Bloomberg Professional Services
- Firms are shifting from collecting more data to defining the decision first, then aligning data to that outcome.
- The ability to connect, standardize and operationalize data is now the main differentiator.
- Multimodal models are expanding what’s possible, but their impact depends on well-structured, connected datasets.
- Firms that embed lineage and control into their data foundations can move faster and deploy AI with confidence.
The conversation around data strategy is shifting. For years, the focus was on accumulation. Now, the emphasis is moving toward how effectively data can be connected, governed and translated into decisions. The firms pulling ahead are those best able to align data with clear business outcomes.
Across financial services, this shift is reshaping how institutions evaluate datasets, design infrastructure and deploy AI. Rather than treating data as an asset in isolation, leading organizations are approaching it as part of an integrated workflow, where insight, not volume, defines value.
These themes emerged clearly during a recent Bloomberg Enterprise Data discussion, where practitioners across banking, quantitative investing and technology shared how their approaches are evolving in practice.
Start with the question, not the data
A persistent challenge in data strategy is what practitioners often describe as the “data deluge”: the trap of accumulating information without a clear view of the decision it is meant to inform.
Clarence Cheung, Global Head of Non Financial Risk Analytics, CIB, HSBC, emphasized that effective data strategies begin with a defined outcome. His team focuses on identifying the specific insight required, what he calls the “aha moment”, before determining which datasets are needed to support it.
“We start by asking what insight we want to draw,” Cheung explains. “And then we work backwards to figure out how to bring you to the ‘aha’ moment, and also at scale.”
This approach reframes how data is evaluated. Fitness for purpose extends beyond coverage to include regulatory usability, timeliness and the ability to align with existing data frameworks. In practice, this reduces noise and ensures that new data contributes directly to decision-making.
For Johnson Wu, Managing Director at Quantbot, the discipline extends further. His firm assesses datasets not only for immediate relevance, but also for durability, considering vendor stability and long-term viability before integration.
Importantly, data that does not meet immediate needs is retained. “Data that is not useful today does not mean that it’s not useful tomorrow,” Wu notes.
As analytical techniques evolve, previously marginal datasets can take on new significance, particularly as AI expands the range of usable inputs.
The integration challenge
If identifying the right data is one hurdle, integrating it remains another.
Fragmentation across vendors, formats and identifiers continues to be one of the most persistent and costly challenges for financial institutions.
Datasets often arrive in incompatible structures, requiring significant effort to standardize and connect. Without a consistent metadata framework, the result is duplication, inefficiency and degraded analytics.
Quantbot’s experience highlights what a more structured approach can achieve. By investing early in a metadata layer capable of linking datasets through common identifiers, the firm has significantly reduced the time required to onboard new data.
Where integration once took months, Wu’s team can now evaluate and incorporate hundreds of datasets within a six-month period.
“As long as you have an ID, doesn’t matter what it is, we can match it most of the time,” Wu says. “That helps to speed up the entire process.”
The implication is that integration is no longer just a technical concern, but a strategic differentiator in how quickly firms can move from data to insight.
AI as an amplifier
As data becomes more connected, AI is extending what can be extracted from it. The current inflection point is not just automation, but the ability to interpret multiple forms of data simultaneously.
Generative AI models can now process images, documents and unstructured text alongside traditional datasets, opening new analytical pathways that were previously impractical.
Cheung describes one application in financial crime detection. By combining geolocation data, imagery and entity relationships, his team can assess whether a business’s physical presence aligns with its reported activity.
“If I’m looking at a picture of a specific location, and I see that this should be a restaurant, but then it shows up to be a warehouse, then there’s some suspicion there,” he explains.
What would previously have required manual investigation can now be evaluated systematically.
Marcela Granados, Global Head of Insurance & Professional Services at Databricks, points to similar advances in investment research. She highlights an example where diverse datasets, from financial filings to alternative data, are brought together in a single analytical environment, allowing users to query information using natural language.
“The cool thing about it is that nowadays, anybody can ask questions around your data in natural language,” Granados says. “You go from automating some of those tasks to orchestrating real business output.”
This broadening of access is changing who can engage with data, while raising the importance of underlying data quality and governance.
Governance as an enabler
Governance is often viewed as a constraint on innovation. In practice, firms with well-structured data foundations are finding the opposite.
When data lineage, access controls and auditability are embedded into systems from the outset, governance can accelerate deployment rather than delay it. Granados notes that a significant proportion of AI use cases in financial services can meet compliance requirements when these elements are already in place.
“When you have multiple teams accessing the data for potentially different use cases, you don’t need to rebuild the tech stack every single time,” she says.
This reduces duplication and allows organizations to scale use cases more efficiently.
At the same time, requirements differ across functions. In financial crime, for example, explainability is essential. Every output must be traceable and defensible.
The challenge is not reducing governance but increasing the speed at which compliant solutions can be developed.
The road ahead
Looking forward, the value of data will increasingly depend on how well it is connected, contextualized and governed.
Cheung points to synthetic media as an emerging risk, where AI-generated content could challenge the integrity of information in financial markets. Wu, meanwhile, emphasizes the continued importance of transparency in models and data usage.
“You cannot be a black box. You need to show exactly what you are doing,” he says.
Granados highlights another shift: the convergence of AI-assisted development and governed data infrastructure, which is expanding access to data-driven analysis across organizations.
What ties these perspectives together is a common foundation. Data remains central to modern financial workflows, but its value is no longer defined by scale alone.
Firms that invest in integration, governance and clarity of purpose are better positioned to translate data into decisions, and to adapt as the pace of change continues to accelerate.