With more than 165 years of operating history, over $200 billion in assets, and a workforce of roughly 23,000 employees, M&T Bank sits at an intersection many legacy financial institutions now face: how to modernize data and AI capabilities at enterprise scale without losing control, trust, or execution discipline.

In this second installment of a three-part conversation, Andrew Foster, Chief Data Officer at M&T Bank, continues his discussion with Parker Thompson, Regional Vice President and General Manager, East at Denodo, shifting the focus from strategy design to execution realities. While part 1 explored M&T’s data strategy, the rollout of AI tools to 17,000 employees, and the governance foundations required to protect customer trust, Part 2 goes deeper into how those ideas are operationalized across the organization.

Foster unpacks the less visible work behind data transformation: stakeholder engagement, federated accountability, pacing organizational change, and building credibility through execution. He also explains how its stewardship model underpins AI readiness, and how a pragmatic two-year strategy horizon helps balance ambition with delivery.

Edited Excerpts

Q: How do you engage stakeholders across the business, especially given that data sits between technology and business teams?

A large part of the success of our data program has been twofold. First is the organization’s willingness to lean in and support what we’re doing. A common challenge for CDOs is getting a voice at the table and getting people invested. That simply hasn’t been a challenge here, and that says more about the culture of the organization than anything else.

Second is building out a strong federated model with data sponsors, data stewards, and data owners. As we pivot more toward AI, the line I use consistently is, “Well-governed data is AI-ready data.” I said recently to our data steward community — we had about a hundred of them gathered — that their job just got about 25% cooler. They’re no longer focused only on well-governed data; they’re focused on AI-ready data. It sounds different, but it’s really the same thing.

That tight integration gives us a strong starting framework for engagement. We’re able to leverage senior leaders across the organization who already have data accountabilities. And that’s a starting point, not a destination. Those people bring high-impact, high-relevance thought leadership, and from there, executive leadership decides who else needs to be involved. The key is that you’re not starting from scratch. You’re starting with people who are already grounded in what you’re trying to achieve.

Q: Some organizations describe this kind of federated ownership model as data mesh. Is that a term you use at M&T?

We don’t really use the term data mesh. What we do have is a strong approach around cloud-enabled data products. On top of that, we deploy a medallion architecture and establish clear accountability at each layer.

To break that down, the bronze layer is fairly raw, unformed data. The silver layer is more curated. The gold layer is highly curated and aligned to specific business use cases. At the gold layer, ownership is very much on the business side — business data, business data products. At the earlier stages, it’s more technology- and data-team enabled.

Q: Are data stewards involved across all layers of that architecture?

They’re always involved because they’re the key contact point for their part of the organization. That doesn’t mean they do all the work or carry all the accountability, but they are key influencers, and they do have defined responsibilities.

We’re talking about roughly a hundred people in formal federated roles, but obviously, there are many more people across a 23,000-person organization who use data. Not everyone has that defined role. People working in analytics or data science engage with my team differently, but they still benefit from the increased AI readiness and stronger governance applied to the data they rely on.

AI is just one example. This applies to anything. Even a basic report follows the same rule: garbage in, garbage out. The data steward role is a major enabler in preventing that from happening.

Q: You’ve actually executed this model in a live production environment. How did you think about timing and the pace of organizational change?

I didn’t need to make major organizational changes. What I needed to do was invest in the data team, so we had the right senior leadership, coverage, and depth across the capabilities we wanted to build. Data architecture is a good example. We needed to bring in a leader and build a team underneath that function.

So it was less about organizational change and more about organizational investment. I’m very deliberate when I talk about a two-year horizon. People don’t tend to think beyond two years in their careers. And at the same time, one year is often too short to deliver meaningful change.

Two years gives you enough runway to execute, but it doesn’t feel like some distant, aspirational future. It’s also important to understand that it’s a two-year strategy, not a fixed two-year task list. The goal is to build a resilient data organization that can adapt to changes you didn’t anticipate, whether that’s shifting priorities or new strategic demands.

Q: So the two-year horizon is as much psychological as it is operational?

One hundred percent. I did a one-year review and a two-year review, and now we have a new data strategy that runs through 2027. I was able to say, the things we committed to in 2023, we delivered.

That creates a positive affirmation that the organization can execute. And when I say “we,” I don’t mean just the data team. This is the data team working with hundreds of people across the bank. But collectively, we were able to say we delivered on the first promise. Now we’re moving on to the second promise.

That second phase is about maximizing value from the new capabilities we’ve put in place, like data observability tools. We’ve also implemented a value capture framework. It gives us a common way to decide where to invest time, money, and effort so we can maximize return.

Q: It almost sounds like a compact with the business around how value will be measured.

Exactly. In the CDO role, you’re judged on execution, but you don’t own all of the execution. You have to influence a lot of people. I often describe it like a computer game where you have a health bar, and that bar represents your credibility. If it drops to zero, that’s why the average tenure of a CDO is only about two and a half years.

It’s critical that when you lay out a strategy, you can actually deliver on it, and that you have partners who can help you get there. Otherwise, you start taking credibility hits. I feel fortunate that we’ve been successful over the last two years, and again, that success ties back to our federated data steward model and our technology partners.

Q: Looking ahead, where are you focusing next across your AI capability areas?

It’s less about a single next step and more about running several activities in parallel. One capability every organization needs is the ability to interrogate unstructured data — things like PDFs, invoices, loan documents, and legal files.

We have three ways to address that. First, copilots can already do a good job of turning unstructured data into structured data for certain use cases, and that capability is broadly available across the organization. Second, we have other technologies powered by large language models that can handle similar tasks. Third, we could build a more common internal capability and deploy it at scale.

What matters more than the specific decision is the process we use to make it. We’re constantly evaluating where a copilot deployment gives us immediate productivity gains, while in parallel deciding where it makes sense to build internally or leverage existing applications. Everything we’re doing right now is inward-facing, focused on staff productivity and efficiency.

CDO Magazine appreciates Andrew Foster for sharing his insights with our global community.