Embedded AI Leadership: Why the Steward Is the Answer to the Execution Gap

The boardroom has received the message. Artificial intelligence is no longer an option but a strategic necessity. In virtually every sector, AI strategies are being developed, innovation budgets freed up and pilot projects launched. Yet the gap between ambition and realisation remains strikingly large: according to McKinsey research, only 11% of AI initiatives at enterprise level reach the production phase. The problem is rarely the technology. The problem is leadership in execution.

This article examines why traditional advisory methods fall short in AI transformation, what embedded AI leadership entails, and how the Steward model helps organisations bridge the gap from strategy to sustainable value creation — without creating lasting dependency on external parties.

The execution gap: where AI strategies stall

Virtually every organisation that seriously embarks on AI recognises the pattern. An ambitious strategy is drawn up, often with the help of a reputable consultancy firm. The report contains impressive analyses, clear priorities and an attractive business case. The CEO and board are enthusiastic. Budget is allocated and a steering committee formed.

And then reality sets in.

The first few weeks, the workshops go well. But as soon as it comes to concrete implementation — setting up data pipelines, aligning models with business processes, navigating governance requirements, persuading departments that fear for their position — momentum begins to evaporate. The consultants delivered a blueprint, but not the builders. The internal organisation lacks the right combination of AI expertise, change management and operational leadership to bring the plan to execution.

This is the execution gap. It manifests in various forms:

• Pilot paradise: Dozens of proof-of-concepts that never scale to production, because no one is responsible for the transition from experiment to operations.
• Governance vacuum: AI models are built without clear ownership, risk frameworks or compliance mechanisms, leading to standstill as soon as risk or legal raises concerns.
• Talent shortage: Data scientists are hired but work in isolation from the business. There is no leader who translates technical capabilities into business value.
• Change fatigue: Employees experience AI as yet another transformation programme imposed from above, without considering their involvement or interests.

The result is predictable: after 12 to 18 months, the board concludes that the millions invested in AI have not yet delivered measurable impact. Confidence in AI as a strategic lever declines, budgets are frozen, and the organisation loses valuable time relative to competitors who did manage to break through the execution phase.

Why traditional advisory falls short

The traditional consultancy model is designed for analysis and recommendation, not for sustained execution. There are three structural reasons why this model fails in AI transformation.

First: the delivery model is episodic. Consultants work in project phases with a beginning and an end. After the strategy phase they leave, and the internal organisation must deliver on its own. But AI transformation is not a project — it is a continuous process of learning, adapting and scaling that takes months to years. What is needed is sustained leadership, not periodic interventions.

Second: knowledge transfer is superficial. Strategy reports contain 'what' and 'why', but rarely 'how' in sufficient depth. The nuances of AI implementation — navigating data quality issues, managing model risks, setting up MLOps, building cross-functional teams — are context-specific and require hands-on guidance, not slide decks.

Third: the incentive structure is misaligned. The traditional consultancy revenue model rewards complexity and duration, not client self-sufficiency. The longer the engagement lasts and the more dependent the client becomes, the better for the consultant. This is fundamentally at odds with what organisations actually need: building their own AI capability that functions independently over time.

What is embedded AI leadership?

Embedded AI leadership is a fundamentally different model. Instead of advice from a distance, an experienced AI transformation leader is placed within the organisation — someone who functions as part of the management team, but with the explicit mandate to build the organisation's capability and ultimately let go.

This model distinguishes itself on multiple dimensions:

• Position: The Steward operates at C-level or directly below it, with a mandate from the board and direct access to decision-making.
• Perspective: Unlike an external adviser, the Steward understands the internal culture, politics and processes because they are part of them daily.
• Accountability: The Steward is not accountable for a report, but for results. Performance is measured against actually realised AI value, not delivered advisory documents.
• Time horizon: A typical Steward engagement lasts 6 to 18 months — long enough to achieve structural change, but with a built-in wind-down mechanism.

The term 'steward' is deliberately chosen. A steward is not an owner but a custodian: someone who bears responsibility for the flourishing of something that does not belong to them, with the ultimate goal that the organisation can continue on its own. This is the essence of the model: building capability, not dependency.

The Steward model: four phases of transformation

The Steward model is structured around four sequential phases, each with its own objectives, deliverables and handover criteria. The four phases — Foundation, Acceleration, Scale and Transfer — form a cohesive whole that takes organisations from initial orientation to independent AI capability.

Phase 1: Foundation (month 1–3)

In the Foundation phase, the Steward creates the preconditions for successful AI transformation. This includes a thorough assessment of current AI maturity, data quality, technical infrastructure and organisational readiness. Simultaneously, the governance framework is established: who owns AI models, how are risks assessed, which ethical frameworks apply, and how is compliance with regulation such as the EU AI Act ensured.

Critical in this phase is identifying two to three 'quick win' projects that deliver demonstrable value within 8 to 12 weeks. These projects are strategically chosen: important enough to maintain board-level attention, yet limited enough in scope to actually achieve results quickly. The aim is to create momentum and let the organisation experience that AI is not an abstract concept but a concrete instrument for value creation.

The Foundation phase also delivers the governance structure: an AI steering committee with clear mandates, a risk classification model for AI applications, and a first version of the AI policy tailored to the organisation's specific context and sector.

Phase 2: Acceleration (month 3–8)

Once the foundations are in place, the focus shifts to acceleration. The quick-win projects are brought to production and learnings are systematically documented. Based on insights from the Foundation phase, a pipeline of AI use cases is prioritised according to business value, feasibility and strategic relevance.

In the Acceleration phase, the Steward actively begins building internal capacity. This means not only hiring or reskilling employees, but also establishing work processes: how do data scientists collaborate with domain experts? How does the handover from model to operations work? Which tooling and platforms become standard? How is model performance monitored after deployment?

Simultaneously, in this phase the business case for AI at organisation level is sharpened. The initial projects deliver hard data on costs, benefits and implementation time, which underpins the business case for further scaling. The Steward uses these results to strengthen board-level support and secure additional budget where needed.

Phase 3: Scale (month 8–14)

The Scale phase focuses on expanding proven AI applications to multiple departments or business units, and industrialising AI operations. What ran manually or semi-automatically in the Acceleration phase is now structurally implemented: CI/CD pipelines for models, automated monitoring, standardised deployment processes and scalable data architecture.

This is also the phase in which organisation-wide change management initiatives are embedded. The Steward facilitates knowledge sharing between departments, establishes communities of practice, and ensures successes are communicated broadly to strengthen the flywheel effect. The culture change needed for successful AI adoption — data-driven decision-making, experimentation culture, cross-functional collaboration — is actively stimulated and measured.

A critical element in the Scale phase is anchoring the governance structure. The AI policy is evaluated and refined based on accumulated experience. Compliance processes are automated where possible. And explicit attention is given to interaction with regulators and other external stakeholders.

Phase 4: Transfer (month 14–18)

The Transfer phase is where the Steward model fundamentally distinguishes itself from other forms of interim management or consultancy. In this phase, full AI leadership is transferred to the internal organisation. Throughout the earlier phases, the Steward has systematically worked on identifying and developing internal leaders who can take over the role.

The handover is not an abrupt moment but a gradual process. The Steward shifts from executing to coaching, from deciding to advising. A formal handover dossier is prepared containing architecture decisions, governance frameworks, ongoing initiatives, risks and recommendations. The internal leader gradually assumes more responsibility, with the Steward as safety net.

The success criterion for the Transfer phase is clear: the organisation functions at a higher level in AI strategy, governance and execution than before the Steward's arrival, and possesses the internal capacity to independently maintain and further develop this level.

Capability building versus dependency

The difference between capability building and dependency creation is the most fundamental distinction in AI transformation services. Yet it is rarely made explicit.

In the traditional model, dependency on the external party grows as the programme progresses. Knowledge resides in the minds of consultants, not in the organisation. Processes run on external capacity. Once the contract ends, a significant portion of AI capability disappears.

The Steward model reverses this dynamic. From day one, the explicit goal is to strengthen the organisation itself. Every decision, every process and every system is designed with the question: 'Can the internal team continue this without us?' This manifests in concrete practices:

• All documentation, code and architecture decisions are recorded in systems that belong to the client, not to the Steward.
• Internal employees are involved in decision-making from the start and collaborate on implementation — not as spectators but as co-owners.
• Training and knowledge sessions are a structural part of the programme, not an optional add-on.
• The Steward's success metrics explicitly include indicators for internal capability building, such as the number of certified internal AI specialists and the degree to which internal teams can independently execute projects.

Results from practice

The impact of embedded AI leadership becomes visible along three dimensions: speed, value and sustainability.

Speed: Organisations working with an embedded Steward bring their first AI application to production on average two to three times faster than organisations relying solely on internal capability or traditional advisory. This is not because the Steward takes over the work, but because they recognise and address the typical pitfalls and delays — governance deadlocks, technical dead ends, organisational resistance — early.

Value: Through systematic prioritisation based on business value, Steward programmes realise significantly higher ROI than bottom-up AI initiatives. The focus on quick wins in the Foundation phase moreover generates early evidence of value, which increases support for further investment. Organisations typically report a payback period of 6 to 9 months on the Steward investment, measured against the value of realised AI applications.

Sustainability: The most distinguishing result is what happens after the Steward's departure. Organisations that have completed the Steward model have a functioning AI governance framework, trained internal teams, standardised processes and a proven pipeline of use cases. These are the building blocks for lasting value creation, independent of external support.

Is your organisation ready for embedded AI leadership?

Embedded AI leadership is not the right choice for every organisation at every moment. The model works best when three conditions are met:

1. Board commitment: The board recognises that AI is a strategic priority and is willing to provide the Steward with the mandate and access needed to operate effectively.
2. Willingness to change: The organisation is prepared to adapt processes, roles and ways of working based on the insights that emerge from the AI transformation.
3. Long-term perspective: There is understanding that sustainable AI value creation is not a matter of weeks but of months, and that the investment in capability building only fully pays back in the medium term.

When these conditions are met, the Steward model offers a proven path from AI ambition to AI reality — without the pitfalls of traditional advisory and without the dependency that many organisations experience with prolonged external engagement.

The question is no longer whether your organisation should adopt AI. The question is who leads the execution, and whether that execution results in lasting capability or temporary activity. The answer to that question determines the difference between AI as a cost item and AI as a competitive advantage.