Is there a clean framework for how banks should think about proving AI’s return on investment? One that works is first measure the business outcome, then attribute it to a clear baseline, and roll accountability up across functions. It’s a good model. But it skips over the question that determines whether any of it actually happens: who, exactly, is going to do this work? 

Most banks didn’t build their org charts for this. The people who understand the business outcome (line-of-business leaders), the people who can build a defensible baseline (data and analytics), and the people who can hold the initiative accountable across silos (a genuine cross-functional owner) rarely report to the same person, and in a lot of institutions, one or more of those seats is simply empty. It’s a staffing gap wearing a measurement costume. 

Step one exposes a talent gap, not a data gap 

Defining the business outcome sounds simple until you try to do it. Is the AI tool for underwriting speed supposed to reduce cycle time, reduce credit losses, or free upunderwriters for more complex files? Each answer points to a different owner and a different skill set.  

Banks that get this step right usually have someone in the room who has done both the operational job and the analytics job – a former branch or lending leader who’s fluent in metrics, or an analytics hire who has spent real time on the business side. That combination is rare, and it’s exactly the kind of hybrid profile that a generic job posting for ‘AI Program Manager’ will not surface. 

Step two is where the attribution talent shortage shows up 

Establishing a credible baseline requires someone who can separate AI’s effect from everything else moving at the same time: rate changes, staffing shifts, seasonal volume. This is a specific skill, closer to a data scientist with domain fluency than a traditional BI analyst, and demand for it has outpaced the supply inside most banking organizations. Institutions that try to fill this with an existing generalist analyst often end up with numbers nobody trusts, which quietly kills the initiative months before anyone admits it’s dead. 

Step three fails without the right kind of leader 

Rolling accountability up across functions is a governance exercise, and governance exercises die without a specific kind of leader: someone senior enough to convene competing stakeholders, credible enough with the board to report hard numbers, and technical enough not to get talked around by either the vendor or the skeptics. That’s an executive hire, not a committee. Banks that treat this as a part-time addition to an existing COO’s or CIO’s plate tend to see AI initiatives stall at the pilot stage. 

What this means for how banks hire 

• Write the AI governance or program-owner role as a hybrid business-and-analytics profile, not a pure technologist role. 

• Recruit the attribution/analytics seat as a specialized skill set, with compensation benchmarked accordingly, treating it like a generic data analyst opening is the fastest way to lose the search to a fintech competitor. 

• Give the cross-functional accountability role real authority and a direct line to executive leadership before the search starts, not after a candidate asks about it. 

The three-step framework for measuring AI’s ROI is sound. But no measurement framework survives an organization that hasn’t hired for it. Before your bank rolls out its next AI initiative, it’s worth asking whether the people who need to own each step are already on staff, or whether that gap is the real reason the last initiative never proved its value.