Most workflows are wrong for automation.
Teams burn quarters on work that was fine already, or work that changes every month, or work that one person understands and nobody can describe.
Seven tests sort this out in about ten minutes.
Test 1: Does it repeat?
Volume beats complexity every time.
A hard task you do twice a year is a poor candidate, however satisfying it would be to solve. An easy task you do four hundred times a month is where the money is.
How many times did this run last month? Under twenty, park it.
Test 2: Can you describe the rules?
Sit with the person who does the work. Ask them to explain the decisions they make.
Listen for how often "it depends" appears.
Some judgment is fine. AI handles fuzzy inputs well. Builds die on judgment nobody can put into words, because the rules live in one person's head and they left in April.
Cannot explain the decision? You can still automate the boring parts around it.
Test 3: Is the input consistent?
Where does the work start?
A structured form is easy. A shared inbox where clients send PDFs, photos of documents, and messages saying "as discussed" is much harder.
Messy input still works. It costs more, breaks more often, and needs a human checking output for longer than anyone expects. Price that in.
Test 4: What does being wrong cost?
This decides how much you build.
Low stakes work, like an internal summary, runs with light checking. A mistake is mildly annoying and someone fixes it.
High stakes work needs a human approving before it goes out. Anything a client sees. Anything with a number that affects money. That is fine. It also shrinks your saving, because the checking stays.
Be honest about which one you have. Teams routinely file client facing work as low stakes and find out the hard way.
Test 5: Is the process stable?
Has this workflow changed in the last six months?
Constant change means constant rebuilding. Automate the stable spine and leave the moving parts alone.
Anything mid restructure fails this test. Wait until it settles.
Test 6: Who owns it?
Name the person who owns this after it is built.
Someone whose objectives include keeping it running. The person who suggested it does not count. Neither does the enthusiastic one from the workshop.
No name, no automation. It works for three weeks and quietly rots.
Test 7: Will people actually use it?
The hardest test, and the one most often skipped.
Ask the person who does this work today whether they want it automated.
Sometimes you get a delighted yes. Sometimes you get a polite yes covering a quiet fear about what happens to their role. That second answer predicts failure better than any technical assessment.
Deal with it in the open before you build. People will work around a tool they resent, and they are creative about it.
Score it
Give each test a mark out of three.
- 3 means clearly yes
- 2 means workable with effort
- 1 means a real problem
- 0 means a blocker
Add it up out of 21.
17 and above. Build it. This is your first project.
12 to 16. Fix the weak tests first. Look at which ones scored low and whether they are fixable.
Below 12. Leave it. Something else in your business scores higher and you have not found it yet.
Any zero. Stop, whatever the total says. A blocker on ownership or willingness sinks a workflow that scores well everywhere else.
The shapes that usually win
Across small and mid sized businesses, the same five keep scoring highest.
- Turning messy input into clean records. Emails into fields, documents into data, notes into a summary.
- First drafts a human edits. Quotes, proposals, responses, reports.
- Research and gathering. Pulling together what is known about a client, a supplier, or a market before a meeting.
- Checking and flagging. Reading a batch and surfacing the three items that need attention.
- Routing. Reading what came in and sending it to the right person with context attached.
The shapes that usually lose need real world judgment about people, change their rules weekly, or replace a process that already works fine.
Before you build
Run all seven tests on your top three candidates. Write the scores down.
Your instinct is often wrong. The workflow that felt obvious scores an eleven. The boring one nobody mentioned scores nineteen.
That is the point of scoring it before you build it.
Score your own workflows in two minutes.
The AI Opportunity Audit runs these tests across your workflows and hands you a ranked list with effort, cost, and payback for each.