A half day for leaders who have told their teams to use AI and now have to stand behind it. Six hands-on labs in three live sessions, and each lab ends with something you keep.
AI-generated image. Every company and person in the labs is fictional.
Where it sits
Each NIA Labs level runs 75 to 90 minutes and stands alone. The Executive AI Lab joins them up for the people who sign off on AI: the same Kellmoor world and the same rule that laptops stay open, with checking and governance built into every lab instead of saved for the end.
What the tools are doing, what they cost, and which one fits your work, tested on a real task.
Ground every claim in a source, and take any plan to replace a team apart task by task.
One page, five decisions, and a rehearsal for the afternoon it goes wrong in public.
Three sessions, two labs each
About 55 to 60 minutes a session. Run them on three days, or as one sitting of about three hours with a break.
Session 1
What is it doing, what does it cost, and which tool fits my work?
You keep: an AI cheat sheet and a toolkit decision card.
Session 2
Is this output right, and what must a person still own?
You keep: team AI quality rules and a replacement reality check.
Session 3
Which AI do we use, who owns it, and what would an auditor see?
You keep: a one-page inventory with a gap list, and an evidence pack checklist.
Six labs, six artefacts
1Daniel Koh, Managing Director
“I told everyone to use AI. Then Aisha asked how it works and what it costs, and I had nothing.”
How a model predicts text, what a heavy day of use costs in tokens, and the four kinds of confident error. Then you ask an assistant about your own organisation and check every claim.
Case file Mata v. Avianca, New York, 2023
You keepMy AI cheat sheet
2Sofia Tan, Head of Sales
“Mine is the best. I use it every day.”
One real task, the same prompt in two to four assistants, names hidden, scored on four criteria: clear, specific, honest about gaps, checkable. Then one connection to your calendar, read only.
Case file Shared chats that turned up in search, 2025
You keepA toolkit decision card, with a re-test date
3Marcus Lee, Head of People
“I got a twelve-page update at nine. By ten I had found three mistakes.”
Mark an AI-drafted update for what is invented, what is ungrounded and what is padding. Re-ask with the source and demand the page. Then write the rules your team will actually follow.
Case file “Workslop”, Harvard Business Review, 2025
You keepOur team AI quality rules
4Ravi Menon, COO
“Twelve people, one year. Everyone says AI can do this now.”
Break a replace-the-team proposal into tasks. Score each for frequency, cost of error and exceptions. Set a level of control, from agent alone to human only, and name the person who owns it.
Case file Klarna, 2024 to 2025
You keepA replacement reality check
5Helen Ward, General Counsel
“I cannot govern what I cannot list.”
One row for every use of AI in your function, then five decisions for the top one: scope, owners, data, human approval and evidence. Pointers to PDPC, IMDA and MAS guidance, not legal advice.
Case file Italy's Garante fines OpenAI, 2024
You keepA one-page inventory and gap list
6The Kellmoor leadership team
A customer chatbot makes a promise in public that policy does not allow.
A tabletop. As the afternoon unfolds, tables decide who pauses it, who speaks and what to preserve. Then they assemble the evidence an auditor would ask for, and find the gaps.
Case file The DPD chatbot, United Kingdom, 2024
You keepAn evidence pack checklist
+Capstone
One action a week drawn from the six labs, with a date next to each, and a five-point governance checklist for your own function. Most people tick one or two points on the day. The plan is how you get to five.
You keepA 30-day plan and a governance checklist, with a 30-day check-in offered
How each lab runs
Once the room knows the rhythm, nobody waits for instructions. The tabletop in Lab 6 runs a little longer.
Fictional or public. Nothing confidential goes into any tool, so every exercise is safe on any assistant.
Word for word, in a fresh chat. Change the wording and you are comparing your prompts, not the tools.
A citation is a claim until you have opened it. Asking the same tool to confirm is not checking.
Watch a lab open
Under forty seconds, then laptops open. Captions are on by default.
Session 1 · Lab 1
Daniel told everyone to use AI. Then Aisha asked how it works and what it costs. In this lab you catch it producing text that only sounds right.
Session 1 · Lab 2
Sofia swears by one assistant, Ravi by another. Four assistants, one real task, and the names stay hidden until the scores are in.
Session 2 · Lab 3
Twelve pages, polished, wrong in three places. When AI makes writing cheap, the reader pays the cost.
Session 2 · Lab 4
Replace the support team within a year? Break the job into tasks, score the risk of each, and decide what a person must still own.
Session 3 · Lab 5
Which AI tools are we using, where, and who owns them? Nobody could say. One page fixes that.
Session 3 · Lab 6
A customer chatbot gets it wrong in public. Who pauses it, who speaks, and what can you prove? A short cut of the opener.
Narration and graphics are AI-generated for NIA Labs.
Case files
One per lab. The lesson is the control that would have helped, not the headline.
Lawyers filed cases a chatbot had invented, after asking the chatbot whether they were real.
Conversations people shared by link turned up in search results once a discoverable option was ticked.
Harvard Business Review on polished AI output that moves no task forward, and who pays to fix it.
An assistant said to do the work of hundreds of agents, then a return to hiring people for customers.
A EUR 15 million privacy fine. The regulator asked about data, legal basis and evidence.
A delivery chatbot swore at a customer and was switched off after an update.
Format
There is no recorded version: Manish runs the room with table helpers, in person or virtual. If you want to test the method first, start with a free lab or a single level.
Try it
15 min
The Kellmoor inbox labs and the Mark the draft taster run in your browser, with no sign-up.
Entry session
75–90 min
Foundations, Real Work or Build and Govern. A private session that stands alone.
Flagship
Half day
Six labs in three sessions of about an hour, or one sitting of about three hours. Six artefacts and a 30-day plan.
Free, in your browser
Short public exercises in the style of the labs. No sign-up.
An AI-drafted pilot update for the Kellmoor leadership team. Find the invented figure, the invented quote and the padding before it goes out.
Open the draft → Practise · about 3 minutesWhat does it cost?Tokens in, tokens out, times the price. Work out what one task and a busy year cost on illustrative prices, and what the review time adds.
Open the calculator →From participants
Your facilitator
Manish advises boards and senior executives on cybersecurity, AI adoption and technology risk. He brings more than 25 years of international experience across financial services, insurance, telecommunications, critical infrastructure and government.
A former partner at PwC and Grant Thornton, he holds Oxford AI Strategy credentials and runs every session himself. The method started with four senior executives and twenty minutes each. All four became daily users.
Sample workbook
Four worksheets from the participant workbook: the AI cheat sheet, the one-page inventory with the five decisions, the 30-day plan and the governance checklist. Leave your work email and the download opens straight away.
