Your team is already using AI. Start by asking, not banning

If your company hasn't approved an AI tool yet, your team has probably approved a few on its own.

That isn't a moral failing. It's the normal starting point, and it's where Phase 1 of our AI Transformation Playbook, Fluency, begins. The goal of the phase, in the playbook's words, is to "make experimentation visible, skillful, and safe enough to learn from — so the organization stops pretending nobody is already using models."

The last post in this series covered the prerequisite: name a sponsor, a chair, and a small team with protected time. This one is the first thing that team does. What follows is the playbook's method, plus what this year's surveys say about how teams are really using AI.

What the 2026 surveys say

Four surveys, four different samples. I'm keeping each figure with its own source, because they measure different things and don't add up.

• Okta commissioned a survey from Apprize360 of 292 executives and 492 knowledge workers in seven countries (March 2026). 90% of executives were confident in their visibility into AI tools, while 52% of workers said they'd used AI tools without approval. The reasons are the useful part: 80% said it's easier to use their own accounts, and 57% said the approval process is too slow or difficult. Okta sells identity security, so read it as a vendor survey.
• PagerDuty commissioned Wakefield Research to survey 1,250 office professionals at companies with $500M+ in revenue (April 2026). Among those who use AI for work, 66% had used AI tools they believed weren't allowed under company policy, and 81% believe leadership plays by different AI rules. Also vendor-commissioned.
• Deloitte UK's GenAI Workforce Survey (25,000 UK workers, fieldwork May–June 2026) found 31% of GenAI users use it without their employer knowing. Its conclusion: "the answer isn't simply to introduce stricter rules."
• TrustedTech, a Microsoft cloud solutions provider, had Censuswide survey 2,001 UK and US employees (March 2026). More than a quarter said they would keep using AI even if their workplace banned it, as reported by Channel Dive. Vendor-commissioned too.

None of these is your company. But they point the same way: the use is already there, and a rule on its own doesn't make it go away. Okta's gap between executive confidence and worker answers is the one I'd keep in mind before assuming your own picture is accurate.

Why a ban backfires

The playbook names this failure mode Ban without alternative. What it looks like: "Shadow AI goes quieter and riskier." The work doesn't stop. It moves to personal accounts, where nobody can coach it, and people stop mentioning it.

Play 1 of Phase 1 puts it plainly: "Prohibition theater drives the work underground; an inventory lets you coach and measure." The playbook calls the play Inventory shadow use (quiet honesty). The quiet part matters. You're not running an audit. You're asking people to tell you what's true.

What to ask your team

Keep it short. Five questions cover most of it:

1. Where does AI already show up in your week? Drafting, summarizing, code assist, support macros, slide cleanup. Give examples so people recognize their own habits.
2. Which tools, and on whose account? Personal, company, free, paid out of pocket.
3. What kind of information goes in? Public material, internal docs, customer details. Ask in categories, not specifics.
4. What has actually helped, and what went wrong? One example of each, if they have one.
5. What would you use if it were approved? This tells you what a sanctioned option has to beat.

Add one more for leaders, and ask it first: what do you use? If PagerDuty's respondents are any guide, people notice when the rules stop at the leadership floor.

How to run it

I'd pair two formats:

• A short anonymous survey to get the breadth. Five questions, ten minutes, with a sentence at the top saying what you'll do with the answers and what you won't.
• A listening session per team, run by that team's fluency owner with someone from the AI team. Bring their real work: their tickets, emails, and docs. You get the stories here, the "I use it for this, but I'm not sure I should."

Say up front that this isn't a disciplinary exercise. Then keep that promise. If someone admits to pasting a customer list into a free chatbot, the response is a better path and a clear rule, not a write-up. The first person who gets punished for being honest ends the inventory.

What to do with the answers

The answers become the first of Phase 1's exit artifacts:

• A shadow-use inventory. Tool, use case, team, type of data. It's a living doc, not a one-time report.
• A one-page acceptable-use note that's readable. What's approved, what data never goes in, and how to ask for something new. Okta's 57% "too slow" figure is a good reason to make that last part fast.
• Sanctioned tools plus craft sessions. This is the playbook's repair for a ban: inventory, sanctioned tools, craft sessions. Run short sessions with your own domain examples, and capture "good / better / don't ship" examples per function.
• A quality bar for anything external. Human review before anything reaches a customer or the public.
• A first gated-agent candidate list. The inventory shows you where repetition already lives.

The playbook's 90-day picture puts the inventory, the acceptable-use note, and named fluency owners in days 1–30.

The other failure modes

Ban without alternative isn't the only trap in Phase 1. The playbook lists four more:

• Prompt theater: workshops with no domain examples. Rebuild them with your tickets, emails, and docs.
• Fake ROI dashboards: "AI productivity" charts with no method. Kill the chart. Keep adoption and quality signals.
• Premature agents: multi-agent demos while fluency is still chaos. Park the demos and finish the candidate list and quality bar.
• No owner: "everyone owns AI." Name people and review monthly.

One more, from the playbook's logging play: keep tracking to "learning telemetry, not surveillance cosplay." Tool, use case, reviewer, outcome. If the inventory starts to feel like monitoring, people go quiet again.

What to do this week

Have the AI team draft the five questions. Have each fluency owner book a listening session. Send the survey with the promise attached. Leaders answer first.

You won't get a perfect picture, and that's fine. A partial, honest one beats a confident one you made up.

Get the playbook

The full Phase 1 section, with all six plays, the exit artifacts, and the failure-modes table, is in the 231 Labs AI Transformation Playbook. It's free at 231labs.com/playbook.