Three simple steps for banks to measure AI's ROI

• Key insight: Measuring AI impact starts on Day 1 of a project.
• What's at stake: Without a clear organizational structure built around the execution of leading initiatives, banks risk confusing AI impact with that from non-AI initiatives.

Lloyds Banking Group has set an ambitious target for its next phase of transformation. Under Accelerate 2030, its "simplify to outperform" strategy includes a digital and AI productivity push and about £2 billion in gross cost saving.

The harder question is how much of those savings can ultimately be attributed to AI.

Banks can make that easier by doing three things before a major AI project starts: Define the business outcome, establish the baseline and make one cross-functional team accountable for the result.

That is where T-shaped teams have an advantage. A T-shaped team brings business and technology functions together around a specific problem. Instead of deploying technology first and looking for benefits later, the team starts with what it is trying to achieve and designs measurement into the workflow. Both revenue and costs are largely self-contained.

T-shaped teams were designed to help financial institutions test, refine and ultimately scale AI-enabled workflows. The same structure also makes them particularly well suited to measuring whether those workflows actually produce business results.

This is the second installment in a three-part series on measuring AI impact in banking. Part 1 examined what banks should measure; this article looks at the organizational structure that makes the measurement process easy and accurate. Part 3 will address how to translate business outcomes into defensible ROI.

Start with the outcome

Many AI projects still begin with the technology. A new model becomes available, employees experiment with it and management later asks whether it created value. That sequence makes measurement unnecessarily difficult.

Consider commercial loan underwriting. The objective might be to speed up application processes without increasing credit losses, reduce turnaround time or improve conversion among qualified borrowers. Those are business outcomes. Hours saved, model accuracy and usage are measures that are tracked along the way. 

A T-shaped team forces that distinction early. The business function knows the workflow and what success should look like. The technology function knows what AI can reliably deliver. The bridging function connects the two and keeps the project focused on the original business problem. 

This is part of the logic behind the T-shaped team model: Cross-functional teams should direct resources toward projects capable of delivering measurable results rather than pursue technology for technology's sake.

Once the desired outcome is explicit, the baseline becomes clearer too. If the goal is to reduce underwriting turnaround time from five days to three without affecting credit quality, the team knows what needs to be measured from the start.

Before deployment, a T-shaped team should agree not only on the outcome but also on what result would justify continuing the project. A metric becomes more useful when management knows what decision it will trigger.

The key message is that measurement starts with project design, not reporting.

Connect AI to the result

The second challenge is attribution. 

A bank rarely introduces AI into a workflow without changing something else. Processes may be redesigned, employees retrained and other technology upgraded at the same time. That makes a simple before and after comparison unreliable.

The T-shaped team helps because the people who understand those changes are working on the same project. The technology team can measure what AI did. Business managers can identify what changed in the workflow. Finance can observe the resulting business performance. The team can then trace an evidence chain:

AI performance -> workflow improvement -> business outcome

If AI makes an analyst 20% faster, for example, that is not yet a business outcome. Did the bank process more applications? Did turnaround time improve? Did the error rate decline? Did customer conversion increase? 

So the measurement question is not about whether AI did something. It is whether AI contributed to the outcome the bank set out to achieve.

Build enterprise impact from the bottom up

NatWest's first-half results illustrate the challenge at a larger scale.

The bank reported around £250 million in gross cost reductions, driven by structural simplifications and continued investment in technology platforms. It also expanded AI-enabled capabilities across onboarding, operations and customer servicing. NatWest did not attribute those cost reductions specifically to AI.

That is understandable. Once AI becomes part of a broader transformation, isolating its contributions becomes harder. 

The best approach is disciplined measurement at the use-case level. If each major project, what's often called a big rock, begins with a defined outcome and the cross-functional team accountable for its results, management has a much stronger basis for determining what AI is contributing across the bank.

That does not mean adding every claimed improvement together. Banks still need to distinguish realized results from forecasts, avoid double counting and use common definitions when results are aggregated. 

That evidence should ultimately support explicit decisions: Continue investing, redesign the workflow or model, strengthen controls, or stop the project. Stage gates help make those decisions before organizational momentum takes over.

This brings us back to Lloyds. Its £2 billion enterprise target sets the ambition, but proving what AI contributes to it requires evidence from the underlying workflows. (Lloyds does highlight a figure called AI generated value, reporting £50 million in 2025 and £100 million in the first half of 2026, but it has not released further details.)

Banks can call the organizational structure something else. But the critical elements that make it work are hard to avoid: business expertise, technology expertise and strategic leaders capable of connecting the two around a measurable outcome. 

Measurement is not the primary purpose of a T-shaped team. Improving the workflow is. But by organizing business and technology expertise around a defined outcome, the team makes the results of that improvement that much easier to measure.

Once that outcome can be measured, another question follows: What is it exactly worth? That is the question we'll tackle in the next installment of this series. Stay tuned.