Depth, not breadth: what the new PYMNTS data tells you to measure

There is a number in the latest PYMNTS Intelligence research that should change how you read your own AI dashboard. It shows that counting deployments tells you almost nothing about whether the money is working.

The report, "AI at Work: Why Deeper Enterprise Use Produces Stronger Returns", is based on a survey of 60 senior technology executives at US enterprises with at least US$1 billion in annual revenue, fielded 14 to 22 July 2026 (Source: PYMNTS Intelligence, AI at Work: Why Deeper Enterprise Use Produces Stronger Returns). The survey measured whether AI was in use across eight business functions and, within those functions, tracked use across 75 specific tasks to capture how much work AI actually affects, not just whether it is present (PYMNTS Intelligence).

The finding that should change your dashboard#Link to this section

Here is the part worth pinning to the wall. Companies with AI embedded in one or two functions use it across an average of 41 of the 75 tasks measured. Companies with AI embedded in three or more functions use it across 40. Virtually no difference (PYMNTS Intelligence). If the amount of AI work predicted returns, these two groups would look similar on returns too.

They do not. Among companies with one or two embedded functions, 55% reported generating returns from AI investments. Among the deeper adopters, that figure reached 93% (PYMNTS Intelligence). Only a quarter of companies that had not embedded AI in any function said it was already delivering a return (PYMNTS Intelligence).

The task volume is the same, and the returns are very different. The variable that moves is depth of use inside a function, not the number of functions or tools you can point to on a slide. As the researchers put it, depth of AI tool usage, not breadth, is driving financial outcomes (PYMNTS Intelligence).

There is a second detail that reframes the whole picture. Only 20% of the 437 deployments examined are embedded in a function's regular work. Another 37% are broad, meaning multiple teams touch AI without making it a core part of daily operations, and the remaining 43% sit in limited use or early exploration (PYMNTS Intelligence). Most of what gets counted as adoption is presence, not practice.

Activity metrics and outcome metrics#Link to this section

Most AI reporting leans on activity metrics, because they are easy to pull: seat counts, licences issued, the number of functions with a tool switched on, logins. These answer the question "is AI present?", and the PYMNTS data shows that question does not separate the organisations seeing returns from everyone else.

The metrics that track with return are depth metrics. They answer a different question: is AI part of how the work actually gets done? A few worth tracking directly:

• Depth per function, not tools per company. For each function, where does AI sit on the path from exploring to embedded? The report placed each function in one of four stages of deployment for exactly this reason (PYMNTS Intelligence).
• Share of real work touched. Of the tasks a team actually does, how many does AI meaningfully affect, and how often? Task coverage without frequency tells you little.
• Self-reported return by function. Ask the people doing the work whether the tool saves time, improves quality, or replaces a slower path. This is where the 55% against 93% gap lives, and it is a number a CFO can follow.
• Unused licences. Deployments in broad or limited use (80% of those PYMNTS examined) aren't waste in themselves, but they are where to look for seats nobody opens.

What good looks like#Link to this section

Good is not a wide footprint. Good is a function where AI is embedded in regular work and the team confirms it is producing a return. If you are choosing between switching AI on in a fourth function and pushing your first two from broad to embedded, this data points at depth.

The researchers name a caution themselves: the survey shows an association, not proof that deeper use caused higher returns (PYMNTS Intelligence). Depth is a strong signal, and it is measurable. It also comes with friction. Companies with three or more embedded functions reported an average of 5.6 barriers to adoption, compared with three among companies with nothing embedded (PYMNTS Intelligence). The organisations that have gone further report more barriers to work through, not fewer.

So the practical move is to stop grading yourself on how many tools you own and start measuring, function by function, whether AI is embedded and whether the people using it say it pays. That is a number you can put in front of a board.