Design an AI Workflow That Sticks · Lesson 1 of 7
Lesson 1 — The workflow audit
30 minutes. You'll leave this lesson with a ranked list of your top 10 AI workflow candidates.
---
Most AI projects fail not because the tools are bad — they fail because the *workflow* was wrong. Wrong place, wrong hook, wrong scope, wrong owner. The audit is how you avoid that.
For any candidate workflow, ask:
- How often does it happen? Once a year is not a workflow. Once a week minimum. Daily is ideal.
- How much does it cost when it slips? A missed customer reply ≠ a missed payroll check. Score the consequence.
- How rule-based is the output? "Write a friendly reply" is fuzzy. "Take this PO and check if line items match our catalog" is rule-y. Both can use AI, but they need different designs.
- Who owns the result? If nobody owns it today, AI won't fix that.
Open a doc. Spend 10 minutes brainstorming every repeatable thing that happens in your business that involves reading, writing, sorting, comparing, explaining, or translating.
Don't filter yet. Just list. Aim for 25-30 items.
Examples to prompt yourself:
Reading:
- Reading customer emails when they come in
- Reviewing invoices before paying them
- Skimming vendor contracts when they arrive
- Reading meeting transcripts after the fact
Writing:
- Drafting quote responses
- Writing follow-up emails after calls
- Composing newsletter copy
- Updating job descriptions
Sorting:
- Triaging the inbox by urgency
- Routing customer requests to the right team member
- Categorizing receipts for the books
- Tagging leads by quality
Comparing:
- Comparing this month's revenue to last month's
- Reconciling bank statements to invoices
- Checking new vendor proposals against current rates
Explaining:
- Explaining your product/service to a new prospect
- Translating technical jargon for non-technical clients
- Writing FAQ answers
- Onboarding new employees
Translating:
- English ↔ Spanish customer communication
- Marketing copy → customer-facing language
- Engineering specs → sales pitch
Now the four-question filter. For each item, score 1-5:
- Frequency (1 = annual, 5 = daily)
- Cost of failure (1 = nobody notices, 5 = customer pays, regulator pays attention, or you lose money)
- Rule clarity (1 = "I just know," 5 = "here's the exact algorithm")
- Owner (1 = nobody, 5 = clearly assigned to one person who would gladly hand off the boring parts)
Total each row. Sort by total. Top 10 wins.
You haven't designed a workflow yet. You've ranked your *candidates* — which is what 95% of operators skip when they jump into building. They build the workflow that sounded fun in the meeting instead of the one that's most likely to stick.
You're going to pick your hook from the top of this list in Lesson 2. But first — a warning.
The top of your ranked list is *usually not* the workflow you should start with. The very top is usually:
- Too central to the business (high cost of failure)
- Too political (multiple owners disagree)
- Too complex (lots of edge cases)
The right starting point is 2nd or 3rd from the top — high frequency, real cost-of-failure, clear rules, clear owner, but not so central that one bad output ruins your week.
In Lesson 2 we pick which one.
→ Lesson 2 — Picking the right hook