Institutions that combine technology with strong governance, high-quality data and experienced oversight are likely to be the best positioned to deliver efficient, scalable and resilient depositary services in the future.

Ultimately, AI may enhance depositary service delivery, but responsibilities remain unchanged. The judgment, challenge and accountability that underpin effective depositary oversight will continue to rest with human oversight, supported by technology rather than being replaced by it.