Chapter 31 · companion worksheet

Derive your industry's boring AI use cases

The 12 boring use cases in the vertical chapters aren't arbitrary lists — they come from a consistent framework: high volume, structured task, measurable output, low judgment requirement at the execution layer, and a human-in-the-loop gate that protects the step requiring real expertise. Run this derivation for any industry.

Step 1 — List your 6 to 8 highest-cost workflows

Start with the workflows that consume the most staff time, generate the most errors, or create the most client friction. Don't filter yet.

Step 2 — Separate judgment steps from non-judgment steps

For each workflow above, list its main steps. Mark each step as J (requires professional judgment, regulatory compliance, or client relationship) or NJ (lookup, comparison, drafting, formatting, extraction, classification, or routing). AI candidates are the NJ steps.

Step 3 — Apply the 5 diagnostic questions

For each NJ step you marked above, answer the five questions. Steps that score yes on 3 or more are strong candidates.

1. Looking things up, comparing, or sorting? If someone on your team spends significant time searching through documentation, comparing options against criteria, or sorting items into categories — that is a Pattern 2 candidate.
2. What do your clients say they want faster? Speed is a leading indicator for a cycle-time compression opportunity. "We need the proposal faster." "Why does onboarding take three weeks?" Go find the workflow behind the delay.
3. Is output quality inconsistently variable? Consistency problems are almost always high-volume, structured-task problems. Humans can't maintain consistent attention across high-volume structured work indefinitely. AI can.
4. Is the human doing work that isn't judgment? Document assembly, data extraction, format standardization, first-draft generation against a template — those are non-judgment steps. Your expensive people should be spending their time on the judgment steps only.
5. What's the first artifact a new client or customer touches — and how was it produced? Client-facing artifacts almost always have significant non-judgment work embedded in their production. Improving them is immediately visible to the client.

Step 4 — Rank and select

Take your AI candidates from Step 2. Rank them by volume (highest first), then by unit cost of human execution (highest first). The top 3 to 5 are your boring use cases. Circle the one at the top for your first pilot.

What you will almost certainly find

• ☐ A document that gets read and summarized repeatedly
• ☐ An intake or triage process where incoming requests need to be classified and routed
• ☐ A reporting cycle where data needs to be assembled, formatted, and distributed
• ☐ A quality or compliance check that happens the same way on every unit of work
• ☐ A first-draft-generation step in a deliverable that follows a consistent structure

Those are your boring use cases. The specifics differ by industry. The structure doesn't.

My #1 candidate

Use case: ______
Workflow it lives in: ______
Volume per week: ______
What "done right" looks like in numbers: ______
HITL gate: ______