FinAi News interviews Emily Steele on the difference between AI that advises and AI that executes
FinAi News reporter Quinn Donoghue spoke with Savana President and COO Emily Steele about where AI actually earns its place in digital banking. The article, published August 7, sets Savana’s thinking against the prevailing direction of the market.
The distinction Steele drew for Donoghue is between advisory AI and operational AI. Advisory AI produces insights and recommendations. It surfaces what a banker might want to know. Operational AI executes the work itself. Savana has spent nearly 18 months in research and development on that second category, and Steele told the publication the choice was deliberate. The R&D has been aimed at what will give financial institutions a real lift rather than at what is currently generating attention.
Why Recommendations Don’t Solve the Underlying Problem
Steele’s argument in the piece is structural. Financial institutions have accumulated technology over decades, much of it through acquisition. Those stacks were never built to talk to each other. Someone still has to bridge the gaps by hand, and Steele told FinAi News who that someone is.
“A banker has become the integration layer at a financial institution.”
That is the condition Savana is targeting. Fix it, Steele argues, and better customer experience and genuine transformation become possible. Layer an advisory tool on top of it instead and the institution has added one more system for the banker to reconcile. As she framed it to Donoghue, that propagates a problem that is already serious.
The Decision Layer
The alternative Steele described begins before any model runs. Every step where an institution interacts with a customer is mapped in advance. Routing is defined. Guardrails and compliance requirements are built in. Once that structure exists, AI can execute inside it rather than making suggestions for someone else to act on. Steele told Donoghue that this is where real efficiency shows up.
The article frames this as Savana prioritizing the decision-making intelligence layer, and it is a meaningfully different bet than embedding conversational tools or summarization across the customer experience.
An Institution’s Own Model, Not a Generic One
Steele also emphasized governance. In the approach she outlined, operational AI ingests an institution’s own policies and procedures rather than running on a general-purpose model. What it can execute is then controlled by the entitlements of the individual banker, which keeps the institution’s existing permission structure intact.
Three use cases are first in line: relationship management, fraud detection, and customer service. Savana is demonstrating these capabilities to banking partners so each institution can deploy them in line with its own guidelines, policies, and workflows.
FinAi News subscribers can read the full article here.
Share This: