- Key insight: Winners in the next phase of AI adoption in banking will be decided less by access to models than by how they redesign their most important workflows.
- What's at stake: Spreading AI thinly across the organization may marginally improve productivity without creating business value that banks can measure, scale and govern.
- Supporting data: McKinsey recommends concentrating AI investments in one to three high-value domains with economic leverage, proprietary data and workflow complexity.
Banks will not capture the full value of artificial intelligence by using it only as a productivity tool. They will need to redesign workflows around what makes their business competitive: proprietary data, domain expertise and the way people and technology work together.
The issue came up during a panel at
My answer was proprietary workflow, enabled by proprietary data.
That does not mean models are unimportant. Banks need access to models appropriate for the work, supported by secure data and sound controls. But models are increasingly available to many institutions. Data becomes strategically valuable only when a bank can use it to improve how decisions are made, work is coordinated, and services are delivered.
The harder-to-replicate advantage lies in the workflow.
As I argued in Orlando, real differentiation will come from how banks redesign workflows around AI, not from the models themselves.
McKinsey flags three characteristics of a promising domain:
- Sufficient economic leverage to improve margins or growth.
- Proprietary data that appreciates with use.
- Workflow complexity that exposes the limits of current operating models.
For banks, this means abandoning the urge to sprinkle AI into every department. Leaders must target workflows where the bank holds valuable data, the economics justify the spend, and current processes are bottlenecked. Once those domains are selected, banks need an organizational structure capable of redesigning them. That is where
The T-shaped team is the organizational structure that allows banks to better measure AI's business impact, scale successful applications and embed governance in the workflow from the beginning.
1. Make the T-shaped team the unit of workflow redesign
A T-shaped team is not another cross-functional committee or a collection of representatives from different departments. It integrates three distinct functions:
- Business: Contributes domain expertise and defines the problem.
- Technology: Develops the models, data infrastructure, and applications.
- Bridging: Connects the two to turn technical possibilities into strategic business initiatives.
This bridging function is critical. It translates between business and tech to prevent promising projects from dying in miscommunication. It prioritizes initiatives based on strategic fit and data advantage. Finally, it secures C-suite buy-in to move from pilot to production.
T-shaped teams are designed to integrate AI and data into core financial processes. They bring specialized professionals together around a defined business problem rather than asking one department to build a tool and hand it to another.
While traditional cross-functional efforts often resemble coordination exercises, T-shaped teams are execution units. They are accountable for outcomes, not just contributions.
The team is organized around a specific workflow within one of the priority domains identified by leadership. It owns the process end-to-end, from initial design through deployment, adoption, measurement, and continuous improvement. While risk, compliance, and legal can participate as needed, the central principle is clear: the team owns the workflow entirely.
That structure makes AI adoption easier to measure because the unit owns the workflows, establishes the baseline and defines the desired outcomes. It makes successful initiatives easier to scale because business, technology and the bridging function jointly manage the path from experimentation to production and successful teams can share their playbook across the organization. It also strengthens governance by incorporating risk, compliance, model validation and human accountability into the workflow when needs arise rather than as an afterthought.
2. Redesign the division of labor
Once assembled, the team's first task is not selecting a model. It is breaking the workflow into component tasks to redefine the division of labor.
While keeping a "human in the loop" is a standard governance baseline, it lacks the specificity needed for workflow design. Banks must categorize activities into three buckets, a framework we explored in Orlando:
- AI-led: Highly repeatable work requiring limited judgment.
- AI + HI (Human Intelligence): Technology supports analysis, but human judgment remains material.
- HI-led: High-stakes decisions requiring direct human accountability.
Take commercial credit underwriting. AI can collect borrower data, summarize financials, flag policy exceptions, and benchmark against similar credits. Credit officers are then freed to evaluate management quality, unusual risks, and the assumptions behind projections.
A T-shaped team should completely rethink how information moves between relationship managers and credit officers. Routine reporting shifts to AI-led work, credit analysis becomes a hybrid AI and HI effort, and consequential decisions, like denying credit, remain strictly HI-led.
3. Distinguish workflow redesign from bolt-on productivity
Providing employees with copilots and chatbots generates useful productivity gains. But these gains do not equal strategic differentiation.
Bolting a chatbot onto an unchanged customer service process or using generative AI to speed up document drafting does not change how a bank makes decisions or coordinates work. This is the "peanut butter" problem described by McKinsey. Spreading AI thinly across the organization yields incremental improvements but often fails to build a competitive moat.
Banks must concentrate investments where proprietary data and workflow complexity justify fundamental redesigns. Recognizing this difference prevents teams from forcing generative AI into processes where simpler, rules-based automation would suffice. The T-shaped team must start with business problems where there is a right technology fix, rather than start with a model and search for a nail.
The value gap in financial AI is rarely a technology problem. It is an operating model problem. The question for bank executives is not how many employees have access to AI. It is whether the bank has identified the few workflows where proprietary data creates a meaningful advantage, and whether it has built the organizational structure needed to measure the results, scale what works and govern the associated risks.
The model may be available to everyone. The real advantage lies in workflows that competitors cannot easily replicate.
