Companies Can't Prove AI ROI

A lot of companies bought artificial intelligence the way people buy a treadmill in January: with a budget, a plan, and no good way to prove what changed three months later. A new report cited by CIO says many firms still cannot show clear return on investment because they never measured the old process well enough to compare it with the new one. (cio.com) KPMG’s latest numbers show the spending is still rising anyway. Its United States first-quarter 2026 Artificial Intelligence Quarterly Pulse says large organizations expect to spend an average of $207 million on artificial intelligence over the next 12 months, nearly double the forecast from a year earlier. (kpmg.com) The strange part is that reported “business value” is much easier to find than hard financial proof. KPMG’s 2026 Global Tech Report says 74% of organizations report business value from their artificial intelligence use cases, but only 24% say they achieve return on investment across multiple use cases. (kpmg.com) That gap usually starts before the software is installed. If a company never tracked how long invoice reviews took, how many customer emails an agent handled per hour, or how often a contract error slipped through, then “faster with artificial intelligence” is just a feeling, not a finance number. (cio.com) Boards are not waiting for perfect spreadsheets. CIO reports that KPMG found three out of four global leaders will prioritize artificial intelligence investment despite economic uncertainty, which means chief information officers are being pushed to spend first and justify later. (cio.com) KPMG’s own surveys show why the pressure keeps building. In April 2025, 93% of surveyed United States leaders said generative artificial intelligence had already improved their company’s competitive position, even while most organizations were still heavy on pilots and light on full deployment. (kpmg.com) That is changing what buyers ask vendors for. Instead of hearing “show me your full platform,” more procurement teams are likely to ask for one narrow job with one owner and one before-and-after scorecard, like reducing claims-processing time from eight days to four or cutting call-summary work from 15 minutes to 3. (cio.com) The winners in that market will be the products that leave an audit trail. A tool that can show baseline volume, labor time, error rate, exception rate, and post-launch results in the same dashboard is easier to defend in a budget meeting than a broad “copilot” that touches everything and measures nothing. (kpmg.com) KPMG’s 2026 pulse data points in the same direction: 65% of respondents said realizing return on investment through scaled use cases is getting harder, even as 54% of organizations have integrated artificial intelligence agents into operations. More deployment is not automatically producing cleaner proof. (kpmg.com) So the next phase of enterprise artificial intelligence probably looks less like one giant companywide bet and more like a stack of small, testable ones. The firms that can say “this workflow used to cost $12 per case and now costs $7” will have a much easier time buying the next tool than the firms still arguing from demos and hope. (cio.com)