Skip to content
Leverage AI

AI at Work · Team workshop kit

Your staff already use AI. Run one day that teaches them to use it well, and safely.

Everything one person in your company needs to run a full-day AI workshop for 8 to 25 colleagues. No outside trainer. No AI expertise needed. It is written to be used with ChatGPT, Claude, Gemini or Microsoft Copilot.

Founding price for the first 10 organisations (10 left). After that: $299 for up to 50 staff, $499 for up to 200. One payment, instant access, invoice included. Full refund within 30 days if it is not useful.

The problem this kit solves

Most teams use AI at a basic level: a question here, an email there. The studies show real gains on single tasks (14% to 40% faster), but small gains across a whole week, because people move only a few tasks to AI and nobody agreed what data is safe to paste. A trainer-led day costs thousands. A video course gets watched by a few people. This kit gives you the day itself, ready to run.

What happens on the day

9:00 to 15:00, about 4 hours 50 minutes of work, two breaks and lunch. Mostly hands-on.

  • 9:00 · Ready check and opening

    Everyone logs in and tests the tool. Four plain facts about how AI works, and where it fails.

  • Pick one real task

    Each person picks one task from their own week. They use it in every block that follows, so they leave with a result they will use on Monday.

  • Write a good brief

    The Brief card: task, audience, context, examples, format, checks. People edit a worked prompt instead of starting from a blank page.

  • Rules for our AI, and make it safe

    The team writes its own rules with green, amber and red data lanes. Then a timed drill: clean 3 fake tickets of personal data in 2 minutes each.

  • Build a skill

    Save instructions and examples as a reusable skill, such as replies in your own voice. A colleague tests it on new data.

  • Writer and checker

    One prompt drafts, a second prompt checks the draft against a list. The rule that catches the dangerous mistake: does a wrong answer look right here?

  • Commit to 30 days

    Each person sets one daily habit. The team names a champion and a weekly 15-minute slot to show one AI win.

What is in the kit

  • Start guide and checklist

    What to prepare 2 weeks, 1 week and 1 day before. The materials list.

  • Pre-work

    How to choose a business AI plan that does not train on your data, the pre-survey, and the data-lanes page your manager signs before the day.

  • Facilitator script, minute by minute

    What to say, what people do, what to watch for, and what to cut if you run late. Includes a dry run for the champion before the day.

  • Slides outline and workbook

    About 25 slides to copy into your slide tool, and the participant workbook in the order of the day.

  • 30 ready prompts for 6 jobs

    Support, sales, operations, finance and admin, HR, and managers. Each with when to use it, its data lane, and how to check the output.

  • Data-safety drill with answer key

    Fake tickets and emails to de-identify, with a timer method. One handout per person, the key for the champion only.

  • Practice inputs by job

    Fake emails, notes and requests for sales, finance and admin, operations, HR and managers, so everyone practises on work that looks like theirs.

  • Objection cheatsheet

    Short, honest answers to what staff will say: AI will take my job, IT banned it, it makes things up, I have no time.

  • Policy pack

    Do's and don'ts, an acknowledgement form, a what-not-to-paste list by industry, and an approved-tools register.

  • Launch pack

    The kickoff email, the invitation, a flyer, and the day-after follow-up.

  • 30-day plan and the boss report

    Confidence scores before and after, a day-30 check, and a one-page result for the sponsor. Time saved counts only when the task owner confirms it.

How it compares with the free vendor kits

Microsoft, OpenAI, Google and Anthropic publish good free material, and you should use it. Each one is built around its own tool, and none is a full-day script for a non-expert with data rules that work in any tool. This kit fills that gap, and it links to their free resources where they help.

Licence and updates

For use inside one organisation. Free updates when the tools change. The founding price covers any organisation up to 200 staff. After the founding round: Team (up to 50 staff) $299, Company (up to 200 staff) $499.

Read a few pages first

Real pages from the kit, free. Open one.

Brief card

Free sample from the AI at Work team kit.

Brief card

Print the card on one page. Keep it on your desk. Use it every time the task matters. For a quick question, you do not need all seven parts. For any work you send to another person, use all seven.


The card

THE BRIEF

1. TASK       What do you want? Start with a verb.
              "Draft a reply...", "Summarise...", "Turn this into a table..."

2. AUDIENCE   Who reads the result? What do they know? What do they need?
              "A small-shop owner who is not technical and is annoyed."

3. CONTEXT    What does the AI not know? Our facts, policy, situation, goal.
              "Our policy: we refund setup fees within 14 days. Not after."

4. EXAMPLES   One good example of what you want, if you have one.
              "Here is a reply I wrote last month that worked well: ..."

5. FORMAT     Length, structure, tone, language.
              "Under 120 words. Three short paragraphs. Friendly, plain."

6. CHECKS     What must be true? What must it never do?
              "Do not promise a date. Do not mention discounts.
               List any fact you were not sure about at the end."

7. ASK ME     "Before you start, ask me up to 5 questions about
   FIRST       anything that is unclear. Then wait for my answers."

LANE CHECK    Green? Go. Amber? Make it safe first, approved tool only.
              Red? Stop. Do not paste.

A fill-in version (copy and paste)

Task: 
Audience: 
Context: 
Example of what I want (optional): 
Format: 
Checks (must be true / must never do): 

Before you start, ask me up to 5 questions about anything that is unclear.
Then wait for my answers.

Why each part matters

PartWhat goes wrong without it
TaskThe AI does something close to what you wanted, but not the same thing.
AudienceThe tone and the detail are wrong: too technical, too long, too formal.
ContextThe AI fills the gaps with guesses. Guesses about policy and facts sound true.
ExamplesThe AI writes in a generic style that does not sound like you or your company.
FormatYou get 600 words when you needed 100, or a list when you needed an email.
ChecksThe AI promises things you cannot deliver, or hides what it was unsure about.
Ask me firstYou answer the questions you did not know you needed to answer. This line often improves a result more than the other six.

Three habits that make every brief better

  1. Give it the facts, then ask it to use only those facts. Add: "Use only the facts above. If something is missing, say so. Do not invent it."
  2. Ask for the doubts. Add: "At the end, list anything you were not sure about." You then know where to check first.
  3. Improve the brief, not only the answer. If the result is wrong, ask yourself which part of the brief was missing. Fix that part. Next time, the first answer is better.

Worked example 1: a support reply (Northwind Services)

The situation. A Northwind customer's card reader was due to arrive on Monday. It is now Thursday. The customer is annoyed. The courier says the parcel arrives tomorrow. Northwind's policy: no refunds for courier delays, but a goodwill credit of one free month of the software is possible for delays over 3 working days.

The customer's email (already made safe):

Subject: Where is my card reader??

I was promised the card reader on Monday. It's Thursday. I have a market
stall on Saturday and I can't take cards without it. This is the second
time something from you has been late. What is going on?

[CUSTOMER-1], [BUSINESS-NAME]

The weak brief

Write a reply to this customer.

What you usually get: a long, formal apology. It may promise a delivery date it cannot know, offer a refund Northwind does not give, or invent a tracking number. It does not sound like Northwind.

The full brief

Task: Draft a reply to the customer email below.

Audience: A small-business owner who sells at a weekend market. Not
technical. Annoyed, and worried about Saturday.

Context:
- The card reader was due Monday. The courier now says it arrives Friday
  (tomorrow), before 18:00. We cannot guarantee the courier's time.
- Our policy: no refunds for courier delays. For delays over 3 working
  days we may give one free month of the software as goodwill. This delay
  is 4 working days, so the free month applies.
- If the reader does not arrive by Friday 18:00, the customer can take
  card payments on Saturday with our free phone app ("Northwind Tap").
  Setup takes about 10 minutes. The guide is at
  help.northwind.example/tap.
- The customer says this is the second late delivery. I cannot confirm
  that. Do not argue with it, and do not confirm it.

Example of our tone: "Thanks for telling us, and sorry for the stress.
Here is what we are doing about it."

Format: Under 150 words. Short paragraphs. Friendly and plain. No
"we apologise for any inconvenience". Sign off as "[MY NAME], Northwind
Support".

Checks:
- Do not promise the parcel will arrive. Say what the courier told us.
- Do not offer a refund.
- Mention the free month and the phone app as a backup plan.
- At the end, under a line, list anything you were not sure about.

Before you start, ask me up to 5 questions about anything that is unclear.
Then wait for my answers.

Typical questions the AI asks: "Should I tell the customer how to claim the free month, or is it added automatically?" "Do you want to give a direct phone number for Saturday?" "Is the free month added to the current month or the next one?" These are exactly the details a customer then asks about. You answer them once, and the reply is complete.

How to check the result (1 minute): Is the delivery described as the courier's estimate, not a promise? No refund? Free month and phone app both mentioned? Under 150 words? Does it sound like us? Read the "not sure" list first.

Make-it-safe drill

Free sample from the AI at Work team kit.

Make-it-safe drill

Goal: every person can make a ticket or an email safe for the amber lane in about 2 minutes. On the day, this drill runs from 11:35 to 11:50. At 11:58, after the rules are saved, the lead champion announces that amber data is allowed in the approved business-plan tool, if the lanes are signed. Before the day, everyone uses Section 1 to make their past outputs safe.

All names, businesses, numbers and addresses below are fake. Phone numbers use the 555-01xx range, which is reserved for fiction. Emails use the reserved example domains.


Section 1: The 2-minute method: Scan, Swap, Test

Put a visible timer on the screen. Each item gets 2 minutes.

Scan (30 seconds)

Read the text once. Underline or highlight everything that could point to a real person or customer:

  • Names: people, businesses, and names hidden inside email addresses and signatures.
  • Contact details: email, phone, address, website, social media handle.
  • Numbers: account, order, invoice, ticket, ID, card, bank, licence plate.
  • Unique details: "the only shop at the station", "our biggest customer in the north", a job title that only one person has.
  • Sensitive details: health, family, money trouble, legal problems, complaints about a named person.
  • Confidential business details: prices or discounts marked confidential, contract terms, deal sizes.

Swap (60 seconds)

Replace each item with a label in square brackets. Use the same label for the same thing each time. Keep what the AI needs for the task.

Replace......with
A person's name[PERSON-1], [PERSON-2] (or a role: [CUSTOMER], [OUR AGENT])
A business name[BUSINESS-1]
Email, phone, address[EMAIL], [PHONE], [ADDRESS]
Account, order, invoice numbers[ACCOUNT], [ORDER], [INVOICE]
An exact date, if the date does not matter"last Monday", "3 days ago"
A unique detailA general one: "a shop at a transport hub", "a mid-size customer"
A sensitive detail the task does not needDelete it.
Card numbers, passwords, bank detailsStop. The text is red. Delete them, and use only what is left, if it is still useful.

Keep the facts the AI needs: what went wrong, what the customer asks, the product name, the error message, how long the delay is.

Test (30 seconds)

Ask the four questions of the confidentiality test:

  1. Could a reader work out which customer or person this is? → must be No.
  2. Is anything in it covered by a contract or NDA, or marked confidential internally? → must be No.
  3. Would we be comfortable if this exact text appeared on a public website with our name on it? → must be Yes.
  4. Is this the minimum text the AI needs? → must be Yes.

All four pass? It is amber. You may paste it into the approved business-plan tool (never a personal account, paid or not). Any fail? Remove more, or treat it as red.

Tips that save time

  • Copy first, then clean. Paste the text into a document in your work account, never a personal app. Clean it there, and delete the document when you finish. Never type into the original email.
  • Delete signatures at once. Most contact details hide in the signature and the forwarded headers ("From:", "To:", "Sent:").
  • Cut before you swap. If the task is "summarise the problem", delete the greetings, the history and the signature first. Less text means less to check.
  • Use your tool's search. Search the cleaned text for "@", for digits, and for capital letters in the middle of a sentence (names). This takes 10 seconds and catches most misses.

Item 2: Chat transcript (setup problem)

[10:42] Customer (Dev Patel, Pellowick Flowers): Hi, the booking page
        stopped showing Saturdays. Order #A-77120.
[10:43] Agent (Priya): Hi Dev, sorry about that. Can you check
        Settings > Opening hours?
[10:46] Customer: It says Saturday 9-17 but the page still hides it.
        We're the only florist in the Harbour Point ferry terminal, so
        Saturday is our busiest day. Sorry for slow replies, my wife had
        surgery last week and I'm doing everything alone.
[10:48] Agent (Priya): No problem at all. I'll escalate to our tech
        team.

Task you want help with: summarise the problem for the tech team, with the steps already tried.


Item 2: answer

[Customer]: The booking page stopped showing Saturdays. [ORDER]
[Our agent]: Asked the customer to check Settings > Opening hours.
[Customer]: Opening hours show Saturday 9-17, but the booking page still
            hides Saturday. Saturday is their busiest day.
[Our agent]: Escalating to the tech team.

Removed: names, business name, order number, the unique location ("the only florist in the Harbour Point ferry terminal"), and the health detail about the customer's wife. The task does not need it, and it is sensitive information about a third person. Kept: the setting already checked, what the page does, why it is urgent. Common misses: the unique location. It identifies the business as surely as its name. Many people also keep the surgery detail because it "explains the slow replies". The tech team does not need it. Lane after cleaning: amber.

The "My computer week" sheet

Free sample from the AI at Work team kit.

The "My computer week" sheet

Send: 2 weeks before. Deadline: 3 days before. Time: 15 minutes. Bring it on the day. It is the starting point for the Brief (Stage 2) and the skill (Stage 4).

This exercise finds the task where AI helps you most. Research shows that AI gains are large on some single tasks and small across a whole week. The difference is the choice of task. A good choice is worth more than any prompt trick. The task you choose here becomes your spine task: on the day you write a brief for it, make its data safe, build a skill for it, and write a checker for it.

Instructions for participants

Step 1. List (4 minutes). Open your email "Sent" folder, your chat app and your calendar for last week. Write up to 8 tasks you did at the computer. Start each task with a verb: "reply to...", "write...", "copy... into...", "check... against...", "summarise...".

Step 2. Score (5 minutes). Fill in each column. Best guesses are fine.

  • Times per week: how often you do it.
  • Minutes each time: how long it takes, on average.
  • Digital in and out? Is the input already on the computer (email, document, sheet), and is the output something you type or paste? Y or N.
  • Cost of an error: L (low: easy to fix, nobody hurt), M (medium: embarrassing or costs time), H (high: costs money, a customer, or breaks a rule).
  • Easy to check? Can you check an AI answer much faster than you can do the task yourself? Y or N.
  • Lane: G (green), A (amber), R (red). If you are not sure, write "?".

Step 3. Filter (1 minute). Cross out any row that is:

  • R (red), or
  • N for "digital in and out", or
  • H for "cost of an error" and N for "easy to check".

Step 4. Rank (2 minutes). For each row you did not cross out, calculate: Minutes per week = times per week × minutes each time. Circle the row with the highest number. That is your spine task: you use it in every block of the day. Also mark your second choice.

Step 5. Check one thing (3 minutes). Ask yourself the five questions of a good AI task:

  1. Is it digital in and digital out?
  2. Do I do it at least weekly?
  3. Does it take at least 10 minutes each time?
  4. Can I check the result in much less time than I need to do the task?
  5. Is it green or amber?

If your lead champion tells you that our day runs in the green lane only, Q5 must be "green". Choose a green task, or the green part of a task (for example: the reply template, not the customer email).

If you answer "no" to any of the five, use your second choice. On the day, your partner asks you the same questions.

Everyone (10 minutes more): bring 2 to 5 good past examples of your spine task's output. For example: emails you sent, summaries or notes you wrote, a table you made. Remove names, numbers and contact details (replace them with [CUSTOMER], [ORDER], [PHONE]). Keep them in a document in your work account, not in the AI. You use them as the examples in your skill at 13:05. You check them again after the drill at 11:35.

  • Support staff: bring 5 replies in different situations (a how-to, a complaint, a delay, a "no", a thank-you). You use them to build the "replies in my voice" skill.
  • No safe examples, or a green-only task with no past output? Use the example past outputs for your function in the full kit. Tell your partner on the day.

Do not write real customer or colleague names on the sheet. Write "reply to a customer about a late order", not "reply to Ms Lee at Quillfern Bakery".

The sheet

#Task I did on the computer last weekTimes per weekMinutes each timeDigital in and out? (Y/N)Cost of an error (L/M/H)Easy to check? (Y/N)Lane (G/A/R)Minutes per week
1
2
3
4
5
6
7
8

My spine task for the day: row # ____ Minutes per week today (my baseline): ____ My second choice: row # ____

Worked example (Northwind Services, a support team member)

#TaskTimes/wkMin eachDigital?Error costEasy to check?LaneMin/wk
1Reply to "how do I..." questions from customers1510YMYA150
2Write the weekly ticket summary for my manager145YLYG (totals only)45
3Phone a customer back about a failed setup615NMNA90 crossed out
4Turn a solved ticket into a help article140YMYA in, G out40
5Update customer bank details in billing35YHNR15 crossed out
6Summarise a long ticket thread to hand to a colleague415YMYA60

Choice: row 1, 150 minutes per week. Second choice: row 6. Row 3 is a phone call, so it is not digital in and out. Row 5 is red, high cost and hard to check, so AI is not used for it.

Run your first AI day this month

Instant online access after payment, with an invoice for your accounts. If the kit is not useful, ask through the contact page within 30 days and you get a full refund.

Get the kit for $149

Questions people ask

Do we need an AI expert to run the day?

No. The script is written for a staff member who uses AI at a basic level. It tells them what to say, what to do and what to cut if time runs short.

Which AI tool does it work with?

It is written to be tool-neutral: the prompts and exercises use plain text that you paste into ChatGPT, Claude, Gemini or Microsoft Copilot. Your champion tests them in your chosen tool during the dry run. The pre-work helps you choose a business plan that does not train on your data.

Is our customer data safe during the workshop?

The day uses fake data only. Before the day, your manager signs the green, amber and red data lanes, and the drill teaches staff to remove personal data before they use AI.

How many people can join?

The day is designed for 8 to 25 people. Larger teams run it more than once with the same kit.

How do we get the files?

After payment you go straight to your private kit page. Bookmark it. You can print any part or save it as a PDF.

Can we get a refund?

Yes. Write to us through the contact page within 30 days, with your receipt number, and you get a full refund.