How to roll out AI to your team is mostly a question of habits, not licenses. Many companies give employees access to an AI tool, send an announcement, and find a few months later that only a minority use it. This guide gives you a 90-day plan for rolling out AI tools to employees: the phases, training that changes how people work, a champions network, usage rules built on the data labels you already have, and the metrics that show whether adoption is real.
How do you roll out AI tools to employees?
Roll out AI in four phases over about 90 days: prepare (approve the tool, write short usage rules, name a sponsor and champions), pilot with one or two teams on real tasks, expand team by team with role-based training, then make AI part of normal processes and measure it. Access alone doesn't create use. Managers who support it, tools that fit existing work, and clear rules do.
Access is not adoption. In Gallup's survey of 23,717 US employees (fielded February 2026, published April 2026), only 46% of individual contributors at organizations that offer AI tools used them a few times a week or more, against 67% of leaders.
Managers make the difference. In the same survey, 78% of employees who strongly agreed their manager supports AI use were frequent users, against 44% of everyone else.
People already use AI, approved or not. In Microsoft and LinkedIn's 2024 Work Trend Index (May 2024, 31,000 knowledge workers), 78% of AI users said they bring their own AI tools to work. A rollout gives them a safe, approved option.
Train on real tasks and on checking. People need to know where AI helps and where it fails.
Measure depth, not logins. Track who uses AI in their core work, by team, and whether quality holds.
[02]
Why AI rollouts stall
Gallup asked employees who don't use the AI their employer provides why not. Among non-users, 46% said they prefer their current way of working, 43% cited concerns about privacy, security or compliance, and 39% said AI can't help with their work. None of these is fixed by more licenses.
Gallup also found three conditions that go with frequent use, each measured as "strongly agree" against everyone else:
Condition | Frequent users when true | Frequent users otherwise |
|---|---|---|
My manager actively supports AI use | 78% | 44% |
AI tools fit well with our systems and processes | 88% | 55% |
My organization has clear AI policies | 68% | 47% |
These are associations from one survey, not proof of cause, but they point to the levers a rollout controls. A large Danish study points the same way. Linking surveys of about 25,000 workers in 11 occupations to employment records, Humlum and Vestergaard (NBER working paper, revised March 2026) found that take-up of AI chatbots almost doubles where employers actively encourage use. Where companies combined encouraged use with an enterprise tool and training, 93% of workers had used AI chatbots at work and 28% used them daily.
Leaders who get value spend accordingly. BCG's October 2024 survey found that the most advanced companies put about 10% of their AI resources into algorithms, 20% into technology and data, and 70% into people and processes.
[03]
When to roll out broadly and when to start small
Roll out broadly a general assistant for writing, summarizing and research that IT and security have approved, used with low-risk data. Everyone gets access, rules and a short training.
Start with one or two teams for tools that touch customer data, connect to business systems, or take actions on their own, such as AI agents. Pick the first teams with the scorecard in how to choose AI use cases.
Wait if there is no approved tool for the data people would need to use, or nobody owns the rollout.
[04]
Step 1: Prepare (weeks 1 and 2)
Approve the tool with IT and security, including which data classes it may handle.
Name the people: an executive sponsor who announces the rollout and explains why, a rollout owner who runs it day to day, and one champion per team.
Pick two or three starter tasks per role, such as drafting client updates, summarizing meeting notes or turning data into a first-draft report.
Measure the baseline: a short survey of who uses AI today and with which tools, and how long the starter tasks take.
Write one page of usage rules (next section).
Usage rules for data you already classify
Your company probably labels data already, for example public, internal, confidential and restricted. Build the AI rules on those labels instead of inventing new ones, so people only learn one system. Use your own company's names and definitions:
Data class (your company's labels) | Company-approved AI tool | Personal or free AI accounts | Examples |
|---|---|---|---|
Public | Yes | Yes | Published web pages, press releases |
Internal | Yes | No | Process documents, meeting notes without client details |
Confidential | Only tools IT has approved for this class | No | Client records, contracts, unreleased financial results |
Restricted | Only with approval from the data owner and security | No | Passwords, access keys, security reports |
Add a few plain rules under the table:
You own the result. Check anything AI helped you produce before you send or publish it.
Never paste passwords, access keys or tokens into any AI tool.
Keep decisions about people with people. Don't use AI alone to decide on hiring, performance, pay, health, credit or legal matters for a person. Involve HR or legal before AI is used in any of these.
Report surprises, not just incidents. Tell your champion when AI gets something badly wrong or when you pasted something you shouldn't have. No blame, so you hear about problems early.
Some uses of AI are regulated in some places. Ask your legal and compliance team what applies to you before you roll this out. For what to tell employees about public AI tools, see is it safe to use ChatGPT at work?
Choose and support champions
A champion is an employee who helps teammates use AI well. Choose people who are respected for their work, not necessarily the most technical. MIT NANDA's 2025 report observed that many of the strongest deployments began with power users who had already experimented with AI tools on their own.
How many: one per team. We suggest one for every 10 to 25 people.
What they do: a 15-minute tip session each week, a list of their team's questions and examples, first-line help, and a report of blockers to the rollout owner.
Time: about two hours a week, written into their goals so it counts as real work.
[05]
Step 2: Pilot with one or two teams (weeks 3 to 6)
Start with teams that have a willing manager and a clear starter task. Run a 60-minute kickoff on their own work, then keep the champions' weekly sessions going. Collect questions, problems and good examples every week, and fix the top issues before you expand.
Training that changes habits
Train by role and task. "Turn these call notes into a client update" sticks. A tour of features does not.
Teach a clear way to ask. A good request states the task, the context, the constraints and the format. How to write clear prompts covers this, and AI prompts for work has examples for common office tasks.
Teach the checking habit. AI states invented facts, numbers and quotes with confidence. Show people how to check AI answers and the common kinds of hallucination.
Show where it fails. In a 2023 BCG experiment with 758 consultants, people using GPT-4 performed about 40% better on a creative task within the tool's abilities, but 23% worse than the control group on a business problem-solving task outside them. Over-trust is a training problem.
Build a shared prompt library. Ten to twenty tested requests per role, owned by the champions and updated monthly.
Protect practice time. Ask managers to set aside 30 minutes a week for the first month.
Pilot exit criteria: no unresolved data incidents, the top problems fixed, at least three documented examples of real work done better or faster, and weekly use in the pilot teams at or above a target you set in advance.
[06]
Step 3: Expand team by team (weeks 7 to 10)
Expand in waves of a few teams at a time, each with its own champion and starter tasks. Use the pilot teams' examples in each kickoff; colleagues trust colleagues more than slides. Brief managers before their teams: give them the reasons for the rollout, the rules, and one request to make of each person, such as "use AI on one real task this week and show the result at our team meeting". Hold office hours twice a week, and update the rules with what the pilot taught you.
[07]
Step 4: Make it normal (weeks 11 to 13)
Write AI into the work: add AI steps to process documents, templates and new-hire onboarding.
Share results at team meetings: time saved on specific tasks, and examples that went well or badly.
Close the side doors. Once an approved tool exists, ask people to move work out of personal AI accounts.
Set a review rhythm: the rollout owner reviews rules, tools and metrics every quarter, and champions meet monthly.
[08]
How to measure AI adoption
Track a few layers, and split every number by team, role and manager. Averages hide the teams that stalled.
Layer | Metric | Where it comes from |
|---|---|---|
Access | Share of the rollout group with an approved account | The tool's admin console |
Active use | Weekly active users as a share of accounts | Admin console usage reports |
Frequency | Share using AI a few times a week or more (Gallup's threshold for frequent use) | Usage reports or a two-question pulse survey |
Depth | Share using AI in at least one core task of their role | Survey or manager check-ins |
Confidence | "I know when to trust AI output", rated 1 to 5 | Pulse survey |
Quality | Errors caught in review, rework, data incidents | Review samples and the incident log |
Shadow use | Share using personal AI tools for work | Anonymous survey |
Value | Time per starter task against the baseline | Time sampling; see how to measure the ROI of AI |
[09]
Worked example: a 120-person agency
Fernwood Marketing is an example company with 120 employees in six teams. It approved one AI assistant, named six champions and followed the four steps. These are example numbers to show what a rollout dashboard looks like:
Metric | Baseline | Day 30 | Day 60 | Day 90 |
|---|---|---|---|---|
Accounts on the approved tool | 0% | 25% (two pilot teams) | 70% | 100% |
Weekly active, as share of accounts | n/a | 80% | 62% | 74% |
Using AI in a core task | 15% | 40% | 42% | 58% |
Using personal AI tools for work | 40% | 30% | 22% | 10% |
Data incidents reported | n/a | 1 | 0 | 0 |
Two things happened that the averages alone would hide. In the first month, someone pasted a client's name and figures into a free tool. Because reporting carried no blame, the champion heard about it the same day, and the rules gained a clearer example. At day 60, weekly use dropped in the two teams whose champions had the least time. The rollout owner moved the champion role to people with more room and asked those managers for one AI task per person per week, and use recovered.
[10]
The 90-day AI adoption plan to copy
Phase | Weeks | Key actions | Owner | Done when |
|---|---|---|---|---|
Prepare | 1-2 | Approve the tool; write one page of usage rules by data class; name sponsor, owner and champions; pick starter tasks per role; measure the baseline | Rollout owner with IT and security | Tool approved, rules published, champions named, baseline recorded |
Pilot | 3-6 | 60-minute kickoff on real tasks; weekly champion sessions; start the prompt library; log issues and examples | Pilot managers and champions | Top issues fixed, no open incidents, three or more real examples, use at your target |
Expand | 7-10 | Waves of teams; brief managers first; role-based training; office hours twice a week; update rules | Rollout owner and team managers | Every team has a champion, starter tasks and tracked use |
Make it normal | 11-13 | AI steps in process documents and onboarding; share results; move work off personal accounts; set quarterly reviews | Rollout owner and sponsor | Metrics reviewed monthly; program owner confirmed for next year |
[11]
Common mistakes
Mistake | Why it fails | Fix |
|---|---|---|
Announcing access and stopping | Old habits win by default | Run the phases with champions and manager requests |
One generic training for everyone | People can't see their own task in it | Train by role on real tasks |
Banning AI without an alternative | Use moves to personal accounts, out of sight | Approve a tool and publish clear rules |
Rules nobody can find or follow | People guess, or avoid AI entirely | One page, built on your data labels, with examples |
Measuring logins only | Logging in is not doing work with AI | Track depth, quality and time per task |
Skipping managers | Manager support goes with frequent use | Brief managers first and give them one request to make |
Treating mistakes as failures | People hide problems until they grow | Blameless reporting to champions |
No time to learn | Busy people fall back to the old way | Protect 30 minutes a week for the first month |
[12]
Frequently asked questions
How do you get employees to use AI tools?
Give them an approved tool, a short set of rules, and two or three real tasks from their own job to start with. Make sure managers support it openly and ask for it, and give each team a champion who shares tips weekly. In Gallup's 2026 survey, manager support and tools that fit existing work went with much higher use.
How long does an AI rollout take?
Plan about 90 days for a company-wide rollout of a general AI assistant: two weeks to prepare, four weeks of pilot, four weeks of expansion, and three weeks to build AI into normal processes. Tools that touch sensitive data or take actions need a longer pilot. Adoption keeps growing after that if you keep measuring.
What should AI usage rules for employees cover?
Which AI tools are approved, which of your company's existing data classes each tool may handle, a ban on pasting passwords and access keys, the duty to check AI output before using it, keeping decisions about people with people, and how to report problems. Keep it to one page and ask your legal and compliance team what else applies.
What is an AI champion?
An AI champion is an employee who helps their team use AI well: sharing short tips, answering questions, collecting good examples and reporting blockers to the rollout owner. Pick respected practitioners rather than the most technical people, give them about two hours a week, and recognize the role in their goals.
How do you measure AI adoption?
Measure access, weekly active use, frequent use (a few times a week or more), depth (use in core tasks), confidence, quality and shadow use of personal tools. Split every metric by team and manager, and compare time on chosen tasks with a baseline. Logins alone do not show adoption.
Should we ban ChatGPT and other free AI tools at work?
Follow your security team's decision on which tools are allowed. A ban with no approved alternative tends to push use out of sight: in Microsoft's 2024 survey, 78% of AI users brought their own tools to work. Approving a tool, publishing clear rules, and then asking people to move their work into it usually works better.
[13]
Next steps
Choose the first teams and tasks: how to choose AI use cases.
Teach the basics of asking well: what is prompt engineering? and how to write clear prompts.
Give staff ready-made examples: AI prompts for work.
Teach checking: how to check AI answers.
Show the results: how to measure the ROI of AI.
See where your company stands: AI maturity model.