Start your first AI workflow with one job your team already does every week. Pick work that repeats, has a clear finish, and can stay a draft while you test it. Then write down what goes in, what should come out, who checks it, and when the tool has to stop. You can test all of that before you connect AI to anything.
Most small teams have tried AI already. Someone uses it to fix an email. Someone else asks it for ideas. That can help, but the work still depends on who remembers to open the tool. A workflow gives one job the same path every time.
What Makes a Good First Job for AI?
"Help with operations" is too big to test. "Turn approved meeting notes into a draft action list" is small enough. Your team knows what goes in. It knows what a good result looks like. It can hold the draft next to the notes and check it before anyone acts.
That makes it a safer first test than letting a tool change an order, set a delivery date, or email a customer on its own.
Before you build anything, ask the person who does the job to show you the last few times they did it. Where do they look for facts? Which parts take judgment? What do they do when a note is missing something? If two people do the job two different ways, agree on one way first. AI should not have to guess your process.
This is how Grit Blueprint starts AI work with building industry teams. We look at the work before the tool. We follow the handoff: who has the facts, who needs the result, and who can say yes to the next step. A showroom note, a special order question, and a meeting action item each need a different answer.
Write the Job in One Sentence
Fill in this line before you open any tool:
When this happens, use these approved inputs to draft this result, then give it to this person to check.
Here is a made-up example for a building supply company. After the weekly operations meeting, the coordinator uses the approved notes to draft an action list. The meeting lead checks it. Only then does the coordinator add the agreed work to the team's task list.
The AI has a small part in this. It pulls out items and puts them in order. It does not decide what the company promised. It does not fill in missing owners or make up dates.
Your first version can be a blank document and a copy and paste step in a tool your company has approved. Prove the draft is useful before you hook it up to anything that can act.
Run a Made-Up Example All the Way Through
Use made-up facts for the first test. Keep customer records, personal employee details, and passwords out of any tool your company has not approved.
Made-up meeting notes: "Morgan will check the showroom sample stock by Thursday. The team talked about updating the delivery handout, but did not pick an owner or a due date. Someone suggested a customer notice, but nobody approved sending it."
What you tell the tool: "Turn these notes into a draft action list. Use only what the notes say. Keep agreed work apart from things we only talked about. If an owner or date is missing, write 'not decided.' Show the sentence each item came from. Do not send anything, create tasks or treat a suggestion as approved. If the notes conflict, flag it instead of picking an answer."
A useful draft would look like this:
| Item | Status | Owner and timing |
|---|---|---|
| Check showroom sample stock | Agreed | Morgan, Thursday (confirm the date) |
| Update delivery handout | Talked about, not assigned | Not decided |
| Send customer notice | Suggested, not approved | Not decided |
This is what a good result should look like for a made-up test. It is not a report from a real system.
The reviewer checks each row against the notes. The customer notice does not turn into a task to send. "Thursday" does not turn into a guessed calendar date. Missing facts stay in plain sight until a person fills them in.
Set a Stop Rule Before You Test
A line in a prompt is not enough to keep a connected tool from acting. So keep the draft and the decision apart. Making the list is one job. Deciding what really belongs on it is another. For the test, keep sending, task creation, and record changes out of the tool's reach.
Write down when the work must stop and go back to a person:
- The source is missing, unclear, or disagrees with another approved source.
- An item would make a promise, share confidential facts, or change a customer record.
- The draft adds a person, date, or step the source does not support.
- The reviewer cannot tell where an answer came from.
When the work stops, the team goes back to doing the job by hand. Name the person who finishes it. A failed test should never leave real work sitting with no owner.
If your workflow will draft anything a customer sees, use a checklist of what a person should check before a sales draft goes out.
Fill In This One-Page Pilot Worksheet
Sit down with the person who does the job and answer each line before you test:
- One job: What exact work are we trying to finish?
- Start: What tells us it is time to run it?
- Inputs: Which approved sources may the tool use? What must stay out?
- Result: What should the draft contain, and in what form?
- Reviewer: Who checks it and decides what happens next?
- Limits: What may the tool never send, change, or assume?
- Stop and fallback: What makes us stop, and who finishes the job by hand?
- Test cases: Which normal, incomplete, and conflicting examples will we try?
- Keep it if: What would make this worth keeping?
If you are not sure which documents count as approved inputs, start with how to set up sources your team can trust.
Run more than the easy example. Try notes with no owner. Try a decision that changed halfway through the meeting. Try a suggestion that sounds like an order. See if the draft keeps those apart.
How Do You Know the Workflow Is Worth Keeping?
Count all of the time. That means getting the notes ready, checking the draft, and fixing it. Compare that with doing the same job by hand. Keep a short log of missed items, made-up details, and times someone had to step in.
If the draft comes back fast but takes longer to fix than the job used to take, drop it or make it smaller. A draft that still needs a check can be worth keeping when the whole job takes less time.
Keep the parts that help. Make the job smaller if the same mistake keeps coming back. Connect the tool to other systems only when you know the risk, have the right approval, and can control what it is allowed to change. When you get to that point, compare an off-the-shelf AI tool with a custom workflow before you pay for either.
For most teams, training comes before any build. Our AI training for building industry teams works the same way as this worksheet. People practice on one task they already do, with real examples they are allowed to use. They learn to check what comes back. If one job needs a tool built for it, custom AI projects start with that one task and a small pilot.
This week, take one real job and fill in the worksheet with the person who does it. If you cannot name the input, the finish, and the reviewer, work that out before you build anything.
Questions and Answers
Pick one job your team does every week that has a clear finish and can stay a draft while you test it. Turning approved meeting notes into a draft action list is a good example. Your team knows what goes in, what a good result looks like and can check the draft against the notes.
No. Your first version can be a blank document and a copy and paste step in a tool your company has approved. Prove the draft is useful first. Connect the tool to other systems only when you know the risk, have approval, and can control what it may change.
A stop rule is a written list of times the work must go back to a person. Examples are a missing or conflicting source, a draft that makes a promise or changes a customer record, or an answer nobody can trace. When it stops, a named person finishes the job by hand.
Count all of the time, including getting inputs ready, checking the draft and fixing it, and compare it with doing the job by hand. Keep a short log of missed items and made-up details. Keep the parts that help and make the job smaller if the same mistake keeps coming back.
For most teams, training comes first. People learn to use AI on a task they already do and to check what it gives back. If one job then needs a tool built for it, start with that single task and a small pilot before you spend more.
Want Help Putting It to Work?
A free call with Stefanie Couch. About 30 minutes.
Not ready to talk? Get the Built to Win Newsletter


