From AI experimentation to scale: Rethinking AI governance

AI has been at the centre of discussions in the news industry for the past few years, but the nature of those discussions is shifting. Up until 2025, many of our clients were primarily focused on identifying effective AI use cases across editorial, reader engagement and monetisation. Our AI Launchpad programmes in 2024 (1st and 2nd edition) and 2025 reflected this phase, helping publishers explore, identify and deploy AI use cases.

Publishers are still experimenting with new products and use cases, but we are now seeing more attention turn to scale. Instead of AI remaining in the hands of small groups of specialists or AI-savvy employees, publishers increasingly want more journalists and teams across the organisation to benefit from the technology.

This shift is also reflected in the evolution of FT Strategies' work. In 2026, our AI Lab, supported by the Google News Initiative, broadened the focus from individual use cases to assessing AI readiness across the organisation, looking at eight dimensions: Leadership commitment, Vision and Alignment, Technology Infrastructure, Content and Data Infrastructure, AI Product and Tooling, Ethical and Legal Risk Management, Adoption, and Governance. Through our discussions with clients across these eight dimensions, Governance emerged as a particularly important area of focus, with many publishers highlighting the need for clearer definitions of who makes decisions about AI, which decisions sit centrally, and how risks, standards and learning are managed.

So, what governance models are we seeing in the industry, what are the trade-offs between them, and why is a ‘Centre of Enablement’ becoming particularly relevant?

What do we mean by AI governance?

AI governance determines how decisions about AI are made: who sets strategic priorities, who can manage experiments or tool deployment, how risks are assessed, and how knowledge is shared across the organisation.

At FT Strategies, we group AI governance into three broad models depending on how centralised decision-making and delivery are: Centre of Excellence, Centre of Enablement, and Decentralised / Embedded Teams. There is no ‘one size fits all’ governance structure. Publishers can combine elements of more than one model or evolve their structure over time. The framework helps publishers understand the trade-offs of each approach and consider which elements best fit their organisation.

1. Centre of Excellence: centralised ownership and oversight

How it works

This is the most centralised of the three models. There is usually a dedicated AI lead and team that manages AI strategy in line with leadership priorities and wider business goals. The central team often oversees AI tool deployment, the AI product roadmap and prioritisation, and gathers ideas from departments across the organisation.

Benefits

• Alignment: Central ownership makes it easier to connect AI investment - whether a new tool, product or experiment - to the wider strategy. This is becoming more important as the number of potential AI use cases grows and the challenge shifts from finding ideas to deciding which ones are worth pursuing.
• Consistency: A central team can apply common standards across AI procurement, development and use. This can also reduce duplication, because one team has visibility across the organisation rather than several departments independently procuring tools or building similar solutions.
• High degree of oversight: Concentrating AI expertise and visibility in one team can strengthen oversight of data, privacy, editorial and business risks. It can also help assess whether proposed AI initiatives comply with regulations and internal policies.

Trade-offs

• Capacity as a potential bottleneck: If most AI development, procurement and approval sits with one central team, the pace of delivery depends on their bandwidth. As AI tools evolve quickly and demand for support grows across departments, this can make it harder to respond to needs at speed.
• Isolation from end users: AI products work best when they are designed around real workflows, so central specialists need close and continuous collaboration with the people who will actually use them. One of the recurring lessons from our AI Launchpad programmes was the importance of involving end users - often the newsroom - early enough to shape the product.
• Dependence on central expertise: If AI knowledge and decision-making remain concentrated in the central team, other departments can become dependent on specialists to identify and implement opportunities. This can reduce local ownership and make it harder to build AI capability across teams.

2. Decentralised / Embedded Teams: greater local autonomy

How it works

At the other end of the spectrum, decision-making and delivery sit predominantly within individual departments or teams. Embedded AI expertise is used to identify, prioritise, develop or procure solutions for local needs. Organisation-wide strategy, policies or guardrails may still exist, but teams have greater autonomy over day-to-day AI experimentation and implementation.

This is different from unmanaged AI use that emerges simply because an organisation has no formal governance structure. A decentralised model can still be deliberately designed, with clear responsibilities and organisation-wide standards.

Benefits

• Agility: Teams can experiment with use cases and tools without waiting for every decision to move through a central function. This can increase the volume and speed of experimentation and, in turn, help departments identify applications that genuinely fit their workflows.
• Local relevance and ownership: Even when two publishers adopt the same AI tool, their workflows, systems, roles and editorial processes are rarely identical. Giving teams more ownership can help them adapt AI to their reality.

Trade-offs

• Inconsistency: Without strong shared guardrails, decentralised teams can interpret policies differently and adopt different standards around areas such as data use, editorial review or AI-generated outputs.
• Duplication: Without sufficient visibility across departments, teams can invest time and resources in tools or use cases that solve similar problems. For example, one publisher we worked with had multiple transcription tools in use across the same organisation.
• Integration and scalability: Independent teams may choose different tools, data approaches and technical solutions. Over time, these can become difficult to integrate, maintain or scale.

3. Centre of Enablement: connecting central expertise with local experimentation

How it works

Sitting between the two models is the ‘Centre of Enablement’. It combines a central hub for strategy, governance, shared standards and specialist expertise with AI capability embedded across different departments. Local teams have room to identify and explore use cases relevant to their needs, while the central function provides guidance, connects expertise and may retain responsibility for higher-risk approvals.

Representatives from different teams can exchange learnings and ideas, raise technical or compliance questions, and take relevant information back into their own departments. The central function does not own every experiment but provides a route for teams to access expertise and maintain visibility over what is happening.

Benefits

• Agility and cross-functional collaboration: Teams retain much of the autonomy of a decentralised model, while regular forums give them access to central expertise. A newsroom team, for example, may be able to explore a use case locally while drawing on product, data or legal support.
• Scalability: The central hub can identify local experiments, tools or practices that have wider relevance and help spread them across departments. This creates a route for successful ideas to move beyond the team that first developed them.

Trade-offs

• Coordination complexity: Responsibilities need to be clear: what can teams decide independently, what needs central input, and when should an experiment be escalated for review? Regular forums can also become burdensome if representatives do not have enough authority, time or support to bring information back to their teams.
• Reliance on active participation: The model depends on the quality of the connections between the central and individual departments. This model only works if information and learning move in both directions.

Why is the ‘Centre of Enablement’ becoming more relevant?

The right structure depends on factors such as organisational size, AI maturity, internal capabilities, culture and risk, to name a few. In our work with publishers, however, we are increasingly seeing interest in elements of the ‘Centre of Enablement’. Its appeal is practical. Local teams can move quickly and identify use cases that fit their workflows, while central expertise, shared guardrails and cross-functional coordination help maintain alignment.

This does not make the ‘Centre of Enablement’ universally relevant. For a publisher at an earlier stage of its AI journey, a more centralised model may provide useful clarity. For an organisation with strong AI capability in teams, greater decentralisation may be appropriate. The important point is that governance should support the organisation's current needs.

Governance becomes more important as AI scales

Scaling AI changes publishers’ operating model, creating the need for sufficient central coordination to manage risk and avoid duplication, but enough distributed capability for teams to innovate in their own workflows. For many organisations, this points to a federated model in which a central function sets the guardrails and connects expertise, with distributed teams owning experimentation, adoption, and knowledge sharing.

Through our work assessing AI readiness with more than 50 publishers, we are building a growing view of how these governance choices play out in practice, alongside first-hand experience from the Financial Times.

If you are reviewing how AI is governed across your organisation, please get in touch.

At FT Strategies, we help publishers move beyond AI experimentation to responsible, organisation-wide adoption. Drawing on our experience assessing AI readiness, we can assess your current capabilities, identify gaps in your operating model and design an AI governance approach that fits your strategy and culture. Get in touch to explore how we can help your organisation scale AI with confidence, ensuring innovation is balanced with effective oversight.