Insight | Financial Services

AI Value Attribution

You Cannot Close a Gap You Cannot Measure

Almost every financial institution now uses AI somewhere: in client service, credit, reporting or software development. Far fewer can say what it costs, which work it does, and where the time it saves actually goes. Without that, the AI value gap stays invisible on the profit and loss statement.

AI value attribution is the ability to link AI spending to the work it performs and the outcomes it changes. It comes before any credible return on investment, and it is a management question before it is a technology one.

Usage Is Not Value

Adoption figures look impressive. In McKinsey's 2026 State of AI survey, 80 percent of respondents report individual productivity gains from AI, but only 6 percent attribute a significant share of EBIT to it. As McKinsey puts it: "Individual gains diminish before they reach the P&L."

A cross-industry survey of 300 senior portfolio and PMO leaders, published in 2026 by the planning software company Tempo, points the same way. Nine in ten were piloting or actively using AI in project delivery, but only a third were handing real delivery work to AI agents. Teams that were simply "using AI" had no advantage over peers that were only piloting or had not started, and did worse on some measures.

At a Glance

80% vs 6% see individual productivity gains from AI, against those who attribute a significant share of EBIT to it McKinsey, Global Banking Annual Review 2026 42% cannot tie AI spend to ROI; 39% cannot see what AI costs Tempo 2026 survey, 300 portfolio and PMO leaders ~10 months average payback across 52 agentic AI use cases in banking, with a three-year ROI of about 2.9x McKinsey, Global Banking Annual Review 2026


Four Blind Spots

The same survey shows where the value disappears:

  • Cost: 39 percent of respondents cannot see what AI costs.
  • Return: 42 percent cannot tie AI spend to ROI.
  • Duplication: 46 percent say their teams duplicate work with AI.
  • Attribution: among teams actively using AI, 55 percent struggle to tell AI work from human work.

The survey excludes many highly regulated industries, so these figures describe the wider economy. Banks face the same four questions, with tighter controls around each of them. And McKinsey adds the decision that follows: "Agents produce a productivity dividend, and the bank has to decide where it goes: more output, different work, lower cost, or a mix."



Why Attribution Comes First

Without attribution, every AI discussion becomes a debate about licenses and pilots. With it, the discussion turns to outcomes. McKinsey notes that "employee adoption is the scorecard many banks are using, but a better scorecard is the business outcomes achieved in a few chosen domains."

Where banks measure at that level, the results are visible. Across 52 agentic AI use cases, McKinsey reports an average investment payback of approximately ten months and a three-year ROI of about 2.9 times. Those numbers exist because someone defined the outcome, measured the cost and tracked the result.

The obstacle is rarely the technology. In McKinsey's 2026 survey of 85 banking executives, data readiness was the most cited obstacle to AI (48 percent), followed by talent (41) and change management (37); technology came last (27). The Tempo survey adds an operating detail: among teams with AI agents in production, the share struggling to tell AI work from human work falls to 16 percent. Attribution gets easier when AI does defined work with a defined owner.



Making AI Value Visible

Attribution does not require a large program. Four steps make the difference:

  • Define the outcome before the tool. Choose a few domains and state which decisions, cycle times or costs should change.
  • Track cost and work together. Link AI spending to the work it performs, so cost, output and owner sit in one view.
  • Decide where the dividend goes. Saved time turns into more output, different work or lower cost only if someone decides which.
  • Give each AI-supported process one owner. One person answers for the outcome, whether the work is done by people, agents or both.

The Execution Resilience Diagnostic™ shows, for one leader or a whole leadership team, where ownership of data, technology and AI decisions is unclear. Where it shows a structural issue, a Decision Rights Reset of four to six weeks assigns those decisions to named owners.



Conclusion: Measure the Work, Then the Return

AI value is not created by usage, and it is rarely lost in the technology. It is created or lost in the decisions about what AI should do, who owns the result, and where the saved time goes.

Institutions that link AI cost and work to outcomes can defend their AI budget, and redirect it when it does not pay. For the wider picture, see the AI value gap in financial services and the whitepaper.



Sources:

  • McKinsey & Company, Akshay Kapoor, Brian Ledbetter, Jonathan Leydon, Klaus Dallerup, Marti Riba, Saptarshi Ganguly, Sid Tiwari, Teresa Diviu and Vik Sohoni, Global Banking Annual Review 2026: The AI disruption dividend, November 2026 (80 percent individual productivity gains against 6 percent significant EBIT impact; "Individual gains diminish before they reach the P&L"; the productivity dividend decision; adoption versus business-outcome scorecards; ten-month payback and 2.9 times three-year ROI across 52 agentic use cases; obstacles cited by 85 banking executives). Source
  • Tempo Software, 2026 State of AI in Portfolio Management, survey of 300 senior project, portfolio and PMO leaders at enterprises in North America and Western Europe, fieldwork by Potloc, June 2026; vendor report, excludes many highly regulated industries (91 percent piloting or using AI, 33 percent delegating delivery work to agents; "using AI" group without advantage; 39 percent cannot see AI costs, 42 percent cannot tie AI spend to ROI, 46 percent duplicate AI work; attribution difficulty of 55 percent falling to 16 percent with agents in production). Source