If you’ve sat through enough “AI strategy” conversations, you’ve probably noticed a pattern. The models keep getting better, the tooling keeps getting easier, and yet many enterprises still struggle to turn AI workloads into measurable business impact.
The gap isn’t ambition. It’s enablement.
Business teams are ready to innovate with generative AI, conversational interfaces, and agentic workflows. They want to ideate, experiment, and move at market pace. But too often, they hit the same wall. The data they need is fragmented across systems, governed in ways that slow progress, and wrapped in complexity that keeps the art of the possible locked inside IT queues.
At AWS re:Invent, I sat down with Raja Palaniswamy, who leads enterprise data and data capabilities at MUFG, to talk about a simple—but powerful—shift: AI innovation will increasingly be led by the business, not IT. And the organizations that win will be the ones that make it safe, fast, and repeatable.
Watch the Full AWS re:Invent Session
In the clip below, Adrian Estala, Field Chief Data and AI Officer at Starburst, and Raja Palaniswamy, Enterprise Data and Data Capabilities Head at MUFG, discuss how business-led data products and a business-facing semantic layer enable faster, safer AI experimentation.
Business-Led AI Doesn’t Mean IT Disappears
Let’s be clear. This isn’t about handing machine learning pipelines or MLOps responsibilities to every line of business. That’s neither realistic nor responsible.
What is realistic is enabling business teams to safely explore GenAI and agentic experiences—grounded in trusted, governed enterprise data that they finally understand. When that happens, innovation accelerates naturally.
You can feel this shift when you walk the room at re:Invent. Product leaders, operators, and executives aren’t asking for another architecture diagram. They’re asking:
- Can we try this new agent?
- Can we ask questions in natural language and trust the answers?
- Can we prototype without waiting months?
When the answer is “not yet,” it’s usually not a tooling problem. It’s a data problem.
A Simple Framework for Business Enablement
In the session, I outlined a framework I’ve used repeatedly—both as a former enterprise CDO and now in my role at Starburst. It’s simple, but powerful when all three layers work together.
1. Access: Federate Data Where It Lives
Data lives everywhere—in the cloud, on premises, in data lakes, in warehouses, and inside SaaS platforms. If every AI initiative depends on migrating data first, experimentation slows to a crawl.
Federation removes that barrier. When teams can access data where it already lives, they can start building and testing immediately—without a multi-year prerequisite.
2. Understanding: A Business-Facing Semantic Layer and Data Products
Access alone isn’t enough. Business teams need to understand the data they’re using.
This is where a modern, business-facing semantic layer makes the difference. When data is organized into logical, well-described data products that reflect how the business thinks, complexity disappears. Legacy systems fade into the background, and confidence replaces hesitation.
I often describe the goal this way. When a business user opens a data catalog or workspace, they should recognize their data—the same way you remember family members when you walk into a room.
3. Impact: Adoption and Experimentation with Guardrails
When access and understanding come together, business enablement follows.
Teams can use data products directly in BI tools, analytics workflows, and increasingly, AI-driven experiences. They can experiment quickly, iterate safely, and focus on outcomes instead of plumbing.
As Raja put it succinctly: Guidance, guardrails, and governance. Those three elements turn experimentation from a risk into a repeatable capability.
MUFG’s Pivot: From Consolidation to Enablement
MUFG’s journey illustrates why this shift matters.
Like many organizations, MUFG initially followed a traditional path: consolidating data physically, curating it centrally, and building from there. But over time, one pressure became impossible to ignore—time-to-market.
As Raja explained, the question became: how do we enable business teams to prepare and use data themselves, without routing every request through IT or a central data office?
The answer was a pivot toward logical data products aligned directly with business products. MUFG introduced a clear distinction:
- Core data products: enterprise-wide, curated foundations (such as customer, channel, or banking domains)
- Functional data products: derivatives built by business teams to answer specific operational or analytical questions
To support this, MUFG embedded business data producers within the business itself—people who understand both the data and the business context. The result is greater self-sufficiency, faster insight, and better alignment between data and decision-making.
Data Literacy is the Multiplier
Technology alone doesn’t drive adoption. Confidence does.
MUFG paired its data product strategy with data literacy initiatives to help business teams understand what data exists, how to access it, and how to use it responsibly. Some teams moved quickly; others needed time. That’s normal.
What matters is the outcome. Business users can now access accurate, timely, high-quality data without waiting in line—and that fundamentally changes how decisions get made.
AI for Data, and Data for AI
One of my favorite moments from the conversation was Raja’s framing: “I use AI for data, and I use data for AI.”
On one side, MUFG is enabling business users to interact with data conversationally—while enforcing strict authentication and authorization. On the other hand, AI is improving the data management lifecycle itself, from lineage and profiling to tagging and discovery.
This is where modern data strategies are heading. AI isn’t something you layer on after the fact. It becomes part of how you govern, understand, and activate data at scale.
Why Experimentation Beats Perfection
Here’s the takeaway I want leaders to internalize: experimentation is a strategic advantage.
If your answer to “Can we try this?” is “Yes, in 18 to 24 months,” you’ve already lost momentum. Not because your teams aren’t capable—but because the pace of innovation won’t wait.
I’d rather see ten AI experiments in ten weeks than one AI project in eighteen months. Some will fail. That’s expected. But you’ll learn faster, incur lower costs, and identify what truly creates value.
With the proper guardrails in place, experimentation isn’t reckless—it’s responsible.
The Question Every Leader Should Ask
If you want to pressure-test your AI strategy, ask this:
What could our business achieve in twelve months if teams had trusted, governed data products they could use without waiting on IT?
That’s the shift. Strategy first. Enable the business. Let experimentation drive progress.
Ready to Put Business-Led AI Into Practice?
If you want to see how a modern data product strategy can empower business teams to experiment with AI—without sacrificing governance or control—start with Starburst Galaxy.
👉 Start a Free Starburst Galaxy Trial
Explore how data federation, business-facing semantics, and governed access can help your teams move faster with confidence.
