Pick a 30-minute slot. Bring the team. Run this exercise. Leave with three workflows to AI-ify and one to leave alone.
Step 1: The four questions
Have everyone list workflows they personally own, then score each one on these four questions (1-5):
- Frequency. How often does this happen? (5 = daily, 1 = annually)
- Predictability. Does it look the same each time, or wildly different? (5 = highly predictable, 1 = bespoke every time)
- Pain. How much do you hate doing it? (5 = soul-crushing, 1 = fine)
- Reviewability. Could a human spot-check the output in a few minutes? (5 = obvious if wrong, 1 = nearly impossible to audit)
Step 2: The math
Multiply the four scores together for each workflow. The math isn't precise — but the ordering matters.
- 250-625: Top candidates. Ship these first.
- 120-249: Second wave. Worth it once the top candidates are live.
- Under 120: Skip. Either too rare, too bespoke, too tolerable, or too hard to verify.
Step 3: The reality check
For your top three, ask one more question: does the data exist?
This is where most AI rollouts die. A workflow can be frequent, predictable, painful, and reviewable — and still un-AI-able because the team has never written down what "good" looks like. If your top candidate has no past examples, no documented standards, no consistent format, demote it. Pick the next one down with cleaner inputs.
The best AI candidate isn't the most painful workflow. It's the most painful workflow that's also legible.
What this commonly surfaces
For most firms we've audited:
- AEC firms get RFP analysis, daily reports, and submittal processing as their top three.
- Professional services get proposals, status reports, and CRM updates.
- Schools and districts get translated communications, lesson differentiation, and progress reports.
- Nonprofits get funder research, LOI drafting, and donor thank-you notes.
If your top three look different than this, that's fine — your business is yours. But if your top three is "executive coaching" or "creative ideation," you've probably scored Predictability and Reviewability too generously. Run it again.
What to do after the audit
Don't try to ship all three. Pick the one that:
- Has the cleanest training data already in hand.
- Will be most visible to leadership when it works.
- Has a clear owner who'll evangelize it.
Ship that one well. Use it as proof for the next two. The firms that ship one AI workflow brilliantly in quarter one ship four in quarter two. The firms that ship four mediocre ones in quarter one stop shipping in quarter two.