Clock is ticking: Why banks can't ignore unstructured data

There's a paradox sitting at the heart of banking.

US financial institutions collectively spend nearly 10% of their revenue on IT , yet productivity struggles, cost-to-income ratios remain stubbornly high, and the promised AI transformation keeps getting deferred. The culprit isn't strategy or talent. More banking leaders are realizing the answer is a strong and trusted data foundation underpinning their strategy and AI efforts.

Specifically, it's the unstructured data problem and it's bigger than many banks acknowledge.

The hidden 80%

According to GartnerTM Data Intelligence Monthly: Executive Insights on Unstructured Data for AI, unstructured data now represents 70%-90% of the enterprise data estimate, growing at 40%-60% per year. In banking and financial services, that means an enormous share of a firm's most valuable institutional knowledge: credit memos, loan documents, call center transcripts, compliance filings, client emails, underwriting submissions. A majority of these exist in a form that legacy systems simply cannot process, analyze or act on today.

As McKinsey has noted, predictive AI models struggle to adapt when data is unstructured and the nature of tasks requires multistep planning, reasoning, and orchestration — the exact conditions that define most high-value banking workflows. The result is a costly irony: banks have more information about their customers, markets, and risks than ever before, yet they can't reliably use it.

The AI-readiness gap

This urgency becomes acute for major AI initiatives — from fraud detection to credit decisioning or customer personalization. A majority of AI projects have been delayed by 3 to 6 months. AI projects depend on clean, accessible, well-governed data. Right now, most banking data simply isn't AI-ready. Banking data is often fragmented across siloed and legacy systems, duplicated, ungoverned, and locked inside formats that models can't access or interpret.

McKinsey is direct about the consequence: organizations must "address significant quality gaps that exist in unstructured data — the very text, documents, and images that agents rely on to understand context and execute tasks" before AI investments can deliver real returns. Skip this step, and even the most sophisticated AI deployment will underperform or miss the mark entirely.

The security exposure compounds the problem. A Ponemon Institute study found that 84% of financial organizations say their unstructured data is accessible by people with no business need for it — a compliance and reputational risk that grows with every new document created.

The agentic AI opportunity for those ready to capture it

What makes the stakes higher today is the arrival of agentic AI. Unlike conventional automation, agentic AI systems can interpret goals, break them into tasks, interact with multiple systems, and execute complex workflows with minimal human intervention. And in banking, the biggest opportunities lie squarely in back-office workflows that require parsing unstructured data — exactly where human time is currently being consumed.

The early results among banks that have addressed their data foundations are striking. McKinsey's analysis of multiagent systems applied to credit memo creation found analyst productivity gains of 20–60% and roughly 30% faster decision-making — achieved by replacing a manual, multi-source data gathering process with agents that extract, synthesize, and draft autonomously. In a separate documented case, a North American bank using a multiagent system for credit risk memos saw credit decisions become 30% faster, while relationship manager productivity more than doubled. PwC's analysis of agentic workflows in finance operations found that AI-driven invoice extraction and processing can slash cycle times by up to 80%, while improving audit trails and reducing compliance risk.

These aren't pilot-project statistics. They're the competitive advantage being built by banks that acted first.

The cost of waiting

McKinsey's Global Banking Annual Review quantifies the divide clearly: AI-first banks are on track for a 4% return on tangible equity advantage over slow movers, who face an increasingly uncompetitive cost base. And McKinsey estimates that between 50 and 60% of bank FTEs are tied in some way to operations — making it the single largest addressable cost pool for AI investment.

The primrose path runs directly through unstructured data. Banks which charge diligently ahead through the less glamorous preliminary and preparative stages of AI will have the advantage later. The ultimate industry leaders will be the ones investing now in assessing, classifying, governing, and activating their unstructured data assets. Fastidious data hygiene isn’t just useful preparation for agentic AI — it’s building the foundation for every intelligent workflow, every agentic automation, and every customer experience improvement that follows. Those that defer are accumulating technical debt with compounding interest.

The data exists. The technology to ingest and properly translate unstructured data for AI exists, too. The gap between them is solvable. But only for institutions willing to address it before their competitors do.

• 

  Cynthia Garcia

  Cynthia Garcia

  Area Vice President, Sales

  Collibra