ITV and Funding Circle on Making EA Stick and Owning AI Governance

Hosted by Heidi Mattison, Ardoq’s Chief Customer Officer in London, this fireside chat brought together two architects who’ve spent years turning their enterprise models into something the wider business actually uses.

Robert Taylor is Director of Architecture at ITV, the UK’s biggest commercial broadcaster and a major international production company. He’s been at ITV for about 15 years, the last 10 in the core Enterprise Architecture team. ITV has used Ardoq as its central enterprise model for four years. Today it underpins the architecture review process, application rationalization, and, increasingly, AI governance.

Mark Davidson is Principal Architect at Funding Circle, one of the world’s largest business lending platforms, connecting investors with SMEs across the UK. He joined five years ago after several domain architecture and technology leadership roles at Lloyd’s of London. Funding Circle has used Ardoq for about three and a half years. It keeps the model current by pulling from systems of record like GitHub, its people system, and its contract management system. That model has supported application rationalization and the due diligence behind the divestiture of its US business, and it now feeds Funding Circle’s work on data as a product and AI governance.

Between them, the two covered how EA earns its place with the business, what it takes to govern AI without slowing everyone down, and why agentic AI depends on an enterprise that machines can actually read.

Watch the full session below or read on for the highlights:

How to Overcome Cultural Resistance to EA

Anchoring Architecture in Live Business Priorities

In Mark’s experience, the organization tends to view architecture in very limited technical terms, and EA involvement in business strategy and organizational design can meet resistance and skepticism. It’s sometimes perceived as architecture teams overstepping their area of expertise.

Often the instinct for EAs then is to strengthen the appeal to reason by doubling down on the architectural basis for their point of view in a given conversation. Unfortunately, this bombardment of technical detail is more likely to confuse rather than compel stakeholders outside of the architecture team. He recommends that EAs fight that instinct and go in the opposite direction, simplifying how they communicate some of these concepts to get buy-in and ease the human side of the transformation process. It’s important for EAs to always anchor discussions with other teams in board-level business topics, be it AI adoption, time to market, or cost savings.

Focus on Visibility Over Perfection to Show Rapid Value and Build Trust

Robert agreed with Mark and added that for ITV, Ardoq had been vital in shifting behaviors and therefore the culture around EA. Implementing the platform allowed them to codify their enterprise model in a collaborative system. Instead of being a diagram in someone’s head or in an inaccessible repository, as is often the case with EA in the past, Ardoq allowed them to share and collaborate on it across the organization. Even though they didn’t have every detail and data point captured at the early stages, this greater transparency already greatly improved alignment and collaboration with stakeholders.

This change in approach was triggered by an internal audit of their estate four years ago. The audit revealed a lack of clarity over technical ownership, and they found that documentation was spread across different pockets and repositories across the organization in varying formats. Some information was logged in Confluence, some in spreadsheets, but they had no single, audit-ready source of truth.

Implementing Ardoq helped them to align and standardize definitions and concepts across the business, such as what an application is and what technologies mean for ITV, and, by doing so, helped drive different behaviors and cultural change around what EA is able to do for the business.

The value of EA input became even clearer in their 2024 cost-saving exercise, where they were able to expand what they had already built around applications, estate, and technical fit scoring to tackle application rationalization and IT cost management. ITV’s leadership was doing a zero-based budgeting process, which could then be used to identify what was missing from the application estate and what those gaps were costing them. They could then use the Gartner TIME model overlaid with cost information to make the application rationalization process clearer, more accountable, and more transparent. ITV’s story is the perfect example of where instead of pursuing complete documentation, focusing on one work stream or critical area helped to more quickly establish EA’s rapport with leadership and other teams such as finance and operations.

How EA Helps with Rationalization and Tackling IT Complexity

Rationalization in financial services usually conjures a familiar picture: a large firm with a complex IT landscape that has undergone several M&A resulting in siloed business units each running their own overlapping systems. Funding Circle’s complexity instead sprang from heavy reliance on bespoke code from in-house engineers. This meant that rationalization for them was quite different from the usual relatively low-hanging fruit of duplicate systems serving the same capabilities.

Despite this difference, their rationalization toolkit is still largely the same as most organizations: Gartner’s TIME analysis and mapping business capability models against the application portfolio. What changes is how hard it is to agree on where the redundancy actually sits. Overlap between a handful of large off-the-shelf systems is easier to spot compared to across a portfolio of microservices and custom-built components.

The clearest proof of EA’s value in rationalization came from a much bigger strategic decision. When Funding Circle chose to divest its US business, due diligence required a complete view of how the organization’s structure connected to its application portfolio, its suppliers, the agreements behind those suppliers’ software, and its infrastructure. Mark’s team drove that process. It was painstaking work, but it changed how the rest of the business saw EA.

“For some people, this was sort of an aha moment. It’s like, all right, that’s what you guys do.”
- Mark Davidson, Principal Architect at Funding Circle

The result, in his words, “has been absolutely transformative.”

How EAs Can Own the AI Governance Challenge and Drive Safe Organizational Adoption at Speed

In Gartner’s 2026 guidance for heads of Enterprise Architecture, it argues that EA must drive AI direction, and it names establishing AI governance as one of three steps on that journey, alongside explaining EA’s role in AI strategy and rapidly discovering AI opportunities.¹

ITV got there in two phases. The first was simply working out how to describe AI. At the time, the pressure from the business was to simply tag certain applications as “AI” or “GenAI” and report on those. Robert pushed back with the argument that tagging meant they wouldn’t be able to separate out different AI capabilities or understand what sort of data AI was accessing where and when. Generative AI that produces imagery and video and a statistical model that filters spam carry very different risks for the business. His stance was that more detail would help them understand the real risks and what ITV needed to be concerned about.

So instead his team built an AI capability model. Like a business capability model, it describes what AI does rather than which tools deliver it. Robert started from an established EU taxonomy, found it too complex to work in practice, and stepped back to something detailed enough to show which applications and technologies deliver which AI capabilities without becoming unusable. That gave legal and privacy a reporting and audit view of AI across ITV’s estate: where the risks might be, and where they probably aren’t.

The second phase came when generative AI features started appearing in almost every application. ITV already had a mature architecture review process, where any new technology goes through divisional and, if needed, group review, and has to be recorded in Ardoq. AI governance initially sat on top of that, and it was heavy and manual: a weekly review involving senior legal, privacy, cyber, and architecture stakeholders, fed by a Google Form, a Google Sheet, and a Jira queue. This was slow and very resource intensive.

So ITV made Ardoq the control plane for AI capability assessments. Using intelligent automation, they made the intake form dynamic, drawing on the applications in Ardoq, the data concepts maintained by ITV’s enterprise data architecture team, the business capabilities the use case supports, and the AI capabilities it will deliver. Submitting the form triggers an automated process with an LLM at its core. It then queries Ardoq for more context on each data concept, business capability, and application, including who owns it and which part of the business uses it, then produces an actionable, ready-to-review document for legal and privacy to sign off.

“There’s still a human in the loop, but we cut out hundreds of hours of time every week and month just through that process alone.”

- Robert Taylor, Director of Architecture at ITV

This closely aligns with what Gartner recommends, for businesses to treat AI governance as an alliance with EA supplying the context that legal, privacy, and cyber teams need to make decisions.¹

EA Teams Should Align With Business Strategy, But Not Be Tethered to It

“As architects, we should be in the business of providing options. We need to be aligned to the business strategy, but we shouldn’t be tethered by it.”

- Mark Davidson, Principal Architect at Funding Circle

Going into the pandemic, Funding Circle’s strategy was to take its successful term loan product into as many territories as possible. Coming out of it, the strategy flipped: focus on a single market and broaden the product portfolio. The technology platform had been built for the first strategy, but that close fit became a cost once the strategy changed, and it’s one Funding Circle is still working through today. In practice, that means EA teams should accept that strategy can change quickly and radically under outside pressure, and focus on building so the organization can respond quickly to those changes. For Mark, this means coming back to the architectural fundamentals and building on the relationships established with leadership.

“By aligning the software that we build and procure to our capability models, we can be in a better position to pivot when the business strategy inevitably pivots.”

- Mark Davidson, Principal Architect at Funding Circle

Organizations Need a Machine-Readable Enterprise

“We’ve got to start thinking about agentic AI as not just chatbots in single applications or AI-enhanced tools. No one wants a smarter kettle or toaster. People want a personal chef. And that means agentic AI capabilities aren’t going to live in just one application.”

 - Robert Taylor, Director of Architecture at ITV

Most organizations’ experience with generative AI so far has happened inside individual tools: an assistant added to one application, or an AI feature in a specific piece of software. Even beyond those tools, people paste context into general-purpose AI assistants and get one-off answers in return. Robert’s view, shaped by ITV’s work over the past year, is that none of this scales. For AI to succeed, it has to be embedded in how the organization works.

The operational reality is that people don’t work exclusively in application silos. A single task can involve several business systems and contexts, so an agent that only knows one application can only help with part of the job. For agentic AI to support how the business actually works, it needs context from across the organization. That means knowing which applications exist, what data they hold, which capabilities they support, and who owns them.

This is where Robert sees EA’s role changing. A machine-readable enterprise, especially one built on a graph, can provide that context to any agent that needs it. Enterprise architects already maintain that model. What’s new is that it now feeds AI agents directly, not only the people reading reports and diagrams.

“Being able to bring context to those agents is a massive change.”

- Robert Taylor, Director of Architecture at ITV

Robert has a clear picture of where this leads. Asking people to remember which application to open in order to use which agent “is bonkers,” he said. “It’s not going to last.” He expects a single interface where employees work with a set of internal agents, and each agent gets the context it needs at inference time from a graph that connects related concepts, rather than from standalone text files describing each function. In his words, the goal is to bring the models to the context, not the context to the models.

An attendee from a heavily regulated organization raised a practical version of the same problem. Regulators love diagrams, and teams increasingly produce them in code-based tools like Mermaid. Mark has had the same debate with Funding Circle’s engineers. Diagrams in code are better than hand-drawn ones, but the components in them often don’t match anything in a repository or the org structure. “Getting from a good, very fully realized graph model to a visualization language like Mermaid is much easier than vice versa,” he said. Start from the graph, and the diagrams take care of themselves.

The need for a machine-readable enterprise changes EA’s relationship with the functions that protect the organization. If risk, controls, policies, and regulations are going to shape how AI agents behave, someone has to encode them in the model, accurately and in the language those teams use. For Robert, that all comes back to building trust and relationships with other teams. Legal, privacy, cyber, and compliance need to be confident that EA can represent their requirements faithfully. Once that’s in place, the model can carry those rules into every AI interaction. “Bringing that to the context of those AI models,” he said, “is, I think, the biggest shift [in EA].”

Sorting the Basics: Capturing Data Points and Populating the Repository

Both architects clearly have well-populated repositories, which led to an obvious question from the audience: where does all that data come from, and who keeps it up to date?

At ITV, the starting point was whatever already existed. During the initial implementation, the team brought in application information from the CMDB and a long list of spreadsheets. The business capability model had a similar history. ITV built one in 2016, but it sat untouched in an Excel sheet, disconnected from the application estate, until the team brought it into Ardoq and connected it.

Keeping that data current depends on process as much as tooling. ITV’s architecture review process, which runs at both divisional and group level, requires that any component shown in a design diagram or listed in a document is also recorded in Ardoq, with a reference back to it. Surveys do much of the remaining work. Robert’s team also holds monthly sessions with architecture and engineering teams to review data quality and catch drift, such as newly added applications with no lifecycle status or technical owner. ITV is also adding integrations with software asset management, the CMDB, FinOps, and GitHub to pull information into Ardoq automatically.

Funding Circle takes a different approach, mainly because of the team they’re engaging in the process. Its large engineering function has little capacity to update records through a UI, so wherever possible, Mark’s team pulls from systems of record instead. In-house application information comes from GitHub through Backstage while people data comes from the HR system. Supplier and agreement details are pulled into Ardoq from their contract management system. They deploy Ardoq’s Surveys for auditing organization-wide application access review.

To make things even easier for their engineers, Funding Circle is exploring domain-specific languages that let teams define architecture information inside their own repositories, so it’s picked up automatically through build pipelines. Some artifacts, such as standards, decisions, and principles, work better as written documents. Those stay in Confluence and are synced to Ardoq so they also exist as components in the model.

The two approaches start in different places: ITV leans on governance and surveys, while Funding Circle leans on automation from systems of record. Both are defined by the same principle, though. The repository stays useful only if keeping it current is part of how work already happens.

Top Takeaways for EA Teams

1. Lead with the business problem, not the metamodel. When faced with skepticism or resistance, try to simplify the message and connect every conversation to a board-level priority, such as AI adoption, time to market, or cost savings.
2. Aim for visibility, not perfection. Choose a single critical workstream, show results there, and let that build the relationships EA needs with finance, operations, and leadership.
3. Model AI by what it does and what it touches. Build an AI capability model detailed enough to show real risk so it can be a reliable resource for legal, privacy, and cyber assessments.
4. Build for options, not just today’s strategy. Strategy can change quickly so align what you build and buy to stable abstractions like capability models, so the organization can pivot.
5. Make the enterprise model machine-readable and self-maintaining. AI agents will need context from across the business, and a knowledge graph like the one powering Ardoq can supply it in ways diagrams and text files can’t.

See how Ardoq supports organizations with AI governance: AI Lens Solution

¹ Gartner, Heads of Enterprise Architecture Must Drive AI Direction, Duncan Robson, 2026.

GARTNER is a registered trademark and service mark of Gartner, Inc. and/or its affiliates in the U.S. and internationally and is used herein with permission. All rights reserved.