The financial world is in the throes of a generative AI (GenAI) revolution. Boardrooms echo with promises of hyper-personalized customer experiences, automated operations, and unprecedented efficiency gains. Industry conferences showcase chatbots writing policies, algorithms predicting market shifts, and large language models (LLMs) drafting investment advice. But behind these glossy presentations lies a sobering truth: Many organizations are building their GenAI castles on sand.

Having led large-scale data and AI transformations across global markets, I’ve witnessed how fragmented data strategies sabotage even the most ambitious AI initiatives. The missing link in many of them is a unified, trusted data foundation.

The illusion of progress

Imagine this: A customer calls their bank to dispute a transaction, only to learn the call center agent sees a different account balance than the mobile app. Or, an LLM recommends a personalized insurance policy based on outdated claims data. These aren’t hypotheticals — they are daily realities for institutions that prioritize AI tools over data integrity.

Many companies are treating GenAI as a standalone miracle cure, delegating it to tech teams while neglecting the underlying data infrastructure. This approach results in flashy pilots that fizzle out because they’re built on siloed, inconsistent data.

Consider the three silent killers of GenAI ROI:

1. Customer distrust and loss of loyalty caused by conflicting data across channels (apps, call centers, agents), can quickly undermine the success of a GenAI initiative.
2. Paralysis of data analysts who waste the majority of their time cleaning and organizing data instead of deriving insights rapidly erodes GenAI momentum and profitability.
3. And finally, GenAI ROI is often undermined by the irregularity of LLMs trained on messy data that produce unreliable, and even risky outputs.

Without a single source of truth, GenAI can quickly become a liability, not an asset.

A 3-step blueprint for a lasting foundation

A mature data strategy isn’t about buying the latest AI tool. Rather, it is about designing systems that serve all users, from customers to analysts, and machines. There’s a clear, three-step blueprint that visionary leaders are using to design a data system that will meet user needs:

1. Demolish data silos

Global financial institutions may have hundreds of systems collecting customer data from sources such as policy administration platforms, claims databases, and mobile apps. Each data source operates in isolation, creating fragmented views of the same customer.

The alternative is to consolidate every touchpoint into one system of truth. This is both a technical lift and a cultural shift. The cultural shift occurs when organizations appoint cross-functional teams (IT, operations, compliance, customer experience) to collaborate on mapping data flows and enforcing governance. The technical lift is a 360-degree customer view that powers everything from chatbots to board reports.

2. Design for non-negotiable users

Most organizations have a multitude of current and potential users of data. Some departments use data to create new products, other teams need data for strategic planning. Some are critical to an organization’s profitability, while others play a secondary role.

While data users differ in their importance within an organization, there are three groups of non-negotiable users common to most large companies. In order to drive technology ROI, these three users must have easy, scalable access to a single source of truth, and they should be the core focus when an organization designs its data foundations.

Customers expect seamless experiences, and a singular data foundation ensures that a policy update on the app reflects instantly in call center systems. Analysts and data scientists need clean, harmonized raw data to build accurate models. Without it, forecasts become guesswork. GenAI LLMs like ChatGPT or Claude require structured, context-rich data in order to produce reliable and useful outcomes.

The common thread between the needs of these three user groups is consistency. When all non-negotiable users access the same truth, the pace of innovation can accelerate.

3. Own your data

It may be tempting for an organization to take the easy route of dumping raw, unstructured data into external LLMs and hoping for the best. However, this cedes control to external parties, inflates costs, and risks regulatory blowback.

The better alternative is to build internal knowledge graphs or vectors, which are structured repositories that organize unstructured data (emails, PDFs, call transcripts) into searchable, machine-ready formats. This lets LLMs generate insights grounded in your business context, not in generic patterns. Structuring your data unlocks scalability.

The ROI of getting it right

When data foundations align with GenAI ambitions, the magic of innovation can occur, and tangible ROI begins to materialize. Hyper-personalization becomes a critical marketing tool. Imagine an LLM that predicts a customer’s life events (marriage, retirement) and suggests tailored financial products.

A solid data foundation provides regulatory agility, enabling automated report generation that cuts compliance costs and audit risks.

Taking a more structured approach to your data foundation also builds investor confidence, and firms with auditable AI governance attract premium valuations.

A call for courageous leaders

The GenAI race will not be won by those with the biggest budgets or shiniest tools. Rather, it will be won by leaders who recognize that data integrity is the bedrock of innovation, and make decisions that prioritize data.

Is your data foundation an enabler or a liability?

Are your LLMs building trust or breeding chaos?

The answers to these questions will define your organization’s next decade. The time to act is now. Partner with data leaders who speak the language of both boardrooms and engineers. Invest in infrastructure that serves humans and machines. And above all, treat data not as a cost center but as the strategic asset of the AI era.

About the Author:

Abhishek Sharma is Group Data Leader at AIA Group. He is responsible for AIA’s data estate transformation, driving the maturity and enhancement of platforms and governance through strategic data, AI and analytics initiatives.