Financial Services Firms Race to Operationalize AI

Turning AI Pilots Into Business Value

Tiran Rothman, vice president of business and financial services for Frost & Sullivan, argues that firms that create real value do not treat AI as a set of isolated experiments; they connect use cases to business processes, core systems, risk controls and measurable outcomes. “Pilots generate interest, but production-ready AI creates impact,” he says. “The difference is execution discipline.”

Aser Blanco, global head of banking for NVIDIA, says firms that stall tend to treat AI as a stand-alone experiment rather than a capability they need to wire into existing workflows and decision-making processes.

“They experiment with hundreds of side projects instead of focusing from day one to transform core processes, and in many cases, they see AI as a tech challenge that IT will solve for them,” he says.

Click the banner below to subscribe to our newsletter for the latest financial services IT insights.

Rothman says an execution gap often emerges between AI strategy and operating reality, where firms may have ambitious AI roadmaps but lack the ownership structures, governance models, funding, data readiness and operational processes needed to move projects from experimentation into production: “In practice, AI stalls when it remains a technology initiative rather than a business transformation.”

Blanco points out that while banks have the strategy decks, in most cases, what they do not yet have is the AI factory, the connected data and the governance fabric required to run intelligence at the scale of an institution that processes billions of transactions a day.

“A model that wins in a sandbox does not survive contact with real-time latency requirements, model risk regimes, fifty years of acquired core systems and the change management work of getting a 200,000-person workforce to actually use it,” he says.

DISCOVER: Turn data into insights and accelerate artificial intelligence initiatives.

Data Problems Slow AI Adoption in Finance

Nearly half of financial institutions identified securing data, insufficient data, integration challenges, data quality and trust in outputs as significant barriers to preparing data for AI projects.

Financial institutions typically have enormous volumes of data, but it’s fragmented across systems that don’t usually work together, Blanco says: “The most common issues are inconsistent labeling, incomplete data lineage and governance frameworks built for compliance reporting rather than AI model training.”

Indranil Bandyopadhyay, a principal analyst at Forrester, says firms often overestimate how prepared their data environments are for AI, particularly as generative AI introduces new requirements related to context, semantics and unstructured data.

“That data was prepared for human consumption, not for AI systems that rely on technologies such as vector databases and multimodal data platforms,” he says.

Bandyopadhyay also notes that AI systems require continuous oversight because they are probabilistic in nature, creating risks for model drift and data drift, meaning organizations cannot simply build once and then move to the next thing.

With 87% of respondents saying AI can improve decision-making and reduce employee workload, financial institutions should prioritize use cases tied to operational efficiency, workflow automation and faster execution, where value is easier to measure and scale. Rothman advises firms to treat AI use cases like an investment portfolio: Each initiative should be assessed by business value, feasibility, risk exposure and scalability.

“The best roadmap is not the longest list of ideas; it is the one that focuses resources on use cases that can deliver measurable value and build reusable capabilities,” he says.

Starting with two or three well-resourced, high-feedback use cases and building shared AI infrastructure around them — for example, a model-ready data platform, scalable compute or a reusable model registry — creates far more durable momentum than pursuing hundreds of small initiatives on separate stacks. “The institutions building AI factories as a shared platform gain a compounding advantage,” Blanco explains. “Every new use case deploys faster because the foundation is already in place.”

UP NEXT: How banks are using AI for help with regulatory compliance.

How To Measure AI ROI Beyond Quick Wins

Most financial institutions are still in the early stages of realizing financial returns from AI, with fewer than half reporting ROI above 50%, underscoring the long-term nature of measuring value from AI investments.

Firms seeking to cut costs by streamlining processes with AI should look at more than just immediate cost reduction when measuring ROI, Rothman advises, including decision quality, speed, risk reduction, customer experience, employee productivity and the creation of reusable data and technology capabilities.

“The real value of AI is not only transactional but also cumulative, because it changes the economics of how knowledge work is delivered,” he says.

From Blanco’s perspective, the most underappreciated form of AI ROI is organizational velocity. “How much faster teams can act on new information, adapt to market changes or respond to emerging risks is critical,” he says. “That’s harder to quantify than cost savings or revenue increases, but it’s often more strategically significant.”

Nearly three-quarters of financial services organizations believe their infrastructure is already prepared to support AI workloads, reflecting growing confidence in the sector’s technical readiness for large-scale AI adoption.

However, Blanco points out that most institutions still have a real infrastructure gap, noting that the compute footprint required for training and serving large-scale models — particularly real-time, GPU-bound inference for fraud and trading — is fundamentally different from the batch-analytics stacks that banks built over the past 20 years.

“Trying to run modern AI on legacy infrastructure hits a hard ceiling fast,” he cautions. “Banks that treat AI infrastructure as enterprise-grade, not project-grade, will lap the rest.”

Jeff Basciano, senior vice president of global financial services industry at Citrix, says some financial institutions struggle due to increasing infrastructure costs, lack of security controls, hardware availability or unpredictable cloud compute costs.

“To be fully equipped to support AI workloads at scale, financial institutions need to put solutions in place that allow them to tailor security, governance and infrastructure needs to the needs of their AI implementations, while providing the flexibility to adapt and scale resources and controls up or down as the business evolves,” Basciano says.

Click the banner below to learn how organizations are unlocking artificial intelligence’s potential.

Best Practices for Scaling AI Agents

Interest in more advanced AI capabilities is accelerating, with 82% of financial institutions actively exploring applications for AI agents as firms look beyond automation toward more autonomous systems and workflows.

Rothman says moving from AI assistants to AI agents is not just a technology upgrade but a governance upgrade.

“Firms need trusted data, clear permissioning, audit trails, human oversight, model risk management and strict boundaries between recommendation and autonomous execution,” he says.

Basciano says to move from pilots to AI agents, firms need to treat autonomy like any other critical service: define what must keep working, map the dependencies and bake in controls that tighten under stress. Just as important, firms need audit-ready traceability — clear accountability for what the system did, why it did it, and how it can be safely paused, rolled back or overridden.

“Having this foundation in place is critical for institutions to be able to trust AI-driven processes the same way they trust their most critical human workflows,” Basciano says.

Blanco notes that the organizations best positioned to move into this space are those that have already invested in explainability and model monitoring in their existing AI programs: “Firms that shortcut governance in early use cases will find it significantly harder to earn the internal and regulatory trust needed to deploy more autonomous systems.”