# Session 1 Handout: See the Process

<p class="meta">FAST-08 AI for Workflow Improvement and Process Automation · 7formations · Instructor: Wadood Chaudhary · All examples fictional</p>

## Why this session matters

Most automation projects that disappoint started by automating the wrong thing, or by automating a broken process as it is. Before you build anything, you need to see the process as it really runs, know where the time goes, and decide which steps are worth automating at all. This session gives you the map, the measurement, and the "when not to automate" test.

## Key concepts

### 1. Map what actually happens

- **Swimlanes:** one row per person, team, or system (Customer, Coordinator, Pricing, Claims, Billing, TMS).
- **Steps:** one sticky note each, verb first ("Re-type request into sheet").
- **Handoffs:** arrows that cross lanes. Each one is a chance to wait or lose information.
- **Waits (W) and error points (red dot):** mark them on the map.

Map the real process, including workarounds, not the version in the procedure manual.

### 2. Measure: touch time, wait time, lead time

| Term | Meaning |
| --- | --- |
| **Touch time** | Minutes someone actively works on the item |
| **Wait time** | Time the item sits with nobody working on it |
| **Lead time** | Arrival to completion, elapsed clock time = touch + wait |

Estimate first, then measure at least 10 real instances (fictional in class). Most people underestimate waiting badly.

### 3. Where time and errors hide

| Waste | Office example | Automation candidate? |
| --- | --- | --- |
| Re-keying | Typing an emailed request into a sheet | Yes: extract and log |
| Waiting | Request sits in a queue for a day | Sometimes: routing, reminders; often a policy fix |
| Rework | Misrouted request forwarded twice | Yes: classify with a review queue |
| Send-backs | Missing PO number, back to the customer | Partly: required fields on the form |
| Searching | Finding the last quote for this lane | Yes: lookup or summarize |

### 4. When not to automate

- Rare (a few times a year) or highly variable work.
- Judgment-heavy decisions: liability, pricing exceptions, discipline.
- A broken process: fix the policy first (for example, a same-day approval rule).
- Anything you cannot check.

## Worked example: Merrimack Freight quote requests (fictional)

Five quote requests measured: average touch time **20.6 minutes** (coordinator 6.2, analyst 14.4); average lead time **about 21 hours**. Waiting is **over 95%** of lead time. The biggest win is not typing faster; it is removing re-keying and getting requests to the analyst sooner with the right details.

## Checklist: is my map ready?

- [ ] Swimlanes for every person, team, and system.
- [ ] Every handoff drawn; waits and error points marked.
- [ ] Estimate recorded before measuring.
- [ ] 10+ measured instances; touch, wait, lead time.
- [ ] Waste labeled; "do not automate" steps identified.

## Do and don't

| Do | Don't |
| --- | --- |
| Map the workarounds | Map the manual |
| Measure waits | Time only the typing |
| Ask "should this step exist?" | Automate every step |
| Keep judgment with people | Hand liability decisions to a flow |

## Try this before session 2

Time one repeatable task you do, five times. Record touch and wait separately.