# Session 4 Handout: Your Use Case

<p class="meta">FAST-07 Industry-Specific AI Applications for Massachusetts Sectors · 7formations · Instructor: Wadood Chaudhary · All examples fictional · Training, not legal advice</p>

## Why this session matters

A good idea for AI at work goes nowhere without a short, honest case that a manager can approve: what the task is, how AI would help, what could go wrong, how you will check it, and how you will know it worked. This session turns what you learned in the sector labs into that one-page brief, tests it with peers, and sets you up to start the following week.

## Key concepts

### 1. The one-page use-case brief

| Section | What to write | Weak | Strong |
| --- | --- | --- | --- |
| Task | What, who, how often | "Reports" | "Nonconformance reports, quality techs, about 12 per week" |
| AI help | Method and approved tool | "Use AI" | "Approved assistant drafts the narrative from sanitized notes using the NCR template" |
| Data rules | From your W6 list | none | "No customer drawings or names; part numbers only" |
| Risks and controls | Failure, check, owner | "Be careful" | "Wrong dimension: tech checks every value against the inspection record" |
| Benefit | Indicator, baseline, target, review | "Saves time" | "40 to 25 min per NCR including check; rework at or below 15%; review day 45" |

### 2. Measuring benefit honestly

- **Pick one indicator:** time per task, error rate, rework rate, or response time.
- **Measure a baseline** before the change (at least a few instances, ideally two weeks).
- **Include checking time** in the new number. A saving that ignores review is not a saving.
- **Watch for side effects:** if time drops but rework doubles, the net benefit may be negative.
- **Set a review date and a decision owner:** continue, change, or stop.

Research shows AI help is uneven: large gains on some tasks, worse results on others (the "jagged frontier"). Measuring your own task is the only way to know which side it falls on.

### 3. Pitch and critique

Three minutes per pitch. Peers ask two questions every time:

1. **What could go wrong?** Add a named risk and a specific control.
2. **How will you know it worked?** Tighten the indicator and the baseline.

Peers score industry accuracy, task quality, risk controls, and measurability.

## Worked example: Merrimack Freight & Logistics (fictional)

**Task:** Daily shipment exception summaries for the operations manager; a dispatcher writes about 15 per day, 6 minutes each (measured over 10 days).

**AI help:** The approved assistant drafts each summary from the sanitized exception log, using the four-field format (ID, what happened, cause as recorded, status).

**Data rules:** Driver and consignee names and phone numbers removed before pasting; customer rates never pasted.

**Risks and controls:** (1) Cause assigned that is not in the log: dispatcher checks each cause; prompt requires "cause not recorded." (2) Wrong shipment ID: dispatcher checks IDs against the dispatch system. (3) Personal details pasted: sanitizing step in the template; weekly spot-check by the operations manager.

**Benefit:** 6 to 3 minutes per summary including the check; zero carrier emails with unsupported blame. Review on day 45 by the operations manager.

## Checklist: is my brief ready?

- [ ] Task, owner, and volume stated.
- [ ] Approved tool named.
- [ ] At least two data rules from my list applied.
- [ ] Each risk has a check and an owner; a stop rule exists.
- [ ] Indicator, baseline, target, review date, decision owner.
- [ ] Sanitized and approved by my manager for sharing.

## Do and don't

| Do | Don't |
| --- | --- |
| Measure before you change anything | Estimate the baseline from memory |
| Count check time | Report draft time only |
| Pick a task you can measure in 45 days | Pick "transform operations" |
| Revise from peer questions | Defend the first draft |

## After the course

Start the baseline next week. Share the brief with your manager and IT. Bring the 45-day results back to your team, whatever they show.