Nov 20 2026
Artificial Intelligence

What Financial Services Should Consider When Measuring ROI for AI

A Google Cloud leader shares insights on how organizations can better capture the success of their artificial intelligence deployments.

After several years of experimentation, many financial services organizations have moved beyond the piloting phase and have documented some measurable returns from using artificial intelligence in key areas such as fraud detection and back-office automation.

At State Farm, for example, agentic AI through Microsoft Copilot has helped employees retrieve vital information for faster underwriting, and reduced the translation cycle for marketing materials from English to Spanish from 12 months to 24 hours. KeyBank’s earlier adoption of Google Cloud set the foundation to launch Gemini to supports its engineers.

Last month, Google Cloud released a report about the ongoing challenges that persist with the adoption of enterprise AI, including the barriers to scaling solutions and measuring long-term ROI. Financial services organizations are already in what the report called “advanced stages of AI adoption” — 93% have strong, organizationwide AI deployment — but scaling further requires them to modernize their underlying IT foundation.

“The step that financial institutions most frequently underestimate is establishing clear, upfront baselines alongside end-to-end workflow and data readiness before scaling,” says Georgina Bulkeley, managing director and global industries lead for financial services at Google Cloud.

Click the banner below to learn more about crafting an effective enterprise AI strategy. 

 

How Has Measuring ROI for AI in the Financial Sector Changed?

Just two years ago, Bulkeley notes, AI ROI had been anchored in “isolated proofs-of-concept and single-task productivity, essentially treating generative AI like an upgraded search bar or a basic FAQ chatbot. ROI was largely theoretical, measured through pilot-user satisfaction scores or narrow task-level time savings.”

Measuring AI value purely through individual seat-based productivity is becoming obsolete. Even if a single employee saves time, how does that translate to broader organizational workflows if there are still bottlenecks because of manual handoffs and legacy approvals?

“Instead of asking, ‘How many minutes did this save an employee?,’ leaders should ask, ‘What does our operating leverage look like at two times transaction volume without adding headcount? How much faster are we onboarding clients, settling claims or launching products?’” Bulkeley says. “Value lives in re-engineered end-to-end processes, not individual productivity.”

READ MORE: What happens when AI agents join your security team?

Now, with the shift in interest from generative to agentic AI, organizations want to deploy specialized AI agents that can handle high-volume workflows, such as claims processing and software modernization, which means tracking real profit-and-loss metrics such as revenue lift and onboarding cycle compression.

“Looking toward 2027, firms need to move from measuring task-level efficiency to end-to-end workflow and ecosystem outcomes,” Bulkeley says. “As multiagent systems and agentic commerce mature, institutions should measure decision velocity, cross-workflow straight-through processing and capital efficiency. For example, measuring how quickly the organization can launch new financial products, adapt to regulatory changes and orchestrate autonomous workflows will be new measures of ROI.”

The days of tracking immediate operational and efficiency metrics — such as cost per interaction, hours saved in documentation and IT developer velocity — make sense for justifying early budget cycles to a CFO, she adds. But these only capture the surface layer of value.

Georgina Bulkeley
Looking toward 2027, firms need to move from measuring task-level efficiency to end-to-end workflow and ecosystem outcomes.”

Georgina Bulkeley Managing Director and Global Industries Lead for Financial Services, Google Cloud

“We see the real sustainable impact in top-line performance and in risk resilience. For example, we’re seeing AI driving revenue gains through hyperpersonalized wealth and retail banking experiences, faster credit underwriting and improved client retention,” she says.

Firms that only measure headcount reduction or faster task completion miss the compounding value of prevented compliance fraud losses, for instance.

DISCOVER: How U.S. Figure Skating modernized its IT foundation.

“When you deploy AI in areas such as anti-money laundering, KYC (know-your-customer) onboarding or regulatory reporting, traditional spreadsheets fail to capture the asymmetric value of avoided downside,” she adds, including preventing regulatory fines or avoiding remediation freezes.

That’s why forward-looking C-suite leaders are shifting their models to quantify their risks. “When an AI system cuts KYC review cycles from days to minutes while producing a complete, verifiable audit trail, that is not only compliance but also direct revenue acceleration and operational capacity,” Bulkeley says.

As organizations shift how they measure AI ROI, that will likely influence how they operate to stay competitive, leaving legacy processes behind for agile, multiagent workflows.

Modern Infrastructure and a Ready Workforce Are Key to AI Success

The organizations that are leading the pack know that infrastructure modernization is vital to agentic AI. Yet 59% of financial services organizations say their IT teams spend more time than is reasonable maintaining, patching and integrating legacy systems, according to Google Cloud’s research. “You cannot get compounding AI returns if your agents are bolted onto fragmented, brittle infrastructure,” Bulkeley says.

55%

The percentage of financial services organizations that will need significant investments to modernize their infrastructure before they can deploy AI at scale

Source: Google Cloud, “The Innovation Paradox: How legacy technology, maintenance burden, and provider dependence are shaping enterprise AI readiness,” October 2026

Infrastructure modernization also involves embedding governance from the start, with cross-departmental collaboration from risk, legal and compliance teams. Strong governance also helps prevent AI agent sprawl and shadow AI, which reduces a firm’s chances for data leakage and duplicated spending. After all, responsible AI and strengthening governance in financial services are no longer thought of as “brakes on innovation” but as “primary accelerators of scale,” she notes.

“Think about the economics: An AI model that cannot be explained or audited cannot be deployed into core banking, wealth advisory or underwriting. Its ROI remains zero because it never leaves the sandbox,” Bulkeley says. “Conversely, when an institution invests in strong guardrails such as data explainability, lineage tracking, deterministic tool-call logging, bias testing and human-in-the-loop escalation paths, it unlocks the ability to deploy agents into high-value, revenue-generating front-office workflows with confidence.”

Finally, Bulkeley notes, workforce readiness and transforming what financial services talent looks like moving forward.

“Across the institutions seeing the strongest returns, we are seeing a deliberate investment in turning domain experts, underwriters, relationship managers, risk officers and operations specialists into active orchestrators and builders of AI workflows rather than passive recipients of IT tools,” she says. “When employees are trained not just on how to prompt a tool, but on how to supervise agents, spot edge cases and redesign their own team workflows, adoption stops being a top-down mandate and becomes an organic engine of continuous improvement.”

zamrznutitonovi/Getty Images
Close

New Research from CDW Explores AI and Cybersecurity

Learn how AI is helping IT teams manage risk and improve resilience.