Rolling AI out across a company works when it is run as a programme with a pilot team, real workflows, a shared skills library and a handover, and fails when it is run as a licence purchase plus a lunch-and-learn. This is the 90-day playbook TAI Labs uses: discovery with the teams, one pilot team taken to working AI workflows in six weeks, a champions network that carries the method to the next teams, an internal knowledge and training system that keeps teaching after the programme, and an executive readiness report at the end. You run it once with us; your own people run it after that.
Why most rollouts stall after the first month
The pattern is the same in almost every company we speak to. A few enthusiasts race ahead with private setups nobody else can use. Everyone else opens the tool twice, gets a generic answer, and goes back to the old way. The manager cannot see who is genuinely up to speed, so the annual training cycle repeats the same introduction next year. McKinsey's 2025 survey found that the small group of high performers differ from everyone else mainly in one thing: they redesign workflows rather than bolt AI onto existing ones. A rollout that sticks is built on that finding.
- Seats are bought before workflows are chosen, so there is nothing specific to learn.
- Training is generic, so it does not change any particular task on Monday morning.
- Nothing is written down, so each person re-invents the prompt and the results vary.
- There is no baseline, so nobody can say whether it worked.
- The knowledge lives in a deck, so the next hire starts from zero.
The 90-day plan
Days 1 to 10: discovery with the teams, not about them
Short interviews with the people doing the work in each candidate team: where the hours go, what they wish they could hand over, what they will not trust a machine with. This produces a ranked list of workflows worth changing, and it is also the seed of the company knowledge system later on.
Days 5 to 15: pick one pilot team and set its baseline
Choose the team with a repeated, measurable workflow and a leader who wants it. Time three or four of its real tasks. Sales, operations, marketing, product and engineering each have a guide of their own with the typical starting workflow.
Days 10 to 20: agree the tools, the data rules and the review points
Which assistant, which systems may be connected and how, which data is off limits, and which outputs a person must review before they leave the team. Security signs this off once, for the whole rollout, instead of per team.
Days 15 to 60: take the pilot team to working workflows
Four live practitioner sessions on the team's own tools, with labs between them and two office hours for what broke. Every workflow is saved as a skill file with its steps, sources and review checklist.
Days 30 to 60: name and train the champions
One or two people per future team join the pilot's later sessions. They learn the method, not just the tools, and they own their team's first workflow after the pilot.
Days 45 to 75: build the knowledge and training system
The skills, prompt libraries, SOPs and session recordings go into one place on your learning platform, organised by role. The internal knowledge system guide covers how to make it answer questions with sources.
Days 60 to 80: the next two teams, run by champions with support
The champions run their teams' first workflow using the pilot's skills as templates, with an instructor available in office hours. This is the test of whether the method transfers.
Days 80 to 90: demo day and the executive readiness report
Each team shows working workflows against its baseline. The report scores the assessed work, names what to do next, and goes to the leadership team with the numbers rather than the adjectives.
Which team to pilot first
The pilot team decides whether the rollout gets a second act, so choose it on evidence rather than seniority. The best pilot has a workflow that repeats weekly, an output a person already reviews, a leader who volunteered, and data the company is comfortable putting in an approved tool.
| Team | First workflow | Why it works as a pilot |
|---|---|---|
| Sales | Account research to meeting brief | Weekly, measurable, a seller already reviews it |
| Operations | Closed deal to customer handover | The most expensive routine failure to fix |
| Marketing | Customer insight to campaign concept | Touches research, copy and creative in one pass |
| Product | Customer feedback to product brief | Evidence-linked decisions are visible immediately |
| Engineering | Brief to tested change with a coding agent | Lives in the repository, so it survives |
The tools that carry a rollout
A company-wide rollout needs fewer tools than people expect: one or two assistants, the systems the teams already run, a connector layer for the operations-shaped workflows, and a place where the skills and training live.
ClaudeThe assistant for long documents, briefs and skills
ChatGPTResearch, drafting and everyday questions
Microsoft CopilotFor teams that live in Microsoft 365
NotionWhere the skills, SOPs and prompt libraries live
SlackWhere champions answer questions in the flow of work
n8nConnect the operations workflows
Custom MCPsApproved, scoped connections to company data- Claude CodeEncode a process as a reusable skill
The champions network
Colleagues learn AI from colleagues. A champion is not the most enthusiastic person; it is the person in each team who owns that team's first workflow, keeps the skill file current, and answers the question in Slack before it becomes a ticket. Champions join the pilot's later sessions, get office hours with an instructor for a quarter, and meet monthly to compare what worked. The network is what makes the rollout run without a vendor after day 90.
- One or two champions per team, named before day 30
- Each champion owns one workflow and its skill file
- A monthly champions call, with the exception logs and the wins
- Office hours with an instructor for the first quarter after the programme
- A path for a champion to become the next intake's instructor
Do it with us once
We run the 90 days with you, once.
Discovery, the pilot team, the champions, the knowledge and training system, and the readiness report. Bring one team and one workflow to a 15-minute call and we will scope the rollout.
The policy people actually follow
Leapsome's 2026 workforce report found that 76% of HR leaders believe their company has a clear AI policy while only 48% of individual contributors agree. The policy that closes that gap is one page, written with the pilot team during the programme, and it says what may go into which tool, what must be reviewed by a person, and who to ask. It is handed to every new joiner with the first workflow, not sent as a PDF from legal.
- Approved tools and the data each may receive, by category
- Outputs that require a named person's review before use
- How sources are cited so a wrong fact can be traced
- What is never automated without a separate scope: sending, paying, changing customer records
- Who answers questions: the champion first, then the owner of the policy
What the company keeps at day 90
- Working workflows in three teams, each with a skill file and a review checklist
- A skills and prompt library on your learning platform, organised by role
- The champions network with its monthly rhythm
- The one-page policy, written with the teams
- Baselines and week-12 comparisons per team
- The programme itself, ready to run for the next teams without us
- The executive readiness report
How to know it worked
DataCamp's 2026 survey of enterprise leaders found organisations with a mature, organisation-wide AI programme report significant ROI twice as often as the rest (42% against 21%). The tools are the same in both groups. What differs is that the programme had a baseline, a measure per team and a report. The ROI guide sets out the measurement in full; the short version is: per team, one time measure, one quality measure and one adoption measure, compared like for like at day 90.
Frequently asked questions
How many teams can a 90-day rollout cover?
One pilot team to working workflows, then two more run by champions with support. Three teams in a quarter is the realistic pace; the champions network is what scales it after that.
Should we start with the most enthusiastic team?
Start with the team whose workflow repeats and whose leader volunteered. Enthusiasm helps; a measurable weekly task and a willing leader matter more.
Do we need to buy a platform first?
No. Most companies run the pilot on the assistants and systems they already have. The learning platform matters at day 45, when the skills and training need a home; we can hand over onto yours or provide one.
What if security will not approve connections?
Then the first workflows run on exports and pasted context, which is enough for the pilot. Connections are added once security has seen a reviewed workflow run for a cycle.
What does it cost?
It depends on the number of teams and the format. Indicative ranges for a 20-person cohort are on the custom AI training page, and the cost guide explains what moves the number.
What happens after day 90?
Your champions run the next teams with the programme on your platform. Office hours with an instructor continue for a quarter. You do not need us again unless you want a new capability taught.
Interview the teams, pilot one of them to working workflows in six weeks, name champions, put the skills and training in one place, let the champions run the next two teams, and end with a report that has numbers in it. That is how TAI Labs rolls AI out across a company in 90 days, once.