AI training for executives, built on your own work
This AI training for executives runs on your own work, such as the bank update or a price review.
What executive AI programs teach
Most results for AI training for executives are business school programs and lists of courses. Columbia Business School’s The Business of AI runs in person over four days, in New York or Paris. Columbia says participants learn the terminology and technical capabilities of generative AI, and build a toolkit for putting AI in place across their organization.
MIT Sloan’s Artificial Intelligence: Implications for Business Strategy is a self-paced online course that runs six weeks, at six to eight hours a week. MIT says the course covers the organizational and managerial side of AI rather than the technical side, with a business-focused project of your own.
Programs like these teach strategy to leaders who plan AI across a large organization. The owner of a plant or a distribution business often has a narrower question. How does this month’s bank update, or the next price review, get done with AI on the company’s own files?
ThriveAI’s training starts from that question. You bring a real task and the documents behind the task, and you do the task with AI.
Jobs an owner or executive can hand to AI
Each of these jobs starts from documents the company already keeps. The assistant writes the draft, and you check the draft against the source before anything leaves the company.
| Job | What you give the assistant | What you check |
|---|---|---|
| A board or bank update | Last quarter’s update, this quarter’s statements and the sales export | Every figure against the statement it came from |
| A pricing decision | A year of quotes and orders from the ERP, with cost and price on each line | The rows behind each margin figure |
| A contract | The whole contract as a PDF, and the clauses you care about | Each clause at the section the assistant cites |
| A customer meeting | The customer’s orders, open quotes, recent emails and CRM notes | Dates and amounts against the ERP |
| A policy | Your rules as rough notes, and any policy you already have | That each rule matches how the plant works |
A board or bank update
Give the assistant last quarter’s update as the model, with this quarter’s statements and the sales export. Ask for the same sections with the new figures, plus a list of every figure that moved by more than 10 percent.
Check each figure against the statement it came from before the update goes to the bank or the board. Save the instruction that worked, so next quarter’s update starts from the same instruction.
A pricing decision from an ERP export
Export a year of quotes and orders with the part, the customer, the cost and the price on each line. Ask which parts or customers sit below the margin you name, and ask for the rows behind each figure.
Claude accepts an Excel upload only when code execution and file creation are turned on, according to Anthropic’s help page on uploads. Check the margin math on a few rows by hand, then decide the price change yourself. For how AI reads the ERP you already run, see AI for ERP.
Reading a contract
Upload the whole contract rather than an excerpt. Ask for the clauses on price changes, late delivery penalties, liability, payment terms and termination, with the section number of each.
Read each clause at the section the assistant cites. An assistant writes a confident summary whether or not the contract says so. A lawyer still reads any contract that carries real risk.
Preparing a customer meeting
Give the assistant the customer’s orders and open quotes from the ERP, the recent emails and the notes in the CRM. Ask for one page that covers what the customer bought this year, what is late, what is still open and what to ask.
Check dates and amounts against the ERP before you repeat them to the customer. For an assistant that prepares every meeting from the calendar and the CRM, see AI chief of staff.
Writing a policy
Write the rules as rough notes first. Say which AI accounts staff may use, which files may go into those accounts and who approves a draft before it goes out.
Ask the assistant to turn the notes into a one-page policy, in French and English if the plant works in both. Check that each rule matches how the plant works, then have the people the policy affects read the draft. For what a policy covers, see AI policy for your company.
Start from the job that takes the most hours
Tell Derik which job takes the most hours in your week and who would attend. He will tell you what a session built on that job would cover.
Start a conversationWhat happens in a working session
A working session runs on your own tasks and documents, on site at your plant or office. Sessions run in French or English, and Derik Lawlis leads every ThriveAI project.
The session starts from the document you brought. You upload the document, write the instruction, read the answer and check the answer against the source. When the answer is wrong, you change the instruction and run the job again.
Along the way you pick up the habits that make an answer checkable. Give the assistant the whole document instead of describing the document. Ask where each figure came from, and save each instruction that works.
The goal is a result that works on your job, such as an instruction that turns the sales export into the bank update. Because the documents are yours, you can tell when a figure is wrong.
What to bring
- One job that takes hours every week or every month, such as the bank update or the quarterly price review.
- The documents behind that job: the last finished version, and the exports or statements the job draws on.
- If the job depends on the ERP, an export of the records the job needs, pulled by whoever runs the ERP.
- A list of the AI accounts you use today, and the plan each account is on.
Bring copies that you are allowed to share with an assistant. If a document holds personal data about employees, such as pay or health records, leave that document out or remove those columns first.
Accounts and settings to check before the session
Check which plan each assistant runs on before you upload a company file. The plan sets what the vendor may do with the files and the questions you send.
Claude
On a personal Claude account (Free, Pro or Max), Anthropic uses chats to improve Claude when the account holder allows that use. To check, open Settings, then Privacy, and look at the model improvement setting.
With that setting on, Anthropic may keep the data in de-identified form for up to five years. On commercial products such as Claude for Work, Anthropic says that by default it does not use inputs or outputs to train models.
The thumbs up and thumbs down buttons are the exception on every plan. Anthropic stores the whole conversation behind a rating for up to five years, and may use that feedback to train models. Leave the buttons alone on a chat that holds company files. On a Team or Enterprise plan, an owner can turn the buttons off with the Rate chats setting.
ChatGPT
In ChatGPT, open Settings, then Data controls, and turn off Improve the model for everyone. OpenAI’s help page says new conversations are then not used to train OpenAI models. OpenAI also says that by default the company does not train on content from ChatGPT Business or Enterprise workspaces.
Microsoft 365 Copilot
Microsoft says prompts, responses and data accessed through Microsoft Graph are not used to train foundation models. Microsoft’s privacy page for Copilot also says Copilot surfaces only the company data that each person has at least view permission for.
Before you ask Copilot about pay or contracts, check who else can open the folders that hold those files. For how ChatGPT, Claude and Copilot compare as a company workspace, see AI workspace for your team.
The same check applies to Gemini or any other assistant: find the vendor’s page on training and retention for the plan you use. For the wider question, see Is ChatGPT safe for confidential information?
How your team follows
Once you use AI on your own work, you can set the rules for everyone else from experience. Staff who use personal AI accounts with company files create shadow AI, and a written rule points those staff to an approved account.
Write the rules into the AI policy, then pick the next job for the team. Choose a job that takes hours every week, such as quoting, purchasing or order entry. Give the people who do that job a session on their own documents.
Keep one rule from the start: a named person approves every message that leaves the company. For where that approval sits in an AI system, see human in the loop.
Show the team a bank update that you drafted with AI, and the checks you ran on that update. Staff then see how much checking the company expects before a draft goes out. For how team sessions work, see AI training for your team. For the weekly jobs of a plant or warehouse manager, see AI for managers.
Working sessions with ThriveAI
ThriveAI is an AI engineering company in Ottawa that builds private AI systems for manufacturers and distributors in Ontario and Quebec. Training runs as working sessions on your own tasks and documents, on site at your plant or office, in French or English.
Derik Lawlis, the founder, leads every ThriveAI project. Each engagement is planned and quoted separately, around the jobs you choose and the people who take part.
He helped me map the flow visually on a whiteboard, then teaching me alongside him in Claude Code, we integrated all those tools into one process.
Fahd Alhattab, Founder, Unicorn Labs.
To try Claude on your own work before you talk to Derik, see how to use Claude at work.