Inside a global investment bank's role redesign.

At a global investment bank, the question was never whether a senior risk leader was working hard enough. It was how much of the function's capacity sat locked in work that AI now handles well. A recent RoleOS analysis put a number on it: $423,000 to $616,000 a year, across one leader's role and the nine-person team beneath it. Almost none of it requires AI to make risk judgments. Most of it comes from removing the work that was never the leader's judgment in the first place.

The role, before the analysis

A senior leader in the bank's risk function, with a dual mandate across scenario analysis and stress testing, leading a team of nine analysts through an active framework overhaul. The work the bank relies on this person for is judgment under regulatory pressure: the calls the Chief Risk Officer and the board look to them to make, and the trust that sits behind those calls.

Around that judgment sat a layer of work the role had quietly absorbed. Board and stakeholder communications. First drafts of regulatory memos. CRO prep and ad hoc risk responses. And beneath the leader, a team carrying a heavy load of structured, repeatable analyst work: peer benchmarking, KRI reporting, loss-event research, model documentation.

What the analysis surfaced

RoleOS broke the role and the team into a task-level inventory and ran each task through the three-way split: automate, augment, keep human. What stood out is that the leader's own time was already mostly irreplaceable. Roughly 70% of it goes to strategic and transformation work that cannot, and should not, be automated. The opportunity was not in the judgment. It was in everything stacked around it.

Keep human

The strategic core stays untouched, and the analysis is emphatic that it should. The judgment calls on emerging risk. The advisory relationship with the CRO and the board. Ownership of every assumption and every regulatory submission. This is the durable moat: the reason the role exists, and the part AI makes more valuable, not less.

Augment

The single biggest fixable problem was communications, consuming two to three hours of the leader's day, every day. First-draft emails, briefings, CRO prep. An executive communications assistant, built inside the bank's approved Microsoft Copilot environment and trained on the leader's voice and context, gives back 1.5 to 2 of those hours daily. Regulatory memo drafting comes next: a first-draft assist that cuts initial drafting time by 50 to 65%. The leader still reviews and owns every word that leaves the desk.

Automate

The largest pool of capacity sits one layer down, across the analyst team. The analysis identified 5.59 full-time-equivalents of structured, repeatable work that AI handles reliably today: benchmarking, reporting, monitoring, documentation. Freeing that capacity does not shrink the team. It redirects nine analysts toward the forward-looking analytical work the function cannot currently produce at scale.

The leader's hardest work was never the bottleneck. The work stacked around it was.

Built for a regulated environment

None of this works in a bank if it cannot survive a regulator's question. So the pilot design carried the governance with it. Every pilot output runs human-in-the-loop: AI drafts and proposes, the leader reviews and approves before anything moves. Nothing enters a regulatory submission without a documented trail of AI inputs, model version, and human sign-off. And before any tool is switched on, week one captures a clean baseline, so the before-and-after is measured, not asserted.

It also stays inside the bank's existing stack. The communications and drafting pilots run on the approved Microsoft Copilot environment. No new systems. No IT approval. No procurement cycle.

The projected impact

Modeled across the leader's role and the nine-person team, the combined capacity in scope runs $423,600 in the conservative case to $616,500 in the moderate case, depending on how much of the identified work is actually captured. The figure that goes in front of a CRO or head of operations is the function-level number. The follow-on engagement turns that aggregate into an executable plan for each analyst role.

These numbers are modeled from the task inventory and the function's loaded labor rates. They are projected capacity, not yet measured post-pilot outcomes. What the analysis establishes with confidence is the structural finding: a significant share of this function, as currently constructed, is work that does not need the people doing it to do it by hand.

What the function gets back

The point of the redesign is not a leaner risk function. It is a sharper one. The leader gets back the hours the transformation actually requires. The analysts get pulled off reporting and onto the forward-looking analysis the function has never had the capacity to run. The role does not shrink. It moves up.

That is the shape of role redesign in a regulated environment. Not fewer people. The same people, aimed at the work only they can do.

Common questions about a risk-function role redesign

Does this hold up in a regulated environment?

That is the first design constraint, not an afterthought. Every pilot output is human-in-the-loop. Nothing reaches a regulatory submission without a documented audit trail of AI inputs, model version, and human sign-off. Tools run inside the bank's own approved environment, under its own controls.

How do you handle confidentiality?

The analysis is conducted at the role level, not the matter level. No client data, no privileged content, no names. Where AI tools enter the workflow, the bank controls deployment and data handling. RoleOS does not touch the bank's work product.

Why start with one role instead of the whole function?

One role keeps the analysis precise and the first pilot low-risk. But the function-level number is the business case. Once a CRO sees the aggregate capacity across the team, the multi-role engagement converts it into a sequenced, executable plan for each analyst position.

How do the projected numbers get tested?

During the pilot. Week one captures a baseline before any tool is deployed. The function then tracks hours against the recovered-time targets at the 30, 60, and 90-day milestones. The model gets calibrated against measured outcomes, not left as a projection.

RoleOS analysis is grounded in research-backed task analysis and a proprietary scoring framework developed across real client engagements. Projected outcomes are anchored to the function's loaded labor rates and modeled from the engagement task inventory.