We bought AI licenses and nobody uses them. What do we do?
Stop counting licenses for a while. Pick one workflow that hurts, with someone inside who owns the result, and work it with the people who do it every day until the old step can be switched off. Usage comes when the tool solves a problem the team already had.
Why it happens
It happens in a lot of companies. ChatGPT, Copilot or Gemini gets bought for everyone, there's a training session, and after a while usage settles on a few people. Most open it now and then to draft an email.
The reasons are almost always these:
- Nobody said what it was for. A license without a specific job is one more browser tab.
- The work stays the same. People write faster, but the proposal still waits for the same approval, the same pricing and the same review. AI sped up the visible part and exposed the same old queue.
- It becomes a double shift. The report gets done the usual way and again with AI, and nobody says when the old one stops.
- Nobody owns the result. If it goes wrong, there's no one to decide whether to fix it, keep going or roll back.
Access, capability and substitution
It helps to separate three things:
- Access: the tool is available. The license buys that.
- Capability: the team produces an acceptable result, again and again. That gets built.
- Substitution: the new cycle passed a threshold and someone can retire an earlier step. That gets decided.
Many companies celebrate the first, work hard on the second and never reach the third. The pilot becomes permanent.
What to do, in order
- Pick one workflow. One, with a name: the weekly sales report, technical answers to customers, quoting a project.
- Name who owns the result. If there's no one, don't start yet.
- Look at how it runs today. How long it takes, what quality it delivers, what controls it has, what exceptions show up.
- Work it with the people who do it. Gather real examples, the sources they use and how they know a result is right.
- Set a date and a threshold. Before starting, agree on what the new cycle has to show to switch off the old one, and who decides that day whether to retire, fix or roll back.
- Move on to the next workflow.
How to tell if it's working
Active users, prompts, tokens and hours saved show how much it's used. To know whether the company changed, ask what important work changed in a way you can verify and which old step stopped being done. If nothing stopped, you're still paying twice.
What usually doesn't help
- Buying more licenses or switching tools.
- Another prompt course for everyone, with no workflow behind it.
Where Business Cyborgs comes in
Business Cyborgs works that step: one workflow at a time, with the people who run it, until it's running inside the company. If there's curiosity but no clear case, we start with a Talk of 60 to 90 minutes. If there's already a workflow that hurts, with an Agent Sprint of 2 to 4 weeks. If there are several workflows, scattered data and a mandate from leadership, with a Transformation of 3 to 6 months.
Some published cases, all anonymized:
Expert technical knowledge turned into a support agent for staff and sales, with curated sources.
A project-based business that went from fragile spreadsheets to agents that help quote, compare suppliers and understand cash and margin.
A consumer goods company that shut down its marketing agency and replaced it with an agent-based operation.
Further reading (in Spanish): BC 40: Don't confuse installation spend with installed capability and BC 56: If nothing stops being done, you didn't migrate work.
FAQ
We bought ChatGPT Enterprise or Copilot and nobody uses it. What do we do?
Pick one specific workflow that hurts, name someone who owns the result and work it with the team that does it until the old step can be switched off. More general training or more licenses rarely change usage if nobody said what work the tool is for.
How do you move a company from AI tool access to real AI adoption?
Separate access, capability and substitution. The license gives access. Capability shows up when the team produces an acceptable result repeatably. Adoption shows when someone can retire an earlier step without adding risk, on an agreed date.
How do you measure AI adoption in a company?
Active users and hours saved show usage. To measure adoption, look at what important work changed in a verifiable way and which old step stopped being done. Before starting each workflow, write down how it runs today and what the new cycle has to show.