Scaling AI in Media

Scaling AI in Media

An Operating Model for Media Leaders Turning AI Ambition into Measurable Business Value

Most media companies now use AI somewhere in their operations. Yet very few can point to it as a measurable driver of enterprise value. The gap isn't the technology; it's the absence of an operating model connecting strategy, governance, platform architecture and workforce enablement.

Media leaders who want AI to move beyond isolated pilots need to do four things in parallel: tie AI investment to a clear business outcome and a KPI framework, establish governance before scaling, consolidate AI on a shared platform rather than using one-off tools and treat change management as a core competency. This article outlines that operating model and a practical agenda for the next 6–12 months.

Why AI Momentum Isn't Translating into Enterprise Value for Publishers

AI adoption in media is no longer in question. A Thomson Reuters Foundation survey of more than 200 journalists across 70 countries found that over 80% use AI tools in their work.

However, audience trust has been hard for publishers to gain. The Reuters Institute’s Generative AI and News Report 2025 found that only 12% of respondents are comfortable with fully AI-generated news, compared with 62% for entirely human-made content. Acceptance rises when journalists remain in the loop, but only 33% of respondents believe journalists routinely check AI outputs before publication. For media brands, that makes visible human oversight and transparent AI use a commercial issue, not merely a compliance exercise.

Financial returns have been equally hard to demonstrate. MIT’s 2025 research found that about 95% of enterprise AI pilots fail to produce a measurable return. While that figure isn’t specific to media, it nonetheless accurately describes the challenge most leaders can relate to: results that depend on one team's expertise, tools that don't survive contact with a second use case and a widening list of pilots that never make it to production.

Meanwhile, the technology isn't standing still. Agentic AI systems that plan and execute multi-step tasks are already shipping, even though many newsrooms haven't finished operationalizing the basics of generative AI.

For a CIO or CDO, the practical question isn't "should we use AI?" It's this: What must be true organizationally for the fifth AI use case to be cheaper, faster and safer to ship than the first? If the honest answer is "nothing has changed since pilot one," that's the signal an operating model is missing, not a technology problem.

Three Gaps Behind the Pilot Trap

Three structural gaps keep showing up in the data, regardless of company size, market or maturity.

Organizational integration is weak. In the Reuters Institute’s survey of U.K. journalists, 60% said their main news outlet had established AI protocols or guidelines for at least one key issue, including human oversight, data privacy or transparency. Yet only 27% reported guidelines covering bias and fairness. The pattern is familiar: High-level principles are increasingly common, but the practical guardrails needed to apply them consistently across editorial workflows often lag behind.

Technological maturity hasn’t caught up with the ambition. The AI vendor market is still consolidating, much like the early internet era. Media companies that had already simplified their IT architecture and cleaned up their data before the GenAI wave are scaling faster now. Companies still running fragmented systems are learning that a working prototype and a production-ready capability are two very different things. AI also expands the attack surface just when many security teams are least prepared to review it. Three of the risks that come up most often are prompt injection, where malicious inputs manipulate an LLM’s responses or logic; data manipulation, where source data or fine-tuning is tampered with to introduce vulnerabilities or bias; and overreliance, where teams make decisions or publish content based on LLM output without adequate human oversight.

Expertise hasn't caught up with intent. The Thomson Reuters Foundation survey identifies a persistent lack of AI training and guidance, with most respondents saying their workplace had no clear AI guidelines. A team that cannot critically assess AI output is also the team least able to build the guardrails needed to catch errors before publication.

The cost of getting this wrong is not hypothetical. Sports Illustrated, Gannett and CNET have all had to walk back AI-generated content publicly after quality or attribution failures reached readers. Each incident traces back to the same root cause: AI was deployed faster than the governance needed to supervise it.

The Operating Model: Four Foundations for Scalable AI

Value doesn't appear in one jump from experimentation to enterprise scale. It builds in stages. Whether an organization reaches the later stages depends on having four foundations in place before scaling starts.

1. Make AI strategy part of business strategy, not a separate track

AI strategy fails the moment it becomes its own workstream. It needs to come directly from corporate and IT strategy, and answer one question before any technology decision is made: What specific business outcome, such as growth, margin, differentiation or a new revenue line, is this investment meant to move?

That question changes how you measure success. ROI still matters, but on its own it's a lagging, incomplete signal. A better KPI framework also tracks operational or editorial efficiency gained, content quality, engagement lift from personalization, staff adoption and how often people actually use the tool, and how stable the model's output is over time. Teams that skip the adoption and quality metrics often find out too late that a technically successful pilot never got used.

2. Put governance in place before you need it

Governance is often seen as a constraint on innovation. In practice, it lets you move faster because it replaces a case-by-case debate with a standing framework. The EU AI Act's explainability requirement is a useful baseline here. In a journalistic context, you need to trace how a data set and a model produced a given output. That matters for compliance, but it matters just as much for editorial accountability.

The elements that matter most: clear, published principles on data quality, fairness, privacy and explainability.

To put these principles into practice, organizations need:

• An interdisciplinary governance board covering legal, IT security, data protection and the business, led by a named owner such as a Chief AI Officer
• Visible sponsorship from senior leadership
• Practical tooling, including quality checks, monitoring dashboards, training and an incident response process
• Periodic review against an external standard such as ISO 42001

3. Build AI on a shared platform, not a collection of once-off tools

Every isolated use case, a chatbot here, a summarization tool there, starts fast and stays small. It rarely gets reused, rarely meets consistent security standards and rarely has its costs controlled. What scales instead is a central AI platform: one place where model access, prompt libraries, agent configuration, data connections and collaboration tools are managed as shared infrastructure instead of rebuilt for every project.

A mature platform typically covers four things: access to multiple LLM providers and self-managed models, configuration and deployment of AI agents against defined workflows, governed access to internal and external data (including retrieval over your own proprietary content), and secure, centralized collaboration tools. Done well, it also enforces your governance model. It controls which providers and use cases are approved, standardizes security checks and gives finance one place to view and manage AI spend.

You don't need a big-bang build to get there. The platform can grow step by step, organized around infrastructure, data, models, serving and security, rather than around individual use cases (see also IT Architecture Transformation for a Media Company).

4. Treat change management as a core AI competency, not a rollout afterthought

The BBC's approach to AI-assisted drafts and summaries keeps journalists responsible for reviewing and editing content before publication. Similarly, the Washington Post’s Heliograf combines automated updates with editorial intervention, allowing journalists to add analysis and override machine-generated text. These examples highlight the need to equip people with the skills to evaluate AI output and exercise sound editorial judgment.

The pattern is the same in both cases: Capability isn't something you deploy; it's something you build with the people who'll use the output every day. That means training tailored to specific roles, timed to land alongside real implementation work, not months before or after.

Are You Ready to Scale? A Readiness Check

• Is your AI strategy clearly derived from, and regularly checked against, your corporate and IT strategy, with near-term and mid-term goals defined?
• Is there at least an informal governance process to decide which use cases get built, and in what order?
• Can your core IT function support new use cases under existing guidelines, rather than treating each one as a custom project?
• Can you demonstrate, rather than assume, that compliance requirements are being met for what's already in production?
• Is there a concrete plan to build the skills each new initiative will need before it launches?

A "no" to any of these doesn't mean stop. It tells you where your next investment should go.

The Executive Agenda: Next 6–12 Months

First, make the business case explicit. Select two or three AI initiatives already underway and ask the KPI question directly. Pause or stop anything that can't answer it.

Second, establish minimum viable governance. You don't need a finished framework. You need a named owner, published principles and a review checkpoint before any new use case goes live.

Third, start consolidating. Identify the two or three technical components that repeat most often across your current AI initiatives, such as model access or data connections, and move them onto shared infrastructure first - that's the start of your platform.

Fourth, build skills alongside delivery, not before delivery. Tie training to the specific initiatives already underway, and bring in the teams who'll use the output from day one.

Conclusion

AI in media has moved from experimentation to strategic necessity. What determines whether that potential becomes a competitive advantage is structure: a business-aligned strategy, governance built ahead of scale, a consolidated platform and workforce capability developed alongside delivery.

Having the proper structure in place turns a portfolio of AI pilots into a sustainable source of business value.