Winners Measure What AI Earns, Not How Much They Use

The numbers this year draw a hard line between activity and value. Writer’s 2026 Enterprise AI Study found that 59% of organizations now spend an average of $1 million or more each year on AI, while only 29% report significant returns (Source: Forbes, Companies Track How Much AI They Use. The Winners Track What It Earns, July 2026). Sit with that gap for a moment. Most companies are writing seven-figure checks, and fewer than a third can point to a return worth naming.

The view from the top is starker. In PwC’s 2026 Global CEO Survey of 4,454 chief executives across 95 countries, only 12% said they had seen both higher revenues and lower expenses from their use of AI, and more than half reported no effect at all (Source: Forbes, Companies Track How Much AI They Use. The Winners Track What It Earns, July 2026). So the story is not that AI does nothing. The story is that only about one in eight leaders can draw a straight line from AI spend to a business outcome. Everyone else is measuring effort and hoping it turns into money.

Usage Is A Vanity Signal

Token counts, seat licenses, active users, and adoption rates all feel like progress. They are easy to pull and they always go up. But they answer the wrong question. They tell you how busy your AI is, not what it earned.

OpenAI’s CFO Sarah Friar made this point plainly in July, arguing that for years software success was measured through adoption, seats, active users, and renewals, and that AI has to be measured by the work it actually accomplishes instead (Source: Fortune, OpenAI’s CFO: 4 questions that reveal if your AI spend is paying off, July 2026). Her proposed metric, useful intelligence per dollar, asks four questions: is AI completing work that matters, what does each successful task cost, can people depend on the result, and does each dollar produce more value as usage grows (Source: Axios, OpenAI’s CFO pitches a new way to measure AI’s value, July 2026). Notice that not one of those questions is about how much you used.

The cost point matters more than it looks. Friar argues that the full cost of a successful task includes AI usage, retries, and the cost of human review, not the token price alone (Source: Axios, OpenAI’s CFO pitches a new way to measure AI’s value, July 2026). A cheap model that needs three attempts and a person to clean it up is not cheap. That hidden human cleanup is exactly where usage dashboards go quiet.

Judge AI As A Capital Allocation, Not A Technology

The leaders who get real returns treat AI spend the way they treat any other claim on capital. It has to compete. It has to earn its place against every other use of the same dollar.

The evidence that scale alone does not buy returns keeps stacking up. Domino Data Lab’s Fifth Annual Enterprise AI Report, based on 639 senior enterprise AI leaders, found that the share of enterprises whose ROI fails to outpace their investment has held at 57% since 2025, even as 93% now report improved production capability, up from 88% (Source: PRNewswire, AI ROI Fails to Outpace Spend for 57% of Enterprises, July 2026). Companies are getting better at shipping AI and no better at profiting from it. Getting a model into production used to be the milestone that mattered, and it is not enough anymore.

McKinsey’s read lands in the same place. Its research found that 64% of companies say AI is driving innovation, but just 39% report a measurable impact on earnings (Source: CNBC, Almost every Fortune 500 tracking AI usage, May 2026). Innovation is a feeling. Earnings are a number. The gap between those two figures is the gap between a story and a defensible case.

What Good Actually Looks Like

Here is what we track when we want a return we can defend, and what each signal tells you.

The job. Every tool gets one line describing the work it does: issues resolved, contracts reviewed, code shipped, hours returned. If you cannot write that line, the tool does not have a job, it has a vibe.

Cost per successful outcome. Not spend, not tokens. The all-in cost of getting the work done right, including the human time spent fixing, checking, and re-prompting. Good looks like this number falling over time.

Dependability. What share of output is ready to use, versus needs correction, versus needs escalation to a person. A tool that is used constantly but corrected constantly is a cost, not a return.

Value at scale. Whether each additional dollar buys more finished work or just more activity. If usage climbs and outcomes flatten, you have found waste.

The honest way to read these signals is to ask them of the people doing the work, quietly and often, because they know where the cleanup hides. That is the whole reason short anonymous micro-surveys exist. Dashboards see the tokens. People see the truth.

If your board asked tomorrow which of your AI tools earned their keep, we would love for you to have the answer ready. Start a GAiGE trial and let your teams tell you, in a few honest minutes, which tools are working and which are just busy.