Companies are racing to adopt artificial intelligence, but many are discovering too late that they have no idea what it actually costs to run. A new report from KPMG reveals a troubling disconnect: while 22% of organizations are now embedding AI into daily workflows, nearly half of all businesses have scaled back or paused AI initiatives after concluding the cost outweighed the value. The problem isn't that AI doesn't work. It's that most companies lack the financial visibility to prove whether it does.

Why Are Companies Struggling to Measure AI's Real Value?

The KPMG Global AI Pulse Q2 2026 report found a striking pattern: organizations with full visibility of their AI operating costs reported established returns on investment at five times the rate of those without that insight. This suggests the issue isn't AI itself, but rather the absence of basic cost tracking and measurement discipline. Finance leaders, payments specialists, and cybersecurity experts all point to the same emerging fault line: rapid deployment without clear accountability.

"More firms are implementing AI into everyday workflows, but that does not mean they are creating value. Proving its worth is difficult without a clear way to measure it. Organisations need to know what AI costs to run, what it is actually producing, and who checked the output before it is used in a decision," said Oliver St Clair-Stannard, VP of Payments AI Strategy and Go-To-Market at RedCompass Labs.

Oliver St Clair-Stannard, VP Payments AI Strategy and Go-To-Market, RedCompass Labs

In regulated industries like banking and payments, this gap becomes critical. Banks must be able to show regulators exactly how AI models behave, what they cost, and how humans oversee their output. Speed with AI is worthless if you cannot clearly articulate what it generated, who approved it, and whether it would survive an audit.

How to Build Financial Discipline Into Your AI Strategy

• Establish Cost Tracking from Day One: Assign responsibility for monitoring AI spending across compute, data, labor, and infrastructure. Without this baseline, you cannot measure return on investment or identify where money is being wasted.
• Link AI Spending to Business Outcomes: Define measurable metrics before deployment. What specific business problem does the AI solve? How much revenue or efficiency gain should it generate? Track actual results against these targets quarterly.
• Involve Finance Leadership in AI Decisions: Treat AI projects with the same financial rigor as any other strategic investment. Finance teams should have oversight of budgets, timelines, and expected returns, not just technology teams.
• Implement Audit and Governance Controls: Document how AI models are used, who approves their outputs, and how decisions are made. This is especially critical in regulated sectors where compliance audits are mandatory.

Aidana Zhakupbekova, Chief Financial Officer at Rydoo, an expense management platform, emphasized this point directly: "Cost visibility is what separates businesses getting real ROI from AI and those still waiting for it. Confidence in AI is high across the board, but confidence doesn't equal proof, and most businesses still can't say with any precision what their AI is actually costing them to run."

Aidana Zhakupbekova, Chief Financial Officer at Rydoo, an expense management platform

"Without that visibility, AI stops being an investment and simply becomes another operational expense. Finance leaders need to be involved from the outset, with clear oversight of how AI spend is tracked, measured and linked to overarching business outcomes," Zhakupbekova explained.

Aidana Zhakupbekova, Chief Financial Officer, Rydoo

What Does This Mean for National AI Strategy?

The cost discipline problem extends beyond individual companies to national-level AI ambitions. Canada's "AI for All" strategy, released in June 2026, reflects a broader global shift toward sovereign AI infrastructure and domestic compute capacity. The strategy allocates $500 million to the BDC LIFT Program to help small and medium-sized enterprises access financing for AI tools, and another $500 million to Regional Development Agencies to accelerate AI adoption and commercialization.

However, this funding only works if companies actually know how to measure whether their AI investments pay off. Canada's strategy also proposes a Public AI Supercomputer and sovereign cloud infrastructure governed by Canadian laws, which could reduce costs and security risks associated with training models on foreign infrastructure. But infrastructure alone won't solve the measurement problem.

Similarly, the UK government's "Cyber Shield" initiative aims to use frontier AI systems for national cyber defense, leveraging AI's ability to identify software vulnerabilities at unprecedented speed. Yet even at the national security level, the same discipline applies: governments must be able to demonstrate that AI-enabled defense delivers measurable security improvements and justifies the investment.

The broader lesson is clear: AI adoption without measurement discipline is not a technology problem. It's a financial governance problem. Companies and governments that treat AI as an experimental budget line, separate from normal financial controls, will continue to see projects stall and budgets balloon. Those that integrate AI spending into standard financial oversight, measure outcomes rigorously, and involve finance leadership from the start will be the ones who actually achieve sustainable returns.

As AI moves deeper into critical systems, from banking to national security, the ability to prove value becomes not just a business imperative but a regulatory and strategic necessity.