AI for managers in a plant or warehouse
AI for managers starts with five weekly jobs: meeting notes, the ERP report, schedule changes, emails and contracts.
Where AI fits in a manager’s week
A manager’s week in a plant or a warehouse holds a lot of writing. There are actions to send after each meeting, the weekly report, notes to the team, and emails to suppliers and customers.
An AI assistant such as ChatGPT, Claude or Microsoft 365 Copilot writes the first draft of each one from the documents you give the assistant. You check the draft, fix what is wrong and send the final version yourself. A named person approves anything the assistant drafts before the draft leaves the plant, the rule that human in the loop describes.
Use the assistant that your company already pays for, on a business plan. A personal account follows consumer data rules, so check the settings first, as AI training for executives explains.
Five weekly jobs to hand to an assistant
Give the assistant the documents behind each job, check the draft against those documents, and keep personal data about employees out of every prompt.
| Job | What to give the assistant | What to check | What never to paste in |
|---|---|---|---|
| Meeting notes to actions | The notes or the transcript, and who attended | Each action, owner and date against the notes | Remarks on a person’s performance, health or discipline |
| The weekly report | The ERP export, and last week’s report as the model | Totals against the totals in the ERP | Pay, hours or absences tied to a name |
| A schedule change | The old and new schedule, and the reason for the change | Dates, shifts and lines against the new schedule | Why a person is absent or on leave |
| Supplier and customer emails | The thread, with the price, quantity and date that apply | Every figure against the order or the ERP | A colleague’s phone number, address or personal situation |
| A first read of a contract | The full contract as a PDF | Each clause at the section the assistant cites | An employment contract or a settlement with a named employee |
Meeting notes to a list of actions
Paste the notes or the transcript, with the list of who attended and each person’s role. Ask for a table of actions with an owner and a due date for each, and nothing that the notes do not say.
Check each action against the notes before you send the list. An assistant fills gaps with plausible guesses, so look for an owner or a date that nobody agreed to. Leave out remarks about a person’s performance, health or discipline, which belong in the HR file.
The weekly report from an ERP export
Export the week’s figures from the ERP, such as shipments, late orders and scrap by line. Attach last week’s report as the model. Ask for the same report with this week’s figures, and a note on each line that missed the target.
An instruction for the weekly report can read: “Here are last week’s report and this week’s ERP export. Write this week’s report with the same sections. For each line, give shipments, late orders and scrap, and mark each line that missed the target. After each figure, name the column it came from. If a figure is missing from the export, write missing instead of a guess.”
Check the totals against the totals the ERP shows, and open a few rows by hand. Claude reads CSV files, and reads Excel files once code execution and file creation are turned on, according to Anthropic’s help page on uploads.
Leave pay, hours and absences by name out of the export, and report by line, shift or department instead. For reports that the ERP does not produce, see AI reporting.
A schedule change explained to the team
Give the assistant the old schedule, the new schedule and the reason for the change. Ask for a short note that tells each line or shift what changes and when, in French and English if the team reads both.
Check every date, shift and line against the new schedule before you post the note. Name lines and shifts in the prompt instead of people, and leave out why anyone is absent or on leave.
Supplier and customer emails
Paste the thread and say what you want from the reply, such as a firm ship date or a price confirmation. Give the facts that the reply needs, such as the price, the quantity and the date from the ERP or the purchase order.
Check every figure and date against the order before you send. You send the email yourself, from your own mailbox. Leave out a colleague’s phone number, home address or personal situation, even when the thread mentions those details.
A first read of a contract
Upload the full contract, such as a supply agreement or a customer’s terms and conditions. Ask for the clauses on price changes, delivery dates, penalties, liability and termination, each with the section number.
Read each clause at the section cited before you act on the summary. Bring the clauses that matter to the owner or to a lawyer before anyone signs. Never upload an employment contract or a settlement with a named employee.
Start with the job that fills your managers’ week
Tell Derik which weekly job takes your managers the most hours and who would attend. He will tell you what a session built on that job would cover.
Start a conversationKeep personal data about employees out
Every job above has the same exception. Personal data about employees stays out of the assistant, whichever plan the company is on.
- Health information, sick notes and requests for accommodation.
- Discipline, warnings and performance reviews.
- Pay, bonuses and hours tied to a name.
- Social insurance numbers, home addresses and personal phone numbers.
Folder permissions matter as well. Microsoft says Copilot surfaces only the company data that each person has at least view permission for. If HR files sit in a folder that every manager can open, Copilot can show those files to every manager.
Write the rule into the company’s AI policy, so every manager follows the same rule. For what a policy covers, see AI policy for your company.
Share the instructions that work
Keep each instruction that works in a shared file, with the job it belongs to and the checks to run. A supervisor who takes over the weekly report then starts from the instruction that worked last week.
In Claude, a project keeps the instructions and the files for one job together. Anthropic describes projects as self-contained workspaces with their own chat histories and knowledge bases. Review the shared file once a month, and remove instructions that no longer match the job.
Can managers be replaced with AI?
Gartner predicted in October 2024 that, through 2026, 20 percent of organizations would use AI to flatten their structure. According to the Gartner prediction, those organizations would eliminate more than half of their current middle management positions.
Gartner expected AI to automate scheduling, reporting and performance monitoring for the remaining workforce. Gartner also listed the challenges, such as remaining managers feeling overwhelmed by additional direct reports. In a plant or a warehouse, scheduling and reporting are the drafts that this page hands to an assistant.
The decisions stay with a manager, such as which order ships first or how to handle a supplier who is late. So does the approval of every draft before the draft leaves the plant. Handing the drafting to AI leaves more of the week for those decisions.
What project managers use AI for
The Project Management Institute (PMI) surveyed 500 project professionals in 12 countries in March and April 2024. According to PMI’s report, Pushing the Limits, project professionals use generative AI extensively to summarize project documents and reports. The report also names reviewing project plans, compiling executive summaries and condensing meeting discussions.
In a plant, the same jobs come up in the project file for a new line, a customer’s launch schedule or a capital project. Give the assistant the documents, ask for the draft, and check the draft against the source, as with the weekly jobs.
Training for managers on their own weekly jobs
ThriveAI trains managers on these jobs in working sessions, on their own tasks and documents. Sessions run on site at your plant or office, in French or English, and Derik Lawlis leads every ThriveAI project.
Each engagement is planned and quoted separately, around the jobs you choose and the people who take part. AI training for your team covers the people who quote, buy and enter orders, and AI training for executives covers the owner’s own work. To start on your own, see how to use Claude at work.
Derik at Thrive was instrumental in taking incredibly messy data we inherited in a business we acquired and, through using AI, organized it in record time in a way that made it reviewable by our team for final review and approval.
Rios-Karim Mercier, Belmont Capital.