Executive Summary
Artificial intelligence (AI) is fundamentally changing how financial institutions operate, compete, and serve clients. Within fund administration and investor servicing, the conversation is moving beyond experimentation with generative AI toward a more consequential question: How can financial institutions responsibly embed AI into the processes that safeguard investor assets, protect confidential information, support regulatory compliance, and sustain market confidence?
This distinction matters. Fund and investor servicing is not an ordinary customer-service environment. Behind every subscription, redemption, transfer, capital call, distribution, investor communication, and ownership record lies an interconnected set of financial, legal, operational, and fiduciary obligations. An error may affect economic entitlements, regulatory obligations, fund liquidity, or confidence in the institution responsible for administering the investment.
The commercial opportunities are nevertheless substantial. AI can accelerate document interpretation, strengthen exception management, improve anti-money-laundering investigations, personalize investor communications, and enable servicing teams to devote greater attention to complex client needs.
Recent evidence illustrates both the opportunity and the challenge. McKinsey’s 2026 global survey found that 80% of respondents reported improved individual productivity from AI, yet only 37% reported a positive contribution to organizational earnings before interest and taxes (Tinkoff et al., 2026). The gap between individual efficiency and enterprise value is particularly relevant to fund servicing, where improvements must translate into measurable service quality, stronger controls, lower operational risk, and better investor outcomes.
Our central argument is that responsible AI must become an integral feature of the fund-servicing operating model, rather than a separate compliance initiative.
The institutions that lead this transformation will not necessarily be those that automate the greatest number of activities. They will be those that establish the clearest boundaries between machine intelligence and accountable human judgment, retain control over authoritative records, and demonstrate that AI-driven efficiency strengthens rather than compromises investor protection.
Ultimately, the competitive differentiator will be institutional trust.
1. Why responsible AI is now a strategic imperative
AI adoption across financial services has reached a point where responsible implementation is no longer an optional consideration. The Bank of England and Financial Conduct Authority (2024) reported that 75% of surveyed UK financial-services firms were using AI, with a further 10% planning adoption over the subsequent three years. Foundation models accounted for 17% of reported AI use cases, while approximately one-third involved third-party implementations. The research also identified an important governance concern: although 84% of respondents had an individual accountable for AI, only 34% reported a complete understanding of the AI systems they used. These findings expose a potentially widening gap between technological adoption and institutional understanding.
More recent evidence reinforces that concern. Stanford’s 2026 AI Index Report recorded 362 documented AI incidents during 2025, compared with 233 in 2024 (Stanford Institute for Human-Centered Artificial Intelligence, 2026). Although these incidents span multiple industries and cannot be treated as a measure of financial-sector failure, the trend illustrates why AI governance must mature alongside adoption. For fund administrators, transfer agents, asset managers, and other servicing providers, three developments make this particularly urgent.
First, investor servicing is becoming increasingly data-intensive and interconnected. Institutional and private-market investors expect timely access to portfolio information, transaction histories, capital-account statements, regulatory documentation, and personalized reporting. Meeting these expectations requires information to move across systems, entities, service providers, and jurisdictions.
Second, operational complexity is increasing. Alternative investment structures, multiple share classes, jurisdiction-specific restrictions, side-letter arrangements, tax classifications, and evolving regulatory requirements create servicing environments in which exceptions can be more consequential than routine transactions.
Third, AI is moving from generating information toward initiating actions. Agentic AI systems can potentially retrieve documents, update workflow statuses, orchestrate tasks, and interact with enterprise applications. While these capabilities promise greater productivity, they also create new questions about decision rights, authorization, and accountability.
The Financial Stability Board (2024, 2025) has highlighted third-party concentration, cybersecurity, data quality, governance, and model risk as important vulnerabilities associated with financial-sector AI adoption. For servicing institutions, these are operational realities rather than abstract technological concerns.
The strategic challenge is therefore to develop an AI operating model that delivers innovation without weakening the control disciplines on which the investment industry depends.
2. Understanding where AI creates real value across the investor lifecycle
Responsible AI begins with a precise understanding of the business processes being transformed. Fund and investor servicing encompasses activities extending from investor onboarding through transaction processing, ownership recordkeeping, ongoing reporting, and eventual exit. These processes vary considerably in terms of complexity, regulatory exposure, data sensitivity, and consequences of failure. An effective AI strategy must reflect those differences.
Investor onboarding and due diligence
Investor onboarding frequently involves collecting, interpreting, validating, and reconciling information from identification documents, subscription agreements, corporate ownership structures, tax declarations, and compliance questionnaires.
AI can support these activities through document classification, information extraction, completeness checks, entity-resolution assistance, and identification of inconsistencies requiring investigation. For example, an AI-enabled workflow might identify a mismatch between a beneficial ownership declaration and supporting corporate documentation, then route the discrepancy to an appropriate compliance specialist. The value lies in improving the speed, consistency, and quality of human review.
Subscriptions, redemptions, and transfer processing
Transaction processing is another promising application.
AI can help classify investor instructions, identify missing documentation, detect unusual patterns, prioritize operational exceptions, and support reconciliation of cash movements against dealing records. However, transaction validity must remain anchored to authoritative systems, verified instructions, contractual terms, and established approval requirements.
Consider an investor submitting a redemption request close to a dealing deadline. AI may assist in interpreting the instruction and identifying the applicable deadline. It should not independently override a cutoff, disregard a liquidity restriction, or approve an exception merely because historical cases appear similar. A transaction that appears operationally routine may carry legal and economic consequences that cannot be determined through statistical pattern recognition alone.
Investor communications and relationship servicing
Generative AI offers substantial potential in drafting responses to investor inquiries, explaining operational procedures, summarizing account activity, and helping service professionals retrieve relevant information.
Academic evidence supports the potential benefits. Brynjolfsson et al. (2025), studying 5,172 customer-support agents, found that access to generative AI assistance increased productivity by approximately 15%, measured by issues resolved per hour.
The benefits were particularly pronounced among less-experienced and lower-skilled employees.
This finding has meaningful implications for investor servicing. AI can help distribute institutional knowledge, reduce dependence on informal expertise, and improve consistency across service teams.
Nevertheless, the research was conducted outside fund administration. Its measured productivity gains should not be assumed to transfer directly to regulated investor-servicing workflows.
In financial services, response quality must also be evaluated against factual accuracy, confidentiality, contractual interpretation, and regulatory obligations.
Fund accounting, reconciliations, and exception management
Although distinct from investor servicing, fund accounting and reconciliation processes directly influence the accuracy and timeliness of investor transactions and reporting.
AI-assisted anomaly detection can improve the identification and prioritization of unusual transactions, valuation movements, reconciliation breaks, and potentially erroneous accounting entries.
In October 2026, BNY reported that AI-supported dynamic benchmarking had reduced false-positive alerts in one fund-accounting application from 90% to below 20%. The institution also reported developing more than 40 AI solutions within its asset-servicing business (BNY, 2026).
These are company-reported outcomes rather than independently established industry benchmarks. Nevertheless, they illustrate a valuable transformation: using AI not simply to process more items, but to direct professional attention toward the exceptions that matter most.
Private-market investor servicing
Private markets introduce additional opportunities and risks.
AI can assist with interpreting limited partnership agreements, comparing side-letter provisions, preparing capital-call communications, summarizing distribution notices, and reconciling investor reporting obligations.
However, preferential liquidity rights, fee arrangements, most-favored-nation elections, waterfall provisions, and investor-specific contractual entitlements demand particular care.
AI-generated summaries should never be treated as substitutes for authoritative agreements or controlled economic calculations.
The appropriate objective is to accelerate interpretation and preparation while preserving human validation and deterministic execution of material financial obligations.
Across all these activities, one principle should prevail: AI should first strengthen the quality of operational decisions before institutions consider delegating more authority to automated systems.
3. A new operating principle: Intelligence must not become the system of record
One of the most important design decisions in responsible AI is separating the technology that interprets information from the systems that establish authoritative financial facts.
Large language models are exceptionally useful for understanding natural language, recognizing contextual relationships, and generating explanations. They are not inherently authoritative transaction ledgers, ownership registries, accounting engines, or legal decision-makers.
This distinction has profound implications for fund servicing.
An AI system may explain a distribution calculation, but the underlying calculation must originate from validated data and approved calculation methodologies. It may help summarize an investor’s holdings, but the official record must remain the controlled investor register. It may prepare a transaction instruction, but execution must depend on independently enforced authorization rules.
I would characterize this as a four-ledger approach to responsible AI.
The data ledger establishes which sources are authoritative, current, complete, and accessible for a particular purpose. This includes investor registers, accounting records, contractual documents, and approved reference data.
The obligation ledger captures applicable contractual provisions, operational policies, regulatory requirements, and service commitments. It ensures that an apparently reasonable AI recommendation cannot override a binding obligation.
The decision ledger identifies who or what made a recommendation, what information informed it, which validations occurred, and who authorized the consequential decision.
The evidence ledger preserves the information necessary to demonstrate the integrity of the process through audit trails, approvals, system logs, exception histories, and retained supporting documentation.
These four elements form the foundations of an AI-enabled control architecture.
Retrieval-augmented generation (RAG), introduced by Lewis et al. (2020), offers a useful architectural mechanism for connecting generative models to external knowledge repositories.
In a servicing environment, RAG can help an AI application draw from approved fund documents, policies, and knowledge bases rather than relying solely on information embedded within a model.
However, retrieval does not automatically guarantee correctness. The system must still establish document authority, validity dates, access permissions, completeness, and whether the retrieved material applies to the specific investor or fund.
More importantly, generated interpretations should be validated against structured business rules before producing consequential actions.
The objective is an architecture in which generative AI provides contextual intelligence while deterministic systems enforce financial and operational truth.
This separation becomes increasingly important as financial institutions experiment with autonomous agents.
Without explicit boundaries, an AI application might acquire the technical ability to initiate activities beyond its appropriate business authority.
Responsible architecture must prevent that possibility by design.
4. The principal risks: What responsible AI must control
Deploying AI within investor servicing creates an interconnected risk landscape that extends beyond traditional model performance.
Inaccuracy and fabricated information
Generative AI systems may produce responses that are fluent and plausible but factually incorrect.
In investor servicing, this could involve an inaccurate redemption deadline, incorrect fee explanation, misinterpretation of an investor-specific provision, or fabricated description of an investment restriction.
The potential consequences go beyond communication quality. Incorrect information may influence investor decisions or trigger inappropriate operational action.
Institutions should therefore require source-grounded responses, deterministic validation where feasible, controlled content libraries, and escalation when information cannot be verified.
A model’s apparent confidence should never be accepted as evidence of factual reliability.
Confidentiality, privacy, and information leakage
Servicing providers routinely process highly sensitive information, including investor identities, account balances, beneficial ownership details, bank instructions, tax documentation, and confidential investment records.
AI applications introduce additional pathways through which information could be exposed.
Potential vulnerabilities include inappropriate model inputs, insufficient tenant separation, insecure integrations, excessive data retention, and unauthorized retrieval of investor-specific records.
The OWASP Foundation (2025) identifies sensitive-information disclosure, prompt injection, and excessive agency among significant risks associated with large language model applications.
The appropriate response is to embed data classification, access controls, encryption, retention restrictions, and information minimization directly into AI architecture.
An institutional investor should never receive another investor’s confidential information simply because an AI system retrieved semantically similar documents.
Bias and inconsistent investor treatment
AI models may reproduce historical patterns that are incomplete, biased, or inconsistent with current policies.
In investor servicing, this may affect the prioritization of requests, treatment of exceptional cases, escalation decisions, or the interpretation of investor documentation.
Fairness does not necessarily mean treating every investor identically. Contractual rights, regulatory classifications, and service agreements may legitimately require different treatment.
Responsible AI must distinguish authorized differentiation from unjustified discrimination or arbitrary inconsistency.
Model drift and changing business conditions
Fund servicing environments do not remain static.
New fund launches, amendments to offering documents, changes in regulations, operational restructurings, and changes in investor populations may cause previously reliable AI applications to perform less effectively.
A system that successfully interprets one fund’s redemption provisions may produce incorrect results for another fund with gates, suspensions, side pockets, or differing notice periods.
Consequently, model validation cannot be a one-time implementation exercise.
Ongoing monitoring, periodic testing, and change management are essential.
Cybersecurity and agentic execution
The transition from conversational AI to agentic AI creates a particularly important security challenge.
An agent connected to investor databases, document repositories, payment workflows, and communication platforms may be able to perform multiple actions in sequence.
If permissions are excessive or instructions are maliciously manipulated, the resulting exposure could be substantial.
A useful safeguard is to apply the principle of least privilege: AI systems should receive only the data access and operational permissions required for their specific functions.
High-consequence transactions should require independent authorization, while applications should incorporate transaction limits, explicit approval gates, and the ability to suspend execution.
The lesson is that responsible AI cannot be secured merely through carefully written prompts. Controls must be implemented in the systems and processes surrounding the model.
5. Human oversight: From nominal approval to meaningful accountability
Human oversight is frequently presented as the default solution to AI-related risk.
Yet assigning a human reviewer does not automatically produce better decisions.
Research by Vaccaro et al. (2024), examining 106 experimental studies, found that combinations of humans and AI did not consistently outperform the better-performing participant acting alone. The research identified particularly important differences between decision-making and content-creation tasks.
Similarly, Lee et al. (2025), surveying 319 knowledge workers, found that greater confidence in generative AI was associated with lower reported engagement in critical thinking.
These findings challenge the simplistic assumption that inserting a human approval step into an automated workflow necessarily creates effective control.
For investor servicing, oversight must be substantive, informed, and proportionate to the consequences of the decision.
A reviewer should understand what the AI recommended, the underlying evidence, the relevant business rules, and the circumstances requiring escalation.
Meaningful oversight also requires the practical ability to reject, amend, or challenge an AI recommendation.
I propose three levels of operational autonomy.
Level 1: AI-assisted preparation. The system summarizes documents, drafts communications, retrieves information, and identifies potential exceptions. Human professionals remain responsible for reviewing and completing the activity.
Level 2: Controlled operational execution. The system performs narrowly defined, lower-risk tasks within established business rules. Exceptions, ambiguous cases, and material changes are escalated, with monitoring and a clear ability to reverse or suspend actions where practicable.
Level 3: Restricted consequential decisions. Activities involving material investor rights, movement of funds, legal interpretations, sanctions decisions, or significant changes to ownership records remain subject to explicit authorization and appropriate independent controls.
These categories are proposed operating-model principles, not legally prescribed classifications.
Their purpose is to make automation authority transparent and consistent with operational risk.
Human oversight must also avoid the opposite problem: unnecessarily rejecting accurate machine-generated recommendations because the technology has previously made mistakes.
Dietvorst et al. (2015) documented this tendency as algorithm aversion.
Effective governance therefore requires evidence-based trust: neither blind acceptance nor blanket rejection, but disciplined assessment of AI performance under clearly defined conditions.
The objective is not to retain manual approval everywhere. It is to place professional judgment where it adds measurable value.
6. Regulatory expectations: Principles are converging, obligations remain jurisdiction-specific
Although AI regulation continues to evolve, a consistent supervisory message is emerging internationally: technology adoption does not displace existing responsibilities for investor protection, operational resilience, confidentiality, governance, or accountability.
For global fund-servicing institutions, responsible AI must consequently be designed against the requirements of the activities performed, the relevant legal entities, and the jurisdictions in which services are provided.
European Union
The EU AI Act establishes a risk-based framework for AI systems, with obligations phased according to system characteristics and regulatory classifications.
Certain transparency requirements and AI-literacy obligations are relevant to financial institutions, while specific high-risk requirements depend on the system’s intended purpose and applicable legal provisions (European Commission, n.d.).
It would be incorrect to assume that every investor-servicing chatbot automatically constitutes a high-risk AI system under the Act.
A documented assessment of the use case and applicable requirements is necessary.
The Digital Operational Resilience Act (DORA), applicable from January 2025, also reinforces ICT risk management, operational resilience, and oversight of relevant technology providers within its scope (European Securities and Markets Authority [ESMA], n.d.).
Importantly, GDPR requirements continue to apply to the processing of personal data, including relevant restrictions and safeguards concerning solely automated decision-making with legal or similarly significant effects.
United States
In the United States, the Securities and Exchange Commission (SEC) amended Regulation S-P in 2024 to strengthen requirements addressing customer-information protection and incident response for covered institutions, including relevant transfer agents, investment companies, and registered investment advisers.
The amended requirements have now passed their phased compliance dates of December 2025 and June 2026, depending on entity classification (SEC, 2024).
The SEC’s fiscal year 2026 examination priorities also emphasized information protection, cybersecurity, emerging financial technology, and compliance with applicable regulatory obligations (SEC, 2025).
The practical implication is that AI deployments must remain consistent with existing recordkeeping, supervisory, privacy, and investor-protection requirements.
European investment services and Hong Kong
ESMA (2024) emphasized that investment firms using AI must continue to comply with relevant MiFID II organizational and conduct requirements, including obligations to act in clients’ best interests.
In Hong Kong, the Securities and Futures Commission (2024) introduced expectations for licensed corporations using generative AI language models in regulated activities, emphasizing senior-management accountability, model risk management, cybersecurity, and third-party risk.
These developments demonstrate the importance of incorporating regulatory expectations into business processes rather than treating compliance as a post-implementation review.
For a global servicing organization, the practical solution is a common enterprise AI control framework supported by jurisdiction-specific requirements and use-case assessments.
The National Institute of Standards and Technology’s AI Risk Management Framework, including its generative AI profile, provides a useful approach through governance, risk mapping, measurement, and management (Autio et al., 2024; Tabassi, 2023).
ISO/IEC 42001:2023 similarly provides a management-system framework for governing AI development and use.
These frameworks can support organizational consistency, but they do not replace applicable financial-services laws or regulatory obligations.
7. Third-party AI risk: Outsourcing technology does not outsource accountability
The fund-servicing industry relies extensively on specialized technology providers, cloud infrastructure, data vendors, outsourced operations, and enterprise platforms.
AI increases the complexity of those relationships.
The Financial Stability Board (2025) has specifically identified the need for stronger monitoring of AI-related third-party dependencies and concentration risks.
This is especially important where multiple financial institutions depend on a relatively small group of foundation-model, cloud, or infrastructure providers.
Technology procurement must therefore extend beyond conventional functionality, cost, and service-level considerations.
Institutions should understand how providers process and retain data, whether customer information may be used for model training, how model changes are communicated, where information is stored, and which subprocessors are involved.
Contracts should address confidentiality, security controls, auditability, incident notifications, business continuity, data portability, and termination arrangements.
Equally important is assessing how the institution would continue servicing investors if a critical AI provider became unavailable.
For example, a servicing team may rely on an AI knowledge assistant to interpret investor queries. If the application fails during a period of heavy redemption activity, the institution must still be capable of accessing authoritative records and meeting contractual deadlines.
This requires documented fallback processes and regular resilience testing.
The future operating model should avoid creating a single point of failure by allowing every critical service activity to depend on the uninterrupted availability of one AI platform.
Responsibility for investor outcomes remains with the institution that has undertaken the relevant obligations, regardless of whether its technology is developed internally or supplied by a third party.
8. Measuring responsible AI: Beyond productivity and cost reduction
A significant weakness in many AI transformation programs is the tendency to measure success primarily through adoption, processing speed, or employee time saved.
These metrics are useful but insufficient.
Within fund and investor servicing, responsible AI should be evaluated against four interconnected dimensions.
Operational performance includes turnaround time, straight-through-processing rates, productivity, exception backlogs, and processing costs.
Control effectiveness includes error rates, false positives, false negatives, unauthorized actions, audit findings, and the effectiveness of escalation procedures.
Investor outcomes include response accuracy, service consistency, complaint resolution, communication quality, accessibility, and fulfillment of contractual service commitments.
Institutional resilience includes system availability, recoverability, concentration exposure, model stability, change-control effectiveness, and the institution’s ability to continue operating without AI.
These dimensions should be assessed together.
An AI solution that reduces processing time by 30% but introduces material errors into investor statements should not be considered successful.
Equally, a solution that reduces unnecessary alerts but misses more material exceptions may represent weaker control effectiveness despite apparent efficiency gains.
A more useful economic measure is net risk-adjusted servicing value.
Conceptually, this comprises efficiency and service benefits, less technology costs, validation and remediation costs, and the expected consequences of operational failures.
The purpose is not to suggest that every risk can be quantified precisely. Rather, it is to prevent institutions from overstating AI benefits while overlooking the costs of maintaining safe and reliable operations.
A properly designed pilot should therefore establish baseline measures before implementation, test against representative historical cases, and monitor performance after deployment.
Where feasible, institutions should compare AI-assisted outcomes with existing processes rather than simply evaluating whether an AI system can produce an apparently credible answer.
This shift from technology metrics to business and control outcomes is essential for translating AI investment into durable institutional value.
9. Building a responsible AI governance model
Responsible AI cannot be the exclusive responsibility of technology, innovation, compliance, or operations teams.
It requires clear collaboration across all these functions.
My proposed governance model has five components.
Business ownership
Every AI application should have an accountable business owner who understands its operational purpose and the consequences of failure.
For an investor-servicing application, ownership should sit with the function responsible for the affected process rather than being delegated entirely to the AI development team.
Enterprise standards
Institutions should maintain consistent standards covering acceptable uses, information classification, model validation, transparency, access control, monitoring, and incident management.
An inventory should identify deployed applications, their business purpose, technical dependencies, risk classification, owners, and approved operating boundaries.
Independent challenge
Risk management, compliance, information security, privacy, and internal audit should exercise appropriate oversight according to their mandates.
The level of independent review should reflect the application’s impact on investors, financial records, regulatory obligations, and operational continuity.
Lifecycle accountability
AI governance should extend from design and procurement through testing, implementation, change management, ongoing monitoring, and eventual retirement.
This matters because AI applications may change through model upgrades, revisions to prompts, altered retrieval sources, or modifications to connected workflows.
A significant change should trigger reassessment rather than automatically inheriting an earlier approval.
Evidence-based governance
Finally, controls must generate usable evidence.
An effective governance framework should allow management, auditors, and supervisors to determine what the system was authorized to do, which information it used, what it actually did, and how failures were addressed.
Governance maturity should be judged by the institution’s ability to demonstrate these facts, not simply by the existence of policy documents.
The CFA Institute (2026) has highlighted the structural effects of AI on professional practice, accountability, and financial decision-making. Its broader message is relevant to servicing institutions: the value of professional judgment will change as analytical capabilities become more widely available.
10. A practical transformation roadmap: From isolated pilots to responsible scale
A sustainable implementation strategy should prioritize business value, control readiness, and organizational learning over the number of applications deployed.
I would recommend a phased approach.
Phase 1: Establish the control foundation
The institution begins by identifying existing AI usage, including unapproved applications and embedded capabilities supplied by technology vendors.
It then establishes ownership, acceptable-use standards, data-handling rules, application classification, and oversight arrangements.
At this stage, institutions should also assess the quality and accessibility of their core servicing data.
Poorly governed investor records will not become reliable simply because a more sophisticated AI interface is introduced.
Phase 2: Prioritize high-value, manageable-risk applications
Initial deployments should concentrate on areas where benefits can be measured and human verification remains practical.
Examples include internal knowledge retrieval, document classification, drafting operational correspondence, reconciliation support, and exception triage.
Use cases should be assessed against expected value, information sensitivity, regulatory implications, potential consequences of error, and technical readiness.
The objective is to build institutional confidence through evidence rather than promotional demonstrations.
Phase 3: Validate in operational conditions
AI applications should be tested against realistic servicing scenarios.
These should include incomplete investor instructions, conflicting records, amended documents, unusual transaction requests, multilingual communications, and attempts to access unauthorized information.
Testing should assess not just routine accuracy but also how the system responds to uncertainty, missing evidence, conflicting rules, and malicious inputs.
Where appropriate, pilots should operate in parallel with existing processes before being permitted to influence live activity.
Phase 4: Integrate with controlled workflows
Once performance is demonstrated, AI should be integrated with established workflow, approval, reconciliation, and recordkeeping systems.
This integration is where productivity gains can become enterprise value.
However, execution rights must remain constrained. AI should not acquire unrestricted access to payment systems or investor registers merely because it can successfully perform preliminary analysis.
Phase 5: Scale through continuous assurance
As applications expand, institutions must maintain independent testing, operational monitoring, incident reporting, and periodic reassessment.
Scaling should be conditional on sustained performance rather than the completion of an implementation milestone.
The 2026 McKinsey research provides important context: 44% of surveyed organizations reported scaling AI across their enterprises, yet only a small minority qualified as high performers based on significant reported financial impact (Tinkoff et al., 2026).
The implication is that scaling technology and realizing value are different management challenges.
For fund servicing, the institutions most likely to succeed will be those that redesign processes while maintaining the integrity of investor records, decision rights, and operational controls.
11. The human dimension: Redefining expertise in an AI-enabled servicing organization
Responsible AI is as much a workforce transformation as a technological one.
Traditional fund-servicing roles have often required employees to develop expertise through repetitive processing, exception resolution, procedural interpretation, and direct interaction with investors.
AI will increasingly automate or accelerate parts of this work.
That evolution creates opportunities to elevate professional roles toward judgment-intensive activities: interpreting complex fund structures, resolving exceptional cases, managing operational risk, supporting institutional relationships, and improving servicing design.
However, institutions must avoid undermining the development of foundational expertise.
If junior professionals depend entirely on AI-generated interpretations, they may become less capable of recognizing unusual transactions, challenging flawed outputs, or understanding the logic behind operational controls.
This risk is consistent with the concerns about critical-thinking practices identified by Lee et al. (2025).
Training must therefore extend beyond technical familiarity with AI tools.
Servicing professionals need to understand the limitations of probabilistic models, recognize inaccurate or unsupported outputs, apply relevant operational and regulatory rules, and escalate decisions appropriately.
They also need the confidence to challenge automation.
At the leadership level, success requires greater collaboration between fund-servicing specialists, operations managers, technologists, compliance professionals, and risk experts.
A particularly important future capability will be the ability to translate detailed operational knowledge into well-controlled digital workflows.
Institutional knowledge that previously existed in individual experience, informal instructions, or isolated procedures must increasingly be documented and converted into accessible, validated, and maintainable knowledge assets.
Done responsibly, AI can strengthen organizational learning, improve succession resilience, and reduce dependence on individual employees.
The goal should be to develop professionals who are better supported by technology without becoming less accountable for the work they perform.
12. The future: From reactive servicing to anticipatory intelligence
The longer-term potential of AI extends beyond making existing servicing activities faster.
It creates the possibility of an anticipatory servicing model in which institutions identify issues earlier, allocate resources more intelligently, and address investor needs before operational problems escalate.
Imagine an environment in which AI identifies a recurring onboarding bottleneck, anticipates a likely increase in investor inquiries following a distribution event, recognizes inconsistent reporting data before investor statements are generated, or highlights emerging patterns in unresolved reconciliation breaks.
Such capabilities could help shift servicing organizations from reactive exception management toward proactive operational assurance.
Agentic AI may also progressively orchestrate defined activities across onboarding, documentation, reconciliation, communications, and case management.
Nevertheless, greater capability should not be confused with unrestricted authority.
The future should be one in which AI systems are more deeply integrated into operational processes but remain governed by explicit business rules, verified information, robust access controls, and clearly assigned accountability.
This is particularly relevant as the industry experiments with tokenized funds, digital transfer agency capabilities, and more connected financial infrastructures.
Digital records and distributed technologies may change how transactions and ownership information are represented. They do not eliminate the need to establish legal entitlement, verify investor identity, manage restrictions, reconcile economic interests, or demonstrate compliance.
Indeed, more integrated technology may make those responsibilities more important.
The next generation of fund-servicing institutions should therefore pursue three forms of intelligence simultaneously: technological intelligence to interpret complex information; operational intelligence to recognize and manage process risks; and institutional intelligence to preserve judgment, accountability, and investor trust.
The combination, rather than any individual AI model, will determine sustainable competitive advantage.
Conclusion: Responsible AI is a business capability, not a technology label
The responsible use of AI in fund and investor servicing represents a defining opportunity for the investment-services industry.
The potential benefits are compelling: faster onboarding, improved document processing, more intelligent exception management, consistent investor communications, enhanced operational resilience, and greater capacity for professionals to focus on complex investor needs.
Yet these benefits cannot be pursued independently of the obligations that make fund servicing a trusted institutional function.
An investor does not ultimately judge a transfer agent or fund administrator by the sophistication of its AI technology. The investor expects accurate records, timely execution, protection of confidential information, reliable reporting, fair treatment, and confidence that commitments will be honored.
AI must strengthen these outcomes.
This requires a fundamental shift in management thinking—from deploying AI applications to designing responsible AI-enabled servicing capabilities.
The distinction is critical.
A technology-led approach asks how many tasks can be automated.
A responsibility-led approach asks which activities should be automated, under what conditions, using which sources of truth, with whose authorization, and supported by what evidence.
The former may produce short-term efficiencies. The latter creates a sustainable foundation for institutional transformation.
In the years ahead, I believe the strongest fund-servicing organizations will be distinguished less by how much AI they deploy than by how effectively they combine automation, professional expertise, governance, and investor-centric service design.
The real promise of AI is not to remove human responsibility from fund and investor servicing. It is to make responsible decisions faster, better informed, more consistent, and more demonstrably trustworthy.
That is the standard against which responsible AI leadership should ultimately be judged.
References
Autio, C., Schwartz, R., Dunietz, J., Jain, S., Stanley, M., Tabassi, E., Hall, P., & Roberts, K. (2024). Artificial intelligence risk management framework: Generative artificial intelligence profile (NIST AI 600-1). National Institute of Standards and Technology. https://doi.org/10.6028/NIST.AI.600-1
Bank of England, & Financial Conduct Authority. (2024, November 21). Artificial intelligence in UK financial services—2024. https://www.bankofengland.co.uk/report/2024/artificial-intelligence-in-uk-financial-services-2024
BNY. (2026, October 5). AI in fund operations: The next chapter. https://www.bny.com/corporate/global/en/insights/ai-fund-operations.html
Brynjolfsson, E., Li, D., & Raymond, L. (2025). Generative AI at work. The Quarterly Journal of Economics, 140(2), 889–942. https://doi.org/10.1093/qje/qjae044
CFA Institute. (2026, July 20). Artificial intelligence and the future of finance: A framework for structural change. https://rpc.cfainstitute.org/research/reports/2026/artificial-intelligence-future-of-finance
Dietvorst, B. J., Simmons, J. P., & Massey, C. (2015). Algorithm aversion: People erroneously avoid algorithms after seeing them err. Journal of Experimental Psychology: General, 144(1), 114–126. https://doi.org/10.1037/xge0000033
European Commission. (n.d.). When does enforcement start? AI Act Service Desk. https://ai-act-service-desk.ec.europa.eu/en/ai-act/faq/when-does-enforcement-start
European Securities and Markets Authority. (n.d.). Digital Operational Resilience Act (DORA). https://www.esma.europa.eu/esmas-activities/digital-finance-and-innovation/digital-operational-resilience-act-dora
European Securities and Markets Authority. (2024, May 30). Public statement on AI and investment services (ESMA35-335435667-5924). https://www.esma.europa.eu/document/public-statement-ai-and-investment-services
Financial Stability Board. (2024, November 14). The financial stability implications of artificial intelligence. https://www.fsb.org/2024/11/the-financial-stability-implications-of-artificial-intelligence/
Financial Stability Board. (2025, October 10). Monitoring adoption of artificial intelligence and related vulnerabilities in the financial sector. https://www.fsb.org/2025/10/monitoring-adoption-of-artificial-intelligence-and-related-vulnerabilities-in-the-financial-sector/
International Organization for Standardization, & International Electrotechnical Commission. (2023). Information technology—Artificial intelligence—Management system (ISO/IEC Standard No. 42001:2023). https://www.iso.org/standard/42001
International Organization of Securities Commissions. (2025). Artificial intelligence in capital markets: Use cases, risks, and challenges (Consultation Report CR/01/2025). https://www.iosco.org/library/pubdocs/pdf/IOSCOPD788.pdf
Lee, H.-P., Sarkar, A., Tankelevitch, L., Drosos, I., Rintel, S., Banks, R., & Wilson, N. (2025). The impact of generative AI on critical thinking: Self-reported reductions in cognitive effort and confidence effects from a survey of knowledge workers. In Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems (Article 1121, pp. 1–22). Association for Computing Machinery. https://doi.org/10.1145/3706598.3713778
Lewis, P., Perez, E., Piktus, A., Petroni, F., Karpukhin, V., Goyal, N., Küttler, H., Lewis, M., Yih, W.-t., Rocktäschel, T., Riedel, S., & Kiela, D. (2020). Retrieval-augmented generation for knowledge-intensive NLP tasks. Advances in Neural Information Processing Systems, 33, 9459–9474. https://arxiv.org/abs/2005.11401
OWASP Foundation. (2025). OWASP top 10 for large language model applications: 2025. https://genai.owasp.org/llm-top-10/
Securities and Futures Commission of Hong Kong. (2024, November 12). Circular to licensed corporations—Use of generative AI language models. https://apps.sfc.hk/edistributionWeb/gateway/EN/circular/intermediaries/supervision/doc?refNo=24EC55
Stanford Institute for Human-Centered Artificial Intelligence. (2026). The 2026 AI index report. Stanford University. https://hai.stanford.edu/ai-index/2026-ai-index-report
Tabassi, E. (2023). Artificial intelligence risk management framework (AI RMF 1.0) (NIST AI 100-1). National Institute of Standards and Technology. https://doi.org/10.6028/NIST.AI.100-1
Tinkoff, D., Van der Veken, L., Chui, M., & Balakrishnan, T. (2026, August 25). The state of AI in 2026: On the road to ROI. McKinsey & Company. https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai
U.S. Securities and Exchange Commission. (2024). Enhancements to Regulation S-P: A small entity compliance guide. https://www.sec.gov/file/regulation-s-p-small-entity-compliance-guide
U.S. Securities and Exchange Commission. (2025, November 17). SEC Division of Examinations announces 2026 priorities. https://www.sec.gov/newsroom/press-releases/2025-132-sec-division-examinations-announces-2026-priorities
Vaccaro, M., Almaatouq, A., & Malone, T. (2024). When combinations of humans and AI are useful: A systematic review and meta-analysis. Nature Human Behaviour, 8, 2293–2303. https://doi.org/10.1038/s41562-024-02024-1