HSBC Head of Product Data Reimagines the 'Single Source of Truth' as a Federated Ecosystem

Even as enterprise AI investment accelerates, a fundamental problem remains: most AI projects still fail, and the vast majority of those failures are tied directly to data. Now that the underlying technology issues—cloud, distributed architectures, data lakes—have largely been addressed, the focus is finally shifting to data. For leaders, the real challenge will be turning those raw inputs into the trusted intelligence that reliable AI demands.

To understand the task at hand, we spoke with Grace Wu, Head of Product and Account Data Asset at HSBC. A global data and AI transformation leader with over 15 years of experience in the financial services industry, Wu specializes in building the enterprise-scale data assets that power digital transformation. In her view, Master Data Management (MDM) is now the foundation for AI success—a major evolution from its traditional role as a back-office data hygiene tool.

For Wu, this new reality prompts a fundamental rethinking of the source of truth itself. "With the development of AI, cloud, and distributed architecture, the 'source of truth' is no longer a static database," Wu says. "It's a dynamic, federated ecosystem." In place of a single, centralized database acting as the undisputed truth, the future is giving way to a far more interconnected model.

The old "garbage-in, garbage-out" mantra takes on new urgency in the age of AI, Wu explains. "Poor data leads to AI bias, errors, reputational risk, and AI hallucinations." And with threats ranging from algorithmic bias to public reputational damage, the risks and the consequences can be significant.

• The Responsible AI rulebook: According to Wu, the built-in quality controls and auditability of MDM make its pre-processing work foundational. "This standardization in data processing, quality controls, and governance is the pre-process that supports AI for decision-making. It's what makes the AI's output more reliable, explainable, and responsible—which is what the whole industry is talking about: Responsible AI."
• Connecting the dots: In this model, the ecosystem functions as an intelligent network held together by a unifying semantic layer and a standard set of rules. Meanwhile, metadata allows AI to understand its full context, Wu explains. "Metadata, on top of clearly consistent data definitions, is what allows an AI to truly understand where data can be captured, how to utilize it, and how to build strong algorithms for decision-making."

By helping manage the complexity of a global product footprint, the federated model can deliver direct business impact. "An AI may be easily deployed in one country," Wu says. "But, to deliver a global product, you have to consider global regulations. This creates a complicated architectural setup with obstacles to making AI work at scale."

• The scaling bottleneck: "A single product may have different naming conventions or restrictions in different jurisdictions," Wu continues. "A proper MDM data model allows that product's data to be stored consistently at a global level while still being customized for each country's specific requirements."
• The co-pilot model: But implementing this new architecture often requires a culture change that goes well beyond the technology, Wu explains. "In today's world, it's about co-piloting with the business, which must be a full participant in sample testing and stress testing. They provide the feedback loop for iterative improvement before a product really goes into the production stage."

Next, Wu describes a process-driven approach born from the realities of operating in a highly regulated industry, where errors can be costly. Before any tool is chosen, she says, a detailed assessment takes place, focusing on ROI, technical feasibility, and change management.

• A human touch: From there, the work moves to prototyping and iterative MVP testing, with the business involved at every step to fine-tune the product before it goes into full production. "For an AI's result to be a meaningful, precise insight, it requires a human in the loop to tell the AI if it's working properly and behaving like a human being."

Ultimately, a collaborative culture that spans the entire organization is a key component for success. "The keyword is 'we.' We have to work together," Wu concludes. "That includes business leadership, IT, legal, compliance, and specialists in the respective functional domains. Everybody has to contribute their domain knowledge to get that AI product to a mature stage."