Seventy per cent of marketing transformation initiatives fail. McKinsey’s research across enterprise AI deployments is unambiguous. The pattern is relentless: brilliant technology, medieval org chart.
Technology is not what breaks. Organisations break when they try to push autonomous AI into structures designed exclusively for humans. Every failed programme carries the same signature: capable engineers, committed executives, and an operating model designed for a world that no longer exists.
Agentic AI does not slot into an existing team like a new marketing automation platform. Generative AI augments individual productivity. Agentic AI introduces autonomous colleagues that perceive market conditions, make decisions, execute campaigns, and learn from results. That demands a different architecture.
This article lays out three battle-tested models for structuring hybrid intelligence within marketing organisations: hub-and-spoke, embedded, and hybrid collaborative. Each carries distinct advantages depending on regulatory environment, digital maturity, and appetite for decentralisation. Choosing well is the most consequential structural decision a CMO will make in 2026.
Why the Old Structures Break
The traditional campaign development process involved creative teams, data analysts, channel managers, and compliance officers in sequential handoffs spanning weeks. The architecture made sense when every decision required human cognition.
Agentic AI inverts the assumption. An AI agent autonomously monitors customer behaviour, identifies micro-segments, generates personalised content variations, and orchestrates omnichannel deployment. The sequential handoff is no longer the bottleneck. The architecture that supported sequential handoffs is now the bottleneck.
Organisations deploying agentic systems into traditional org charts report a predictable failure pattern. Campaign velocity does not change because the human review cycles still take weeks. Quality of customer engagement does not improve because the human creative direction never reaches the agents. The technology works. The structure prevents the technology from compounding.
The fix is not training. It is architecture.
The Hub-and-Spoke Model
The hub-and-spoke architecture centralises AI capabilities in a dedicated centre of excellence serving the broader marketing organisation. It works particularly well for organisations beginning their agentic AI journey or those in highly regulated industries requiring stringent governance.
A central AI hub houses specialists — AI trainers, prompt engineers, performance analysts, governance leads. These experts work with agentic systems serving multiple marketing teams across the organisation. The spokes represent traditional marketing functions — brand, demand generation, customer experience, analytics — that submit requests to and receive outputs from the central hub.
Centralised oversight ensures consistent ethical standards, maintains regulatory compliance, and allows rapid deployment of updates across all AI agents. The model creates economies of scale. Rather than each team investing in AI expertise, the hub provides shared services. LVMH’s partnership with Google Cloud exemplifies this approach, enabling AI-driven personalisation and campaign management across its portfolio of brands through a centralised human-AI operating model.
The financial returns typically show thirty to fifty per cent cost savings through centralised infrastructure and shared expertise. The true value lies in risk mitigation. Centralised governance significantly reduces the probability of brand damage or regulatory violations from unsupervised AI actions.
The trade-off is speed. At a financial information provider with market-moving events demanding same-day campaign adjustments, hub-and-spoke cannot deliver the cadence the business requires. In B2B contexts with complex, high-value campaigns, hub-and-spoke works. In B2C environments with live consumer engagement, it can throttle the agility the model was meant to enable.
The Embedded Model
The embedded model distributes AI capabilities directly into functional marketing teams. Each team — brand, performance marketing, customer experience — incorporates both human members and dedicated AI agents as integral parts of its structure.
This approach treats agentic AI as full team members rather than external services. Embedded AI agents participate in daily stand-ups through human facilitators, contribute to strategy discussions, and autonomously execute within defined parameters. The philosophical shift from AI as instrument to AI as colleague changes how organisations approach capability development at every level.
A luxury fashion house exemplifies this model. Their brand team includes an AI agent specialising in trend analysis and creative inspiration. Their e-commerce team features an AI agent managing dynamic pricing and inventory predictions. These agents do not just execute tasks. They proactively identify opportunities, suggest strategic pivots, and learn from team feedback.
The embedded model demands significant investment in training and change management but yields exceptional results in organisations with strong digital maturity and collaborative cultures. Early adopters customise AI agents to align with their team’s communication styles and brand voice. A performance marketing team’s agent might prioritise efficiency and testing velocity. A brand team’s agent might emphasise consistency and creative coherence.
The risk is fragmentation. Without central guardrails, one team’s AI agent can optimise for metrics that directly undermine another team’s objectives. The performance agent optimising for short-term conversion can damage brand equity the brand team is trying to compound. Embedded models without coordination mechanisms produce locally excellent and globally incoherent output.
The Hybrid Collaborative Model
The hybrid collaborative model combines elements of both hub-and-spoke and embedded structures. A lean central AI excellence team establishes standards and develops capabilities, whilst AI specialists and agents are embedded within functional teams.
A multinational technology company pioneered this approach, creating what it calls “AI-powered squads”. Each squad includes marketers, an AI facilitator, and multiple specialised AI agents. A central AI council provides governance, shares operating standards, and ensures interoperability between agents across squads.
The structure enables both standardisation and customisation. Teams develop specialised human-AI capabilities whilst maintaining organisational coherence. Network effects emerge as AI agents learn from each other. When the content team’s agent discovers a particularly effective message variation, that learning can be rapidly shared with the social media team’s agent. Collective intelligence amplifies the impact of individual improvements.
WPP has adopted hybrid models, centralising AI expertise while embedding AI capabilities within teams to balance innovation with governance. The central team acts as an innovation lab, experimenting with emerging capabilities before deploying them to embedded teams. Embedded specialists ensure agents remain closely aligned with business needs.
The investment required exceeds either pure hub-and-spoke or embedded models. The returns also tend to exceed both. Organisations report significant improvements in speed and efficiency for new campaigns whilst maintaining governance standards.
The model requires sophisticated coordination mechanisms, clear role definition between central and embedded resources, and mature organisational change management. Without those, it becomes the worst of both architectures rather than the best of either.
The Human-AI Division of Excellence
The key to successful collaboration lies not in determining what AI can do but in understanding what humans and AI each do best. This is about complementary excellence, not replacement.
Where humans excel. Creative vision grounded in cultural context. Emotional intelligence in crisis communications. Ethical judgement under genuine uncertainty. Strategic thinking with incomplete information. Relationship building with stakeholders. Brand intuition that survives a thousand A/B tests.
Where AI excels. Pattern recognition across millions of data points. Real-time optimisation at machine speed. Consistency at scale. Tireless execution of well-defined workflows. Multi-objective optimisation when the objectives have been clearly specified.
The dividing line is not fixed. As models improve, more tasks migrate from human-only to human-AI to AI-led. The CMO’s job is not to predict where the line will be in two years. It is to design an organisation that re-evaluates the line every quarter and reallocates work accordingly.
The Decision Rights Matrix
Clear decision rights prevent confusion, ensure accountability, and build trust in human-AI operations. The framework must balance AI autonomy with human oversight, enabling speed whilst maintaining control.
AI Autonomous Decisions. Real-time bid adjustments within defined budget parameters. Content variation selection based on performance data. Campaign timing optimisation within approved windows. Audience segment refinement within privacy guidelines.
AI-Recommended, Human-Approved Decisions. Budget reallocation above specified thresholds. Major creative pivots or message changes. New channel activation or partnership opportunities. Targeting strategy modifications.
Human-Only Decisions. Brand positioning and core value propositions. Crisis response and sensitive communications. Ethical considerations and values-based choices. Strategic partnerships and long-term investments.
The matrix is not theoretical. It is the operating system that determines what gets done at machine speed and what gets done at human speed. Get the matrix wrong and you either throttle the agents into uselessness or expose the brand to autonomous decisions it cannot afford to make.
Choosing Your Architectural Model
A structured assessment helps identify the right starting point.
Hub-and-Spoke indicators. Highly regulated industry. Limited AI expertise currently in the organisation. Strong need for standardisation and governance. Centralised decision-making culture. Budget constraints requiring shared services. Early stage of AI maturity.
Embedded model indicators. Digitally mature organisation. Decentralised, autonomous team culture. Existing AI expertise distributed across teams. Need for rapid market response. Strong innovation culture. Comfort with managed risk.
Hybrid model indicators. Large, complex organisation. Mix of experimental and traditional teams. Need for both control and agility. Sufficient budget for dual investment. Mature change management capabilities. Ambition to lead industry transformation.
Score each set. The highest score indicates the most suitable model. Tied scores suggest a phased approach — start with the simpler model, evolve toward complexity. Low scores across all models indicate a need for foundational capability building before any architectural commitment.
Most large enterprises start with hub-and-spoke and evolve toward hybrid as maturity grows. B2B companies often emphasise central governance more heavily. B2C organisations grant more autonomy to embedded teams. Start-ups rarely need the complexity. Embedded models typically suffice.
Three Architectural Decisions, Not Discussions
These cannot be delegated to a planning committee. They sit on the CMO’s desk.
Map your current architecture against the three models. Most organisations are running a hybrid model by accident — partially centralised, partially embedded, with no explicit coordination layer. Diagnosing the accident is the first step to designing the intentional version.
Run the Decision Rights Matrix exercise on every AI deployment currently in production. Where is the matrix explicit and documented? Where is it inferred and inconsistent? Inconsistency is where governance failures live.
Identify the single AI role you have not yet hired but the architecture clearly demands. AI trainers, prompt engineers, performance analysts, governance leads — each of these did not exist three years ago and will be ubiquitous within three more. The talent decision made this quarter compounds.
The gap between marketing’s evolution and enterprise capacity to support it marks the true challenge. Which of the three models most closely matches your organisation’s current reality, and which one should it match eighteen months from now?
The gap between those two answers is your architectural transformation roadmap. The organisations that get this decision right do not do so through planning committees or strategy papers. They get it right by running the architecture itself as an experiment, measuring what compounds and what throttles, then systematically evolving the operating model in response to real results.
Keep Reading
That’s all for this week book chapter summary, come back next Monday for the next chapter summary.
The Agentic CMO - Second Edition is available today in hardcover, paperback and ebook.
Disclaimer: The views and opinions expressed in The Agentic CMO, Chronicles of Change and on my social media accounts are my own and do not necessarily reflect the official policy or position of S&P Global.