NOTE: This is an installable Claude skill, not a document to fill in. Download it above, install it in Claude, and ask it to help you plan your AI Week. The playbook below is what the skill works through with you.

Planning a company-wide AI Week

An AI Week takes every team in the organisation out of normal work for one full day to learn and build with AI. Each team takes its day on a different date, spread over two to four weeks. The morning is structured learning and getting tools connected. The afternoon is building solutions for that team’s real work. The last 30 minutes is a show and tell.

We built this skill from running AI Weeks in enterprise organisations, across multiple offices and timezones. What follows is the playbook the skill works through with you, including the things we would do differently.

The nine stages

1. Sponsorship

Strong, visible leadership is the single largest determinant of whether an AI Week lands. To go faster, sometimes you need to slow down, and only the most senior executive can credibly say “this is important, I am taking a day out, all of our leaders are taking the day out, and you should too.”

Test sponsorship against three questions:

1. Who says it? A CEO or COO announcing it reads as a company priority. A Head of IT or L&D announcing it reads as optional training.
2. Are leaders attending their team’s full day? A leader who drops in for the first hour tells the room this is not important.
3. Is the day protected? Meetings cancelled, cover arranged, out-of-office on.

If any of those is soft, fix the sponsorship before fixing the agenda. An AI Week with weak sponsorship produces a good day and no change.

2. Scope and calendar shape

Work out cohorts, sites and sequencing:

• Cohort size. 10 to 20 people per room. Above 25, the afternoon build stops being supportable.
• Sites. Run one per office rather than one central event. Being in the room matters more than getting the whole company together.
• Calendar. One team per day, two or three teams per week, over two to four weeks. Facilitators improve between runs and the business stays open.
• Sequencing. Start with a team that will succeed, ideally one with an enthusiastic leader and obviously AI-shaped work. Their show and tell becomes the promotional material for everyone who follows.

Group by the work, not by AI experience. The temptation is to split people into beginner and advanced streams. Grouping by team works better, because the discussion in the room is then about the work that team does.

The refinement we would make: experienced users find the opening modules basic, and a few will say so. Handle that inside the team day rather than by splitting the cohort. Compress the framing for teams you know are ahead, give the afternoon an advanced brief such as building an agent rather than a skill, and use the already-fluent people as informal helpers in the room.

3. Customisation per team

Run a 30 to 45 minute briefing with the senior leader of every team in scope, across sales, marketing, engineering, legal, finance and talent. Get:

• What that leader is focused on this quarter, their OKRs and their KPIs.
• The three to five processes their team spends the most time on.
• The systems the team works in daily, so connectors can be pre-approved.
• Real, non-confidential sample material: a document, a dataset, a ticket.

This is where AI Weeks lose the most credibility. The most common criticism from participants is that examples were generic. A worked example about a hotel is a fine teaching device, but a finance team wants financial data extraction, and a data team wants tables and datasets, not paragraphs of prose. Generic examples lead people to conclude the tool does not fit their work.

Add one live demonstration per team of a use case relevant to that function, run before the build session starts. Seeing a relevant workflow built end to end is what unlocks people’s own ideas.

4. Readiness and logistics

Nothing wastes a morning faster than provisioning.

Access: licences activated and verified per named person, not per seat count. Desktop app installed in advance. Connectors and integrations approved by IT and security beforehand. Admin approvals for extensions, plugins and MCP servers cleared. A named IT contact reachable on the day.

Room and facilitation: one facilitator per 10 to 12 people. You cannot beat sitting next to someone, seeing where they are stuck, and unblocking them, and this matters most for senior leaders who will not ask for help in front of their team. Have both facilitators in the room; a remote second facilitator is noticeably worse and participants notice. Power at every seat, wifi tested at full room load, food on site.

Pre-work: a six-question survey a week ahead covering the processes each person owns, the hours those take, the documents they use, the one task they would most like to hand over, how often they use AI today, and their confidence on a 1 to 5 scale. The last two questions are the baseline for every capability claim you make later.

5. Communications

Two rules. The “why” comes from the sponsor in their own voice, not from IT. And every message after day one carries somebody’s real, named example, because peer proof travels and corporate encouragement does not.

6. The build session and show and tell

The afternoon is where the week earns its cost. Give people time to build for the tasks in front of them: a prompt, a pre-built skill, their own skill, or a first live artifact connecting more than one system.

• Build time. Two to three hours. One team will find three hours too long and the next will want more, so set it per team and give facilitators a mid-session checkpoint.
• Capture as they build. Log what each thing does, who owns it, what it replaces, and what production-ready would need. Without this you have a good day and no pipeline.
• Show and tell, 30 minutes, every day, without exception. Seeing what a colleague built, and how they brought it to life, is what gives other people the confidence that they can too. Record it with permission and use it in the comms for the teams that follow.

7. Metrics and ROI

Measure at four points: a baseline before, activity on the day, a survey in week one, and platform and process measures at 30, 60 and 90 days.

Be honest about which numbers are which. Confidence uplift, attendance and self-reported hours saved are leading indicators. They tell you the week worked and they are the right thing to report to the business. They are not a P&L number.

For the business case, pick two or three processes during planning, baseline them on volume, cycle time and rework rate before the week, and re-measure at 60 and 90 days. Two processes measured properly beat a hundred self-reported estimates. Include the cost of participant time in the cost side; it is usually the largest line, and a sponsor who finds it missing stops trusting the rest.

8. What happens next

Plan this before you start, so you can promote it during the week itself.

• Focused workshops on the highest-impact processes, chosen with senior leaders during the briefings and named during the AI Week.
• Regular drop-in sessions where anyone brings an early idea and gets help. Low ceremony, recurring, in the calendar before the week starts.
• A backlog where ideas land, scored so the good ones get picked up. Whoever owns it must visibly respond in the first fortnight, or it dies.

9. The executive readout

Book the slot before the week starts. Structure it in seven parts: what we did and why in the sponsor’s words; coverage as a percentage per team rather than a headcount; capability shift, confidence before and after; what got built, the count plus three named examples with named owners; the pipeline, scored, with the top five going forward; value, with leading indicators labelled as such and the instrumented processes named; and what you need next in decisions, budget and sponsor actions.

Implementation notes

Adapt it. The skill is plain Markdown. Change the stages, add your own governance requirements, swap in your own cost assumptions, and re-upload it as your internal version. Organisations running a second or third AI Week normally do.

Start the sponsorship conversation first. Stages 2 through 9 are all recoverable. Stage 1 is not. If your sponsor will not take the day themselves, that is the conversation to have before you build a calendar.

Do not skip the baseline. One survey, six questions, a week before the first team’s day. It costs almost nothing and it is the only thing standing between you and having no capability measure at all when someone asks whether the week worked.