# Session 1 Handout: Using AI as a Manager

<p class="meta">FAST-09 AI for Managers and Leaders · 7formations · Instructor: Wadood Chaudhary · All examples fictional · Training, not legal advice</p>

## Why this session matters

Your team watches what you do with AI more than what you say about it. Managers also handle the most sensitive information in a small business: reviews, pay, discipline, health notes, bids. Using AI well yourself, and visibly following the data rules, is the foundation for leading anyone else.

## Key concepts

### 1. The manager's data line

| Never in a prompt | Fine in the approved tool |
| --- | --- |
| Employee records, reviews, discipline and attendance notes | Your own sanitized notes |
| Salaries, pay, and benefits data | Generic descriptions ("a team member missed deadlines") |
| Customer personal information (G.L. c. 93H, 201 CMR 17.00) | Public documents: state guidance, vendor brochures |
| PHI, bids, unreleased financials, confidential drawings | Fictional or template text |

Sanitize first: roles instead of names, no ID numbers, nothing that points to one person. An employee ID still identifies someone. Personnel records are regulated under G.L. c. 149, § 52C.

### 2. Three management tasks

1. **Team update from your notes.** Fence your facts; tell the tool not to add dates, rules, or approvals.
2. **Summary of a long document** (policy, vendor proposal). Ask for the points that commit money, time, or data, then check each one against the original.
3. **Talking points for a hard conversation.** Generic situation only. Ask for an opening, open questions, and a close. Remove anything that diagnoses the person.

### 3. Judge every output on three things

| Check | Ask |
| --- | --- |
| **Accuracy** | Are every fact, number, date, and commitment right? |
| **Fairness** | Could any wording disadvantage someone (age, disability, gender, national origin)? |
| **Tone** | Is it right for the audience and the news? |

### 4. Supervising AI-assisted work

- Ask: "Which parts were AI-assisted, and how did you check them?"
- Check facts, numbers, commitments, and that no restricted data was used.
- Hold the sender accountable, not the tool.
- Coach better prompts and checks. Don't ban it quietly.

## Worked example (fictional)

A ClaimDraft Pro proposal summary came back as "affordable per-user pricing, flexible terms." The manager checked it against the original and found **auto-renewal unless cancelled 90 days before**, **up to 9% increases at each renewal**, and **customer content used to improve the vendor's service unless you opt out**. None of these were in the summary. The review note on W1: "Summary dropped renewal, price increase, and data-use clauses; rewrote with a terms checklist."

## Checklist

- [ ] Sanitized every input; nothing that identifies a person.
- [ ] Fenced the facts and told the tool not to invent.
- [ ] Checked each output for accuracy, fairness, and tone.
- [ ] Wrote specific keep / fix / discard notes on W1.
- [ ] Noted where AI did *not* save time.

## Do and don't

| Do | Don't |
| --- | --- |
| Describe situations generically | Paste a review or HR file |
| Check commitments before sending | Let AI promise refunds, dates, or approvals |
| Ask how work was checked | Blame the tool for a sent error |
| Coach a better check | Quietly ban AI after one mistake |
| Record meetings only with everyone's consent | Use an AI recorder in a one-on-one without consent (G.L. c. 272, § 99) |

## Try this before session 2

List five places in your team's week where AI might help: the task, who does it, and roughly how long it takes.