Why context, not just data volume, is key to successful AI

"Data is the fuel for AI" has become a popular catchphrase. Unlocking it, integrating it, and governing has become increasingly important in the AI era. 

Financial institutions collect massive volumes of information — transaction records, market events, customer documents, system logs — but data volume alone creates no value. Structure and context turn raw data into fuel for business decisions and AI use cases.

Three financial services leaders representing institutions with a combined 640 years of history recently discussed this challenge at a Google Cloud industry panel. These executives from CME Group, KeyBank, and BNY revealed how adding context to raw data unlocks AI's true potential.

Context matters

Sunil Kataria, Chief Information Officer at CME Group, processes 30 billion market events daily. On a single Friday, his systems captured 5.5 billion market data events and 2.8 billion order entry events. CME stores 20 years of these snapshots.

"We're not in the business of just predicting, we're not in the business of forecasting," Kataria explained. "We're in the business of identifying worst-case situations for a given portfolio."

The value comes from adding historical context — understanding how markets performed during the 1980s stagflation, then applying that pattern to today's environment. Raw market events become useful when connected to specific risk scenarios. That's the refining process.

Three approaches to building context

Mining institutional knowledge at scale

Eric Hirschhorn, Head of Corporate Engineering at BNY inherited 240 years of institutional knowledge living in 3 million PDFs — documents containing valuable information but lacking structure. "Think about documents in boxes," Hirschhorn said. Without context about what each document contained or how it related to other information, those PDFs might as well have been in storage.

His team used machine learning and large language models to transform those documents into searchable knowledge bases, finding connections that were invisible when information lived in isolated files.

Connecting data to business outcomes

Ankit Goel, Chief Data and Analytics Officer at KeyBank, learned that data initiatives fail when they exist in isolation. His approach changed when he stopped treating data as a separate function and started embedding it directly into business strategy.

When KeyBank's commercial banking team works on accelerating loan approvals, they start with a business goal — faster client response times — then identify how quality data enables it. Using AI, they reduced analyst research time from 60% to 25%. "We're not going to work on any effort without having a line of sight into value," Goel explained.

Turning governance into a data layer

Before any team at BNY touches AI tools, their ideas go through a data engineering review board examining privacy, ethics, compliance, legal considerations, and data usage. Many organizations see this as bureaucracy. Hirschhorn sees it as adding essential context.

BNY's governance approach provides context about data provenance, usage rights, and appropriate applications. "When we explain that to our customers, they're fired up and they say, 'Hey, could you just tell us how you're doing that?'" Hirschhorn said.

Teams move faster because they understand the boundaries and requirements upfront. What looked like a bottleneck became a competitive advantage.

The refining advantage

Organizations that treat data as raw inventory will fall behind those that refine it into actionable intelligence. The key isn't volume — CME, KeyBank, and BNY all collect massive amounts of information. The key is understanding what that data means and applying it to real business decisions.

These financial services leaders understand the importance of refining data into the right context for the right application. Market data needs historical context for risk modeling. Transaction data needs process context for optimization. Unstructured documents need relationship context for knowledge mining.

AI accelerates context creation at scale and changes what's possible. Kataria's team models worst-case portfolio scenarios against years of market snapshots in ways that would not be practical in a manual process. Goel's team identifies transaction patterns that human analysts would never spot across billions of data points. Hirschhorn's team extracts meaning from 3 million PDFs that would take years for humans to review manually.

The technology doesn't just process data faster — it adds layers of understanding that were previously impossible. And the refining process works in both directions. AI needs refined data — data with context — to produce useful results. But AI also refines data by adding context at scale. The relationship is symbiotic.

Three principles for refining your data

The pattern across CME, KeyBank, and BNY points to three essential practices:

Start with what your data represents. Before CME could model risk scenarios, they needed to understand what 30 billion daily market events actually meant for portfolio behavior under different conditions. Before BNY could mine 240 years of institutional knowledge, they needed to understand what information those 3 million documents contained and how pieces related to each other.

Connect data to business outcomes from day one. KeyBank learned this lesson directly: data initiatives that exist in isolation fail. When their commercial banking team wanted faster loan approvals, they started with the business goal. Business leaders champion data initiatives when they understand exactly how contextual information enables their goals.

Build governance that adds value, not just restrictions. BNY's review board process could slow teams down. Instead, it accelerates them. Governance can become a competitive advantage when it provides context about data provenance, usage rights, and appropriate applications.

What comes next

"It's untenable to ignore AI," Kataria said simply. "It's going to be an integral part of every institution."

The organizations winning this shift aren't just adopting new technology — they're getting better at refining what they already have. They understand their data's context: what it means, where it came from, how it relates to business goals. 

The question isn't whether to adopt AI. The question is whether you understand your data well enough to put AI to work.