The companies winning with AI this week aren't the ones spending the most. They're the ones who can tell you exactly what it costs.
That's the finding buried in KPMG's latest AI Pulse, and it's the most useful sentence in a month of AI research: leaders with strong cost visibility are 5x more likely to report established ROI — 15% vs. 3%. The gap between winning and wasting money on AI isn't budget. It's bookkeeping.
Sit with that for a second, because it flips the usual story. We assume the companies getting returns are the ones with the biggest AI teams and the deepest pockets. The data says the opposite: the ones who simply track what their AI costs and what it returns pull away from everyone else. And almost nobody is doing it — only about 1 in 3 enterprises has full visibility into what their AI actually costs to run. Here's the week's data, and what it means if you're a professional trying to make your first AI spend actually pay.
The numbers worth knowing this week
Cost visibility is the single biggest predictor of AI ROI. KPMG pegs average enterprise AI spend at $188M, yet only 7% of leaders report fully established ROI. The ones who do share a habit: strong visibility into AI costs makes an organization 5x more likely to report established returns (15% vs. 3%) — while 42% have only partial visibility into what they're spending (KPMG)
The ROI-vs-spend problem has plateaued for two straight years. Domino's Fifth Annual Enterprise AI Report (639 senior AI leaders) found 57% of enterprises still say their ROI doesn't outpace their investment — unchanged since 2025 — even as 93% now report improved production capability (up from 88%). The bottleneck is a "last-mile" gap between models in production and the people who'd actually turn them into value (Domino Data Lab)
Confidence is rising — but most of it is still a forecast, not a receipt. A SAP / Oxford Economics survey of 2,600 leaders across 13 countries found companies expect to drive 21% ROI this year (~$6.3M), up from 16%, with projected agentic AI returns reaching $17.6M within two years. Worth reading as expected ROI, not yet realized. On the realized side, satisfaction with current AI ROI hit 69%, with AI now assisting 30% of tasks (up from 25%) (SAP/Oxford via CIO Dive)
Customer service still has the clearest named payoffs — read the fine print. Rocket Money reports nearly $1M in annual efficiency gains at a 68% resolution rate, and Anthropic is cited saving 1,700+ support hours in month one — though both figures are vendor-published customer stories, not independent audits. Gartner previously forecast conversational AI could cut $80B in contact-center labor costs in 2026, but its newer 2026 guidance warns GenAI cost-per-resolution may actually rise over time — full-automation ROI isn't guaranteed (Marketplace, Fin/Rocket Money, Gartner 2026)
The pattern that matters
Read these together and the dividing line is no longer "is AI working?" It's "can you prove what it cost you?"
Everyone can feel a productivity bump — 93% of enterprises report better production capability. Far fewer can turn that into a number they'd defend, which is exactly why 57% still say ROI doesn't beat spend, and why 42% can't fully see what they're paying. The winners aren't a smarter tier of company. They're the ones keeping the receipts. When you can see cost against outcome, you cut what's dead weight and double down on what's landing — and your ROI follows. When you can't, you're guessing with a bigger and bigger checkbook.
The EMDUIT take — for the professional just getting started: You don't have $188M or a finance team building AI cost dashboards. But you have the exact same problem in miniature — and this week it comes with the exact same fix.
Be honest: right now, could you say what you spend on AI each month, and what it gave you back? Most solo professionals can't. There's the ChatGPT subscription, maybe a transcription tool, a writing assistant bundled into something else, a "just trying it" trial that quietly renewed. That's your cost-visibility gap — and it's why AI feels vaguely worth it but never obviously so.
The Fortune 500 fix scales all the way down to your desk: keep the receipts. One line in a note or a spreadsheet — tool, monthly cost, what it actually completed for you. That's it. That single habit is what makes an enterprise 5x more likely to see real returns, and it costs you nothing but ten minutes. Once you can see cost against outcome, the hard calls make themselves: the tool that clearly saved you three hours on intake stays and expands; the one you can't remember using gets cancelled.
And borrow enterprise skepticism while you're at it. The flashiest AI wins this week — the $1M customer-service numbers — came from the vendors selling the tools. That doesn't make them false, but it's a reminder: trust your own receipts over anyone's case study. Your ledger is the only ROI report that's actually about you.
The winners this quarter aren't spending more. They're measuring better. You can start doing that today, for free — and it's the cheapest edge in all of this week's data.
Sources & methodology
Sources: KPMG AI Pulse · Domino Fifth Annual Enterprise AI Report · SAP/Oxford Economics via CIO Dive · Customer-service ROI via Marketplace, Fin/Rocket Money (vendor-published) · Gartner 2026 guidance · Futurum enterprise AI ROI research. Compiled from reports published July 21–25, 2026, plus recent 2026 benchmark reports from KPMG, SAP/Oxford Economics, Futurum, and Gartner. The Futurum data is from earlier in 2026; customer-service figures are vendor-published case studies.