Are you actually using AI, or just paying for it?

Every week I talk to executives who tell me their company has rolled out AI tools. When I ask how it is going, the answer is usually some version of: "People are using it some. We think. It is hard to tell."

That uncertainty is usually a bad sign. If AI was working well, they would know. The wins would be obvious. People would be talking about it.

The hard truth is that most companies are not getting the return on their AI investment that they think they are, or that they could be. They are paying for capability they are not using.

The utilization gap is real and it is large

Industry surveys consistently put enterprise AI license utilization rates around 20-25%. That means at a company where every knowledge worker has an AI license, roughly three out of four either do not use it at all or use it so sporadically that it has no meaningful impact on their work.

That is a brutal number. It means that if you are spending $100,000 per year on AI licenses, you are getting $23,000 worth of value at the average utilization rate. The other $77,000 is essentially a sunk cost for a tool that sits open in a browser tab and gets used occasionally for drafting emails.

I see this in almost every organization I work with. Not because the people are lazy or the technology is bad. Because purchasing licenses and actually getting value from those licenses are two completely different things, and most companies only did the first one.

Five questions that reveal where you actually stand

Most leaders do not have a clear picture of their AI utilization because the signals are easy to misread. Here are five questions worth asking yourself honestly.

1. Can your team describe a specific workflow they do faster because of AI?

Not "do they use AI sometimes." Can they name a specific task, describe how they do it with AI, and tell you how long it used to take versus how long it takes now?

If you ask that question in your next all-hands and get blank stares or vague answers, you have a utilization problem. Real adoption produces specific stories. "I used to spend two hours building the competitive brief before a pitch. Now it takes me 25 minutes." That is what good adoption looks like.

2. Do your managers talk about AI in their one-on-ones?

AI adoption lives or dies at the manager level. If your managers are not asking their reports about AI workflows, not celebrating wins, and not modeling usage themselves, the tools are not going to stick regardless of how good they are.

The quickest way to find out is to ask a few managers directly: "When did you last talk to your team about how they use AI?" If the answer is "in the kickoff training we did six months ago," that tells you something.

3. Have you ever calculated what your team's most time-consuming repetitive tasks are worth in salary?

Take one task. Something your team does regularly that takes a few hours and is largely mechanical: building reports, reviewing documents, doing research, writing first drafts. Calculate how many hours per week it consumes across your team, multiply by the average hourly cost of the people doing it, and you have a dollar figure for what that task costs you annually.

Now ask: is that task running faster than it was before you bought AI licenses? If not, you have your answer on whether the investment is working.

4. Do new employees get AI workflow training, or just tool access?

How a company onboards new employees to AI is a good proxy for how seriously they take adoption. Giving someone a license and a link to the help docs is not training. It is the equivalent of hiring someone and handing them car keys without telling them where anything is or what routes to take.

If your onboarding process includes "here is your ChatGPT license" but not "here are the three workflows our team uses AI for, here are the prompts, here is how you QC the output," you should expect that new employee to add their license to the utilization gap.

5. What happens when the AI produces bad output?

In teams with good adoption, people know how to handle AI errors. They have a mental model for when the tool is reliable and when to check its work carefully. They treat it like a junior colleague rather than an oracle.

In teams with poor adoption, bad AI output becomes a reason to stop using the tool. Someone gets a weird result, mentions it in a meeting, and suddenly half the team is skeptical. Without a support structure, the skepticism spreads and utilization falls.

If your team does not have a clear answer for "what do we do when AI gets something wrong," that is a gap worth closing.

What low utilization usually looks like in practice

When I do AI workflow audits, the patterns of underutilization tend to fall into a few categories.

The most common is the shallow use case. People use AI for things like "help me write a better email" or "explain this concept to me" but not for the higher-value work that would actually move the needle. They are using a power tool to drive finishing nails.

The second is the trained-but-not-embedded group. They went through an AI training session, thought it was interesting, used the tools for a week or two, and then drifted back to their old workflows. Without structured integration into their day-to-day work, the new behavior did not stick.

The third is the skeptics who got burned. They tried to use AI for something important, got a hallucination or a confidently wrong answer, and decided the tools are not ready for their work. These people need a different conversation about use cases and QC rather than more general training.

The difference between organizations that get value and those that do not

I have worked with organizations that have transformed productivity through AI and organizations that have essentially wasted their AI budgets. The difference almost never comes down to the tools they chose.

What separates high-utilization organizations is consistent internal structure around AI. Someone owns the adoption problem. There are specific workflows that have been built, tested, and documented. New employees learn those workflows on day one. Managers are expected to support and reinforce AI use. Wins get shared. Problems get addressed.

Low-utilization organizations bought licenses, did a one-time training, and hoped behavior would change on its own. It does not work that way with any significant change, and AI is not an exception.

What to do if you are in the utilization gap

The good news is that low utilization is fixable. It requires structured effort, not another software purchase.

Start by auditing where your team's time actually goes. What are the 3-5 most time-consuming manual tasks? Pick the one with the highest time cost and the clearest quality standard. Build a specific AI workflow for that task: the prompt, the process, the verification step. Test it. Train a small group on it. Measure the results.

Once that first workflow is running well and people are talking about the time savings, expand to the next one. Adoption spreads from demonstrated value, not from mandates or license counts.

The companies getting real value from AI right now are the ones who treated implementation as a project, not an assumption. If you bought licenses and expected transformation to follow automatically, you need to go back and do the implementation work.

The window for early-mover advantage is not closed, but it is closing. Teams that figure this out in 2026 will have a structural speed advantage over teams still running the same manual processes they were in 2024. The tools are good enough. The question is whether you have built the workflows to use them.