Umesh Babaria has spent more than 20 years in banking — corporate finance and strategy at M&T Bank, retail product management, line-of-business CFO, and now Head of Consumer Analytics and Physical Distribution/Market Strategy at KeyBank. That winding path shaped the contrarian ideas he brought to the BankSpaces stage: implementing AI isn't the hard part of banking. Changing decisions is.
Technology, he argued, is more capable and available than ever. But banks' ability to change decisions in line with what that technology offers hasn't caught up.
"AI for the sake of AI is not going to be super helpful," he said.
Babaria shared a chart comparing the adoption curve of the internet with that of ChatGPT — the latter dramatically steeper. He likened it to his own kids: his 9-year-old son is the internet; his 4-year-old daughter, learning from everything around her, is ChatGPT.
The takeaway for bankers is twofold. Your customers are on the ChatGPT curve — so what curve are you on? And because that curve is steeper, the window to adjust your operating model is far shorter than it was in the internet era. "Being a fast follower is not going to be good enough."
His CFO instincts sharpen the point: producing insights creates zero value on its own. Value arrives only when someone acts — a frontline banker, a back-office teammate, a client. "Insight without action is just expensive awareness."
ATMs were supposed to kill branches 30 years ago; digital was supposed to finish the job 20 years ago. Now branches are sexy again. The real question is how AI brings physical and digital together into one go-to-market engine: right client, right time, right channel, right message — a strategy Babaria believes is nearer to execution than ever before.
That means resisting the frontline instinct to "give me leads" for every client (some prefer never to hear from a banker) while building genuine advice skills in the branch, because even digital-native Gen Z wants to look someone in the eye for complex needs.
KeyBank's journey started two years ago with a central team building governance and guardrails. In 2025, Babaria pushed for a shift: an AI team embedded inside the consumer business itself. His mantra — "business-led, tech-enabled. Not tech-led and tech-enabled," he said.
The results so far: 30+ ideation sessions producing roughly 250 AI ideas, about 30% addressable through Copilot alone — paired with prescriptive, job-function-specific guidance rather than open-ended access. Early wins are tangible: small business underwriting proposals cut from two hours to five minutes via engineered prompts (with human review preserved), and a merchant statement tool — traditional AI, not gen AI — that generates a pricing proposal in 90 seconds, right in front of the client. AI champions in each line of business carry the work forward.
In the branches, the focus is freeing bankers for meaningful conversations: automating data-gathering and note-taking, pre-call planning, and push-button performance insights so branch managers spend time coaching instead of decoding reports.
Babaria closed where he began: don't be fascinated by the technology. Be fascinated by the problem — and honest about whether your organization has the structure and willingness to act. Because AI is not the hardest part. Changing decisions is.