Agentic AI: From vision to reality for asset servicing

Introduction

The conversation around AI in asset servicing is shifting from potential to practical application. Asset servicers no longer ask simply what AI can do; they ask where they can embed it in day-to-day operations to reduce manual effort, strengthen control execution, and address operational challenges that years of incremental process improvement have not resolved.

Individual AI capabilities already deliver measurable benefits across asset servicing. Firms can advance adoption by connecting these capabilities through multi-agent systems, where specialized AI agents link data, checks, and actions across processing workflows instead of operating in isolation.

In this second article in the series, we draw on real-world use cases from a leading European financial institution and Deloitte’s Operate services to explore how firms can evolve standalone AI capabilities into connected multi-agent systems and scale them responsibly.

AI capabilities in asset servicing today

Since the earlier publication, "Agentic AI: Transforming asset servicing," Deloitte has continued to develop and implement AI use cases through both client engagements and managed services operations supporting investment management and asset-servicing activities.

Many processes and roles across transfer agency, fund accounting, management companies, and depositary banking have evolved organically around the limitations of existing systems rather than through intentional design. New regulatory requirements, data security constraints, and technology debt often push the industry toward tactical solutions that meet new demands without disrupting service delivery. By breaking these functions into discrete tasks and control points, teams can identify where to introduce new building blocks and where to restructure existing ones, creating a more solid operational foundation.

The following selection of AI capabilities from our work illustrates how these building blocks function individually and the impact they can have on the business. We designed each AI capability as a self-contained component for specific tasks. Teams can combine and extend these components as operational needs evolve. This modular approach enables teams to connect individual capabilities quickly into broader process chains. In most cases, these capabilities do not replace core servicing platforms or existing compliance systems. Instead, they complement them by handling unstructured inputs, supporting exception management, and improving coordination between tasks that still rely on manual review and handoffs.

Across our work with clients, firms already apply AI in different parts of the asset-servicing operating model, although mainly through discrete capabilities.

^{Source: Deloitte AI capabilities developments. Expand image}

These capabilities already create value individually. Across the market, firms still focus AI adoption in asset servicing on targeted use cases and connected workflows rather than fully autonomous processes. They now need to determine how far they can connect these capabilities across the operating model. Instead of shifting immediately to fully autonomous processes, firms can build progressively on established capabilities by linking inputs, decisions, and outputs where a clear operational case exists and where teams have put the necessary data, controls, and human oversight in place.

From standalone AI capabilities to multi-agent systems

Most AI capabilities in asset servicing still operate in isolation. In many setups, one tool handles a narrow task, produces an output, and passes the process onward. A multi-agent approach, illustrated in Figure 1, connects existing capabilities into a broader workflow so each step builds on the previous one instead of restarting at every handoff. This approach especially benefits processes where core systems already perform individual checks effectively, but teams still rely on fragmented data and manual handoffs for investigation, escalation, and cross-team coordination. One agent extracts data from a document, a second validates it against regulatory rules, a third compares it across sources, and a fourth determines the next action. Each agent operates within a defined scope, while shared context, orchestration, and control points coordinate the workflow. These mechanisms validate outputs, escalate exceptions, and require human approval when necessary.

^{Figure 1: Isolated task execution vs. connected workflow execution
Most AI tools in asset servicing today handle a single task in isolation, producing a useful output but with no shared memory or coordinated handoff. A multi-agent approach changes this by connecting specialized agents into a managed workflow, where context is shared and cases move toward a coordinated outcome. Expand image}

The coordination layer connects the agents. Figure 2 shows how this layer spans existing infrastructure, including legacy systems, data platforms, documents, and external sources, without requiring firms to replace that infrastructure. It maintains continuity across the workflow, routes information and exceptions to the appropriate next step, and preserves human oversight when decisions carry regulatory, client, or reputational weight.

^{Figure 2: The agentic AI layer across existing infrastructure
Rather than replacing existing systems, an agentic AI layer sits across them, drawing from legacy platforms, data warehouses, documents, and external sources. Specialized agents extract, check, compare, and route within a shared structure that maintains workflow orchestration, decision support, and human control points throughout. Expand image}

The next section applies this model to three functional areas: client and investor KYC, investment management onboarding, and depositary oversight. Each area involves multiple steps and handoffs, often with limited shared context across the process.

Multi-agent systems applied to operational processes

1. Client and investor KYC

In asset servicing, client and investor KYC involves several distinct steps: teams gather documentation, extract relevant data, identify missing information, check alignment with policy and regulatory requirements, and coordinate follow-up. In most operating models, different people or systems handle these steps separately, which limits visibility across the full case. A multi-agent approach can add value by allowing agents to pass context forward, from document intake to case resolution, so that each stage builds on the previous one rather than starting fresh at every handoff.

This approach accelerates onboarding, promotes more consistent treatment across cases, and strengthens process control. Reviewers continue to validate exceptions, approve risk-sensitive decisions, and handle complex cases, while the workflow systematically coordinates the steps leading to those decisions.

When transfer agents, management companies, and depositaries perform the same tasks across several services within one financial institution, a coordinated workflow can consolidate duplicate processes into a single run while preserving each entity’s regulatory responsibilities.

The broader direction is consistent with developments across the industry. In its 2026 partnership announcement with Mistral AI, BNP Paribas identified KYC in retail banking as a priority area for generative AI¹, specifically targeting the automation of lower-value tasks and improved workflow orchestration, while preserving specialist validation within strict security and compliance boundaries. While the example comes from retail banking, the underlying logic is relevant to asset servicing: automate routine steps, improve coordination, and keep specialist review where risk is highest.

2. Investment management onboarding

Fund onboarding ranks among the most document-intensive processes in asset servicing. Teams collect fund documentation, extract key information, validate disclosures, assess regulatory alignment, and prepare materials for downstream use. These activities typically span multiple teams and systems, which conduct manual reviews, reconcile information across sources, and coordinate follow-up in parallel but not always in sync. Delays often stem not from any single task but from the effort required to keep information consistent throughout the process.

Agents can share context across the workflow instead of running each activity independently. One agent extracts key fields and disclosures from source documents. Another compares information across prospectuses, KIIDs, KIDs, marketing documents, and translated versions. A third assesses regulatory and internal alignment. A fourth consolidates findings into a structured output that supports remediation and approval decisions. The workflow still proceeds in stages, but each stage uses the context created by the previous one.

Fenergo Fen-AI for Client Lifecycle Management² shows how firms can apply AI agents to complex institutional onboarding. The solution uses document intelligence and automated AML screening to support client lifecycle management while maintaining regulatory controls. It can accelerate onboarding timelines and improve consistency across documentation. Human judgment remains central, but teams no longer need to coordinate and compare the documents that support those decisions entirely by hand.

3. Depositary oversight

Depositary teams perform one of asset servicing’s most control-intensive functions. They reconcile data across funds and counterparties, monitor investment compliance and cash flows, review NAVs and transactions, conduct due diligence, and identify and document irregularities. In many operating models, teams draw on multiple data sources, spreadsheets, and separate control processes to investigate exceptions manually and maintain evidence of completed controls. As fund activity grows in volume and complexity, teams need more time to perform oversight and face greater difficulty scaling while maintaining consistent controls and traceability.

Integrated oversight platforms already create value in this area. Platforms such as Depowise bring multiple oversight activities together and use automated data collection, reconciliation, compliance monitoring, exception management, and workflow capabilities to identify discrepancies and direct teams to cases that require attention. These platforms do not change the depositary’s regulatory role. Instead, they bring supporting data, exceptions, and evidence together so teams can focus their review on the cases that matter most.

ING Luxembourg recently deployed the Depowise platform across³ its UCITS and AIF activities, demonstrating how these capabilities can support a more integrated, technology-enabled oversight model.

These capabilities can automate and standardize oversight, identify discrepancies and exceptions faster, strengthen auditability, and reduce manual effort for depositary teams.

Across all three examples, the same pattern emerges: AI capabilities already support many individual steps. A multi-agent approach changes how those steps connect and enables firms to manage more of the coordination systematically rather than manually.

The foundation for greater scalability

The capabilities described in this article already deliver results through targeted use cases and, increasingly, through connected workflows that combine them. However, firms need the right foundation to make these capabilities work. In regulated environments, they must establish clear governance, strong data protection, defined accountability, and human oversight at critical decision points, including client communications, remediation, and regulatory action. When firms build these elements into the design rather than adding them afterward, they create systems that stakeholders can use with confidence.

Firms also need consistency to scale. When they build use cases on a common technology foundation and follow shared standards, they can connect, govern, and reuse them more easily. This consistency also makes the modular approach described throughout this article practical rather than purely architectural. By breaking existing processes into component parts, evaluating each one, and identifying missing or underserved links, firms can determine step by step where to introduce new capabilities and where to improve coordination among existing ones. AI can support more than document-heavy compliance tasks; over time, it can reshape how firms deliver asset-servicing operations across onboarding, investor servicing, fund administration, and oversight.

Leadership teams now need to balance in-house innovation with strategic partnerships and scale in a controlled, auditable way that aligns with operational reality. Asset servicers increasingly treat this work as a practical transformation agenda that demands both operational experience and technology expertise.