Only 6% of leaders have a framework that tracks AI returns consistently: what one looks like

The numbers in Battery Ventures' latest enterprise technology spending survey are worth sitting with for a moment. According to the survey, 98% of respondents are increasing what they spend on AI over the next 12 months, and not a single CXO surveyed is cutting back (Source: Battery Ventures, Measuring AI ROI: The Next Big Conundrum for Enterprises, 2026).

Here is the other number from the same survey. Only 6% of leaders have a well-defined measurement framework tracking ROI consistently across the organisation. Another 42% are measuring in some areas but not consistently, and 43% are still defining how to measure at all (Battery Ventures, Measuring AI ROI). Just over half, 53%, say they see clear ROI from some area of AI investment today (Battery Ventures, Measuring AI ROI). Seeing returns somewhere is not the same as being able to prove them everywhere.

That gap between spending and proof is the whole reason a measurement discipline exists. So let us be concrete about what closing it involves.

What separates the six percent from everyone else#Link to this section

By the survey's own definition, the 6% have a repeatable method applied across the organisation, rather than a one-off deck pulled together the week before a board meeting.

A revealing detail sits in the Battery data. It found that 62% of enterprises already report AI ROI to the board, yet only 53% can actually point to a clear return (Source: Battery Ventures, State of Enterprise Tech Spending Survey, 2026). More people are presenting numbers than have numbers that stand up. The reporting exists; the framework underneath it often doesn't.

KPMG's June 2026 pulse of more than 2,145 leaders across 20 countries found only 7% report established AI ROI, which lines up almost exactly with Battery's 6% (Source: KPMG, Global AI Pulse, June 2026). The same KPMG research found that organisations with cost monitoring dashboards and embedded cost reviews were five times more likely to report established ROI, 15% against 3% (KPMG). Consistent measurement is a practice you can install, and in KPMG's data it goes with established ROI.

What to track, and how to read it#Link to this section

Most teams measure the easy things. The Battery survey shows cost savings at 76% and employee productivity at 73% are the dominant lenses, with speed of delivery close behind at 61%. Revenue attribution at 38%, quality improvement at 40% and customer satisfaction at 37% are still secondary (Battery Ventures, State of Enterprise Tech Spending Survey). In other words, most measurement captures effort and efficiency, not business value.

We would push you to track four things, tool by tool and team by team.

First, adoption that means something. Not seat counts, but who actually uses a tool in a normal week and who quietly stopped. A licence nobody opens is a cost hiding in plain sight.

Second, impact as the people doing the work experience it. Ask whether a tool saved time this week, and roughly how much. Short, anonymous, and repeated. One answer is a data point; a trend line is evidence.

Third, a baseline captured before the tool goes live. This is the one everyone skips. Plug and Play's 2026 survey found that a use case which goes live without a pre-deployment metric can't be measured later, because there is nothing left to compare against (Source: Plug and Play, 2026, reported by Forbes). If you take one thing from this post, take that.

Fourth, waste. The licences paid for and unused, the overlapping tools doing the same job, the pilots that never scaled. This is often the quickest money to find.

What good looks like is simple to describe. You can name, for any AI tool, what it costs, how many people use it, what they say it does for their work, and how that compares to where you started. You can do it this quarter and next quarter using the same method, so the numbers are comparable over time. And when someone asks a harder question, and Battery's own conclusion is that the scrutiny is coming, you have an answer that holds up (Battery Ventures, 2026).

A counterintuitive warning from the same Plug and Play research. The narrowest deployments are often the worst measured, with 74% of companies running AI in a single function saying ROI is too early to measure or isn't tracked at all (Plug and Play). Small scope should make attribution easier, not harder. If you aren't measuring the simple cases, the complex ones will be hopeless.

Battery's 6% have a framework that tracks returns consistently across the organisation. That's a method, and any organisation can adopt one.