Teams usually stop using AI tools because nobody connected the tools to specific tasks people own, taught the habits that make output reliable, said clearly what is allowed, or made time to change how the work is done.
That is good news. Each of those causes is fixable without buying anything new.
Who this is for
COOs, heads of people and operations leaders who rolled out ChatGPT, Claude, Copilot or Gemini and are now looking at low or uneven usage.
The short version
- Low adoption is rarely a tool problem. It is usually a relevance, skill, permission, time or proof problem.
- Diagnose before you fix. Each cause needs a different response, and the wrong fix wastes goodwill.
- Restart small: one team, real tasks, visible follow-up.
The five reasons adoption stalls
We find it useful to check five things, roughly in this order. A stalled rollout is often stuck on two or three at once.
1. Relevance: people can’t see where it fits. The tool was introduced as “use AI for everything,” so it ends up used for nothing in particular. People are busy, and a general invitation loses to a specific deadline.
How to check: ask five people to name one task they now do with AI. Vague answers (“emails, sometimes”) point here.
Fix: give each role two or three named tasks to try, drawn from their actual week.
2. Skill: early results were disappointing. Someone asked a hard question with no context, got a generic or wrong answer, and concluded the tool isn’t good enough. Usually the gap is in how the task was set up, not in the tool.
How to check: watch someone use it on a real task. Short prompts, no examples and no iteration are the tell.
Fix: hands-on practice on real work, focused on three habits: give context, ask for structure, and check against the source.
3. Permission: careful people aren’t sure what’s allowed. If there is no clear guidance on client data, confidential documents or approved tools, conscientious staff tend to avoid AI entirely, while others use it quietly with no guardrails. Both outcomes are bad.
How to check: ask, “Could you paste a client document into this tool?” If answers differ across the team, guidance is missing.
Fix: a one-page “what’s allowed” guide with approved tools, data categories and a named person to ask.
4. Time: there is no room to learn. Changing a workflow costs time before it saves time. If the first attempt makes a task slower, people under deadline pressure will revert.
How to check: ask whether anyone has been given time, even an hour a week, to practice.
Fix: protect a small, regular block for practice and sharing, and lead by example from the top.
5. Proof: nobody has seen it work here. Abstract demos don’t persuade skeptical professionals. Seeing a respected colleague save real time on a familiar task does.
How to check: can anyone point to a specific internal example of AI improving a piece of work?
Fix: collect and share small wins, with the actual prompt and output, in a place people already look.
A quick diagnostic
Score each reason from 1 (not a problem) to 3 (clearly a problem). Start with whichever scores highest, and don’t try to fix all five at once.
| Reason | What you’d hear | First fix |
|---|---|---|
| Relevance | “I’m not sure what to use it for” | Named tasks per role |
| Skill | “It gave me rubbish” | Hands-on training on real work |
| Permission | “Am I allowed to do that?” | One-page usage guide |
| Time | “I’ll try it when things calm down” | Protected practice time |
| Proof | “Has anyone here actually used it?” | Shared internal examples |
A 30-day restart plan
- Week 1: Pick one team and two or three of its recurring tasks. Publish the one-page usage guide if it doesn’t exist.
- Week 2: Run a hands-on session using those real tasks. Everyone leaves with a prompt or checklist they will use that week.
- Week 3: Short check-ins. What worked, what didn’t, and where did people get stuck? Adjust the prompts.
- Week 4: Share the results internally, including what didn’t work. Decide whether to extend to a second team.
Measure the change in terms of work: which tasks are now faster, better or less painful. Seat utilization is a weak signal on its own.
Common mistakes
- Sending a reminder email. Encouragement doesn’t fix relevance, skill or permission.
- Buying a different tool. Switching vendors rarely helps if the underlying causes stay the same.
- Blaming the holdouts. Skepticism is often a reasonable response to poor early results.
- Training everyone in one generic session. Different roles need different tasks and examples.
- Leaving senior people out. If partners and leaders don’t use it visibly, others notice.
When you do not need us
If the diagnostic points mainly to permission or time, you can likely fix it yourself: write the usage guide, protect an hour a week, and ask leaders to share how they use AI. Bring in help when skill is the main blocker and nobody internally has the time or experience to teach it well.
How Ruby helps
- AI training: hands-on sessions built around each team’s real tasks, with follow-up so habits stick.
- AI strategy: identifying the workflows where adoption will matter most, plus a clear usage guide.
- AI engineering: for the recurring work that people shouldn’t have to prompt by hand at all.
A practical way to restart is one hour of hands-on AI training with a single team.