How AI forces asset managers to rethink execution

I previously stressed that artificial intelligence is not a tooling problem for asset managers, but an operating model problem and that strategy in the age of AI increasingly manifests itself through operating architecture rather than ambition alone. These perspectives inevitably lead to a more difficult and practical question: how does such a transformation actually unfold in a non-AI-native organisation?

Click here for the previous articles in this series: How AI forces asset managers to rethink their operating models and How AI forces asset managers to rethink their strategy.

This article addresses that question. Not from the perspective of technology-first firms, but from that of incumbent asset managers operating in regulated, trust-based environments. The objective is sustainable transformation rather than speed. The focus is therefore on sequencing, organisational realism and execution discipline.

Executive summary

This third article moves from strategy to execution, addressing the most difficult question facing incumbent asset managers: how AI transformation actually unfolds in a non-AI-native organisation. Building on the earlier argument that AI is an operating-model challenge and that strategy increasingly manifests through operating architecture, the article focuses on realism, sequencing and execution discipline rather than speed.

The central premise is that AI transformation is cumulative rather than linear. It cannot be delivered through isolated initiatives or time-bound programmes, but emerges through reinforcing changes across data, architecture, governance and people. When AI is treated as a programme, organisations often achieve technical success without organisational impact - proofs of concept that fail to alter how decisions are made.

The article introduces a pragmatic, phased transformation journey. Phase 0 and Phase 1 are positioned as non-negotiable foundations. Strategic grounding must precede technical ambition, ensuring that AI efforts are anchored in decision quality, complexity management and long-term positioning rather than in use-case enthusiasm. Data-centricity follows as the true starting point of execution: establishing consistent definitions, ownership, quality, lineage and security. Without this, AI-enabled learning cannot scale.

Subsequent phases describe how an AI-enabled operating layer institutionalises learning, how AI insight is deliberately integrated into decision processes and how governance evolves. Throughout, the article emphasises that AI augments human judgement rather than replacing fiduciary responsibility and that governance and security are prerequisites for scaling sustainably, not constraints.

A critical execution risk is staffing. The article argues that AI transformation cannot be absorbed ‘on the side’ by existing teams. A dedicated, cross-functional transformation team is required to coordinate strategy, data, analytics, governance and organisational change. While external partners can accelerate progress, strategic ownership and learning must remain internal. Speed without institutional learning creates dependency and erodes long-term optionality.

Finally, the article explains why the journey is intentionally non-linear. Data, governance, architecture and strategy continue to evolve together. Progress must therefore be measured not only through traditional output KPIs, but through indicators of learning capacity: reduced decision latency, increased reuse of insight, improved consistency and faster adaptation.

The concluding message is clear. For non-AI-native asset managers, AI transformation is not about becoming technology firms. It is about becoming learning organisations capable of managing complexity responsibly at scale. In this way capability rather than any individual model or application is the true source of durable advantage in the age of AI.

Why AI transformation fails when treated as a programme

Many AI initiatives fail not because of insufficient intent or talent, but because they are framed as time-bound programmes. Roadmaps are defined, budgets allocated and pilots launched, with the implicit expectation that transformation will follow.

This logic misunderstands the nature of AI. AI does not introduce a new capability that can simply be deployed. It introduces new learning dynamics that only materialise once systems, data and organisations interact over time.

For asset managers, this mismatch is particularly pronounced. Programme thinking prioritises predictability and control, while AI-enabled learning requires iteration and feedback. The result is often technical success without organisational impact: proofs of concept that work in isolation but do not structurally change how decisions are made.

Transformation must therefore be approached as an evolution of the operating model, not as a project with a defined end date.

Phase 0: Strategic grounding before technical ambition

Before any technical investment, organisations must clarify why they are pursuing AI. This is not a question of use cases, but of strategic intent. Executive sponsorship is a pre-requisite. Strong communication is an accelerator.

Which decisions should the organisation become better at over time? Where does complexity accumulate? Where does learning stall? What types of judgement must remain human, and where can systems augment consistency and insight?

For non-AI-native asset managers, this phase creates alignment. Without it, AI initiatives gravitate towards what is technically interesting rather than what is strategically relevant. Expectations fragment across teams and governance becomes reactive.

Phase 1: Data-centricity as the true starting point

Despite common narratives, AI transformation does not start with models. It starts with data.

As discussed in the previous articles, most asset managers are data-rich but data-poor in practice. Data exists across portfolio management systems, risk platforms, market feeds, client systems and reporting tools, but it is fragmented, inconsistently defined and difficult to mobilise.

Phase 1 therefore focuses on data-centricity:

establishing consistent data definitions across the organisation;

clarifying ownership and accountability for critical data domains;

designing reusable data pipelines rather than point-to-point integrations;

ensuring data quality, lineage and traceability.

This work is technically demanding, politically sensitive and often invisible. Yet without it, AI-driven learning cannot emerge.

Data security and confidentiality are integral to this phase. Access controls, encryption, segregation of sensitive datasets and compliance with regulatory expectations must be designed into data architecture from the outset. Security is not an add-on; it is a prerequisite for trust and scale.

Crucially, data-centricity must be business-led and IT-enabled. Domain experts define meaning and relevance; IT provides infrastructure and ensures robustness, scalability and security.

Phase 0 and Phase 1 are not preliminary exercises that can be accelerated or bypassed; they constitute the non-negotiable foundations upon which all subsequent AI capabilities depend.

Phase 2: Building the AI-enabled operating layer

Once data foundations are in place, organisations can begin to construct what is often referred to as an AI-enabled operating layer or ‘AI factory’. For asset managers, this is not an industrial automation engine, but a mechanism for institutionalising learning.

This phase typically involves:

creating shared analytical environments;

standardising model development, validation and monitoring practices;

embedding feedback loops between decisions and outcomes;

promoting a data culture across the entity;

enabling reuse of analytical components across teams.

This layer must coexist with existing governance structures. Risk ownership, fiduciary responsibility and accountability remain human. AI augments decision-making; it does not replace it.

Phase 3: Integrating AI into decision processes

AI creates value only when it influences decisions. This phase focuses on embedding AI-enabled insight into daily workflows.

For asset managers, this integration must be deliberate. Rather than automating decisions, organisations should explicitly define:

which decisions remain fully human;

which are supported by AI-generated insight;

and where AI operates within clearly bounded domains.

This phase often brings cultural tensions to the surface. Professionals may perceive AI as reductive or intrusive. Addressing these concerns requires transparency, explainability and clarity about roles.

Successful integration treats AI as a partner in judgement, not as a substitute for expertise.

Phase 4: Governance, ethics and control at scale

As AI becomes embedded, governance shifts from oversight to design. Ethical considerations such as bias, transparency and accountability become operational requirements.

This phase involves:

defining acceptable-use boundaries;

ensuring explainability and auditability of AI-supported decisions;

monitoring drift and unintended effects;

embedding accountability into workflows and systems.

For European asset managers, this phase aligns naturally with existing governance traditions. When embedded early, governance enables scale rather than constraining it.

Phase 5: Organisational adaptation and talent evolution

AI transformation ultimately reshapes roles, not headcount. As systems absorb routine coordination and analysis, human effort shifts towards interpretation, oversight and strategic judgement.

This phase requires:

redefining roles and expectations;

investing in hybrid profiles that bridge domain and data expertise;

adjusting performance metrics to reward learning and collaboration.

Framing AI as a means of amplifying professional judgement build engagement.

Staffing the transformation: why a dedicated team is essential

One of the most underestimated aspects of AI transformation is staffing. Many organisations assume that AI initiatives can be absorbed by existing teams, layered on top of current responsibilities. Although upskilling through training is intended to drive progress, in practice it typically leads to fragmented initiatives and slow momentum.

AI transformation is organisationally demanding. It requires sustained focus, cross-functional coordination and the ability to navigate technical, regulatory and cultural complexity simultaneously. Expecting this work to happen ‘on the side’ is unrealistic.

For non-AI-native asset managers, a dedicated transformation team is therefore a prerequisite for execution.

Core roles in a dedicated AI transformation team

AI / Digital Transformation Lead Provides overall direction and coherence. Aligns AI initiatives with strategy and creates an interface with executive leadership.

AI / Analytics Lead Oversees model development and validation. Ensures methodological rigour and close alignment with domain expertise.

Business Domain Lead(s) Senior professionals from investment, risk or operations who articulate where complexity accumulates and ensure that AI efforts reflect real decision logic.

Data Product Owner Defines data semantics, quality standards and prioritisation. Treats data as a strategic product rather than a technical artefact.

Data Engineering Lead Designs scalable, secure and auditable data pipelines. Ensures lineage, traceability and compliance.

Governance & Risk Liaison Embeds regulatory, ethical and risk considerations into system design rather than acting as an ex-post gatekeeper.

External partners can accelerate execution, but strategic ownership must remain internal. AI transformation builds institutional capability; it cannot be fully outsourced. Everything related to internal domain expertise could come from internal people, but some targeted recruiting will remain necessary to credibly roll out AI strategy. There is always the tension between buy or build. In fact, most incumbent asset managers will, at some point, rely on external partners to accelerate their AI transformation. This is both rational and often necessary. External expertise can shorten learning curves, reduce execution risk and bring proven practices into the organisation.

However, this introduces a fundamental trade-off between speed and institutional learning. Buying capability accelerates delivery, but risks externalising the very learning that AI transformation is meant to institutionalise. When core architectural decisions, data semantics or model logic remain outside the organisation, dependency increases and strategic optionality decreases.

Why the journey is non-linear by design

Although presented sequentially, these phases overlap and reinforce one another. Data work continues as models evolve. Governance adapts as use cases expand. Strategy is refined as learning accumulates.

This non-linearity is not a flaw. It is the defining characteristic of AI-enabled transformation. Attempts to force linearity often lead to stagnation. For asset managers, patience and architectural discipline are strategic virtues.

Finally, measuring progress, from output to learning capacity, requires new learning related KPIs. Traditional KPIs focus on output: efficiency gains, cost reduction or performance improvements. While relevant, they fail to capture AI’s core contribution. More meaningful indicators include:

reduced decision latency;

increased reuse of insight across teams;

improved consistency in decision-making and prediction quality;

faster adaptation to new information.

These metrics reflect learning capacity rather than short-term results.

Conclusion

AI transformation in asset management is neither fast nor linear. It requires deliberate design, sustained investment and organisational focus.

The journey starts with data, matures through operating architecture and is carried by people. Governance and security are not obstacles, but anchors.

For non-AI-native asset managers, the objective is not to become a technology company. It is to become learning organisations capable of managing complexity responsibly at scale.

That is the real promise - and the real challenge - of AI.