Our Take
Banking executives correctly treat agentic AI as sophisticated automation rather than autonomous decision-making, focusing on measurable ROI from grunt work instead of chasing the hype.
Why it matters
Financial institutions face regulatory pressure and zero fault tolerance, making their conservative approach a preview of how other risk-averse industries will adopt AI agents. Their focus on back-office automation reveals where the actual near-term value lies.
Do this week
Risk managers: audit your highest-volume, lowest-stakes processes this week so you can identify automation candidates that won't trigger compliance issues.
Banks deploy agents for grunt work only
Major financial institutions are limiting agentic AI to high-volume back-office tasks while maintaining human oversight for all decisions. Speaking at SAS's annual conference, JPMorganChase's Adolfo Lopez called agentic AI "in its terrible twos stage," with most implementations remaining "assisted or delegative" rather than autonomous.
Specific use cases include KYC onboarding, compliance documentation, and claims processing. Allianz deployed agents to handle high-volume, low-risk claims while keeping humans as decision-makers for catastrophe claims. The insurer aims to cut commercial underwriting timelines from six to eight weeks down to six to eight hours (company-reported).
AWS's Sri Raghavan emphasized that "human in the lead" defines current deployments, warning that excessive autonomy in high-risk situations can go "spectacularly wrong." He identified KYC compliance as immediate low-hanging fruit for automation efforts.
Regulated industries set the template
Banks operate under extreme regulatory scrutiny with zero fault tolerance, making their approach a bellwether for other risk-averse sectors. As SAS banking advisor Stephen Greer noted, financial services deals with "very complex problems that have extremely high consequences for failure," shaping implementation decisions.
The focus on back-office automation over customer-facing applications reveals where organizations find measurable ROI. Rather than pursuing personalized investment recommendations or other high-stakes use cases, banks prioritize tasks where human reviewers can catch mistakes before they become costly.
Talent shortages add urgency to automation efforts, particularly in underwriting where processing bottlenecks affect business operations.
Start small, fix data first
Lopez recommended organizations "find that one process that eats up a significant amount of high-impact human hours, and automate that process." The iterative approach prioritizes proven value over ambitious scope.
Data rationalization emerges as a prerequisite. Raghavan cited a client managing 146 data sources across spreadsheets, flat files, and platforms, arguing that cleaning up these assets ensures effective AI rollouts.
On build versus buy decisions, JPMorganChase uses external models but develops control layers internally. "The guardrails that actually feed the data are all being built in-house, but the model is coming from outside," Lopez explained. This hybrid approach lets banks maintain control over risk management while accessing external AI capabilities.