# Session 3 Handout: Learning from Cases and Building a Safe Workflow

<p class="meta">FAST-16 Working Well with AI · 7formations · Instructor: Wadood Chaudhary · All examples fictional except cited public cases · Suggested reading: Svekis, AI Literacy for All Employees (2026), chapters 13 and 15</p>

## Why this session matters

Most AI problems at work are not exotic. They repeat: a tool trusted too much, a check skipped, a group affected unfairly, people not told. Learning from cases lets you spot the pattern before it happens to you, and a five-step workflow makes good habits routine.

## Key concepts

### 1. What the cases show

| Case | Benefit | What went wrong | Safeguard |
| --- | --- | --- | --- |
| Cape Harbor Inn chatbot (fictional) | Fewer routine calls | Wrong refund rule | Ground in policy; route refunds to a person |
| North Shore Builders screening (fictional) | Faster hiring | Applicants with career gaps rarely ranked high | Monitor outcomes; person reviews all qualified applicants |
| Lowell BioSupply forecasting (fictional) | Less overstock | Missed a supplier shock | Forecasts are inputs; add outside factors |
| Bayside invoice automation (fictional) | Hours saved | Errors on one vendor's layout | Route exceptions to a clerk |
| Merrimack monitoring (fictional) | Data on productivity | Privacy concerns, lost trust | Explain, involve staff, limit data |
| Mata v. Avianca (S.D.N.Y., 2023) | n/a | Invented case citations filed | Verify every citation |
| Moffatt v. Air Canada (B.C. CRT, 2024) | n/a | Chatbot gave wrong fare policy; airline held responsible | The deployer owns its tool's answers |

### 2. Success factors and common challenges

Success: clear goals, trained employees, transparency, ongoing monitoring. Challenges: over-reliance, bias, privacy concerns, poor data quality, and resistance when people are not involved.

### 3. The five-step safe workflow

| Step | Ask yourself |
| --- | --- |
| Define | What is the task? What are the stakes? Is AI a good fit? |
| Instruct | Four elements, gap rule, only approved and necessary data |
| Review | Five-step evaluation against a named source |
| Judge | Would I sign this? Who else must? When do I discard it? |
| Document | AI-use note where policy requires it |

### 4. Documenting AI use

A short note such as "Drafted with [approved tool]; figures checked against the March report by [role]" lets others trace how the work was made. It supports accountability; it does not move responsibility to the tool.

## After the course

- Finish W4 and keep W5 at your workstation.
- Next steps: FAST-03 (Prompt Engineering for the Workplace) for source-bound prompting at a higher level, or FAST-17 (Responsible AI for All Staff).