The Same-Shape Test

“It’s faster if I just do it.”

It is, the first time. It is also the sentence that keeps a Monday task on your plate for three years.

Before you hand anything to AI, you need to know which tasks can go. A tool list can’t tell you that, but a test can.

What to delegate to AI: three questions

Take one task. Ask three things. Am I starting it with the same kind of input every time? Does it end with the same kind of result? Does it go to the same next person or system?

Three yeses mean the task has a shape. A task with a shape can be written as a rule, and a system can follow a rule. One no means the task still needs you, for now.

The test says nothing about how hard a task is. A client status update takes real thought. It still starts with project notes, ends as a short update, and goes to the client. Hard to write, easy to hand over. The rule for it costs something, though. Plan on an hour or two to write down what a good update looks like. That is a rough guess, and it changes with the task.

A worked example

This is a made-up Monday list, here to show the shape of the test. It is not a client story.

A consultant writes down 10 tasks. Weekly status updates. Answering “where is my invoice.” Kickoff emails for new clients. Scheduling. Summarizing call notes. Filing signed contracts. Deciding whether to take a project outside her usual work. Writing a proposal for a kind of client she has never served. Handling an upset client. Reviewing a team member’s deliverable before it ships.

Six pass cleanly: the updates, the invoice question, the kickoff emails, the scheduling, the call notes, and the contract filing. Same input, same result, same next person, every time. Those six are Silent Sinkholes. Say the updates take 90 minutes a week, the invoice question 45, kickoff emails 60, scheduling 90, call notes 60, and contract filing 30. None of them looks like a problem alone. Together they are a little over six hours.

Three fail. The take-or-pass decision gets new facts every time. The new proposal has no pattern yet. The upset client needs a relationship. AI can help draft the reply, and the relationship stays with you.

The tenth is the interesting one. The deliverable review passes all three questions, but it carries your taste. It moves once you write down what good looks like. After that, a system can check against your standard and send the result along.

So six tasks move now, one moves after a written standard, and three stay with you.

The worth-it check

Passing the test tells you a task can be handed over. Whether it should be depends on a second check.

Compare two numbers: the hours to set it up and the minutes it costs you each week. A task that takes four minutes a week and two hours to set up pays you back in 30 weeks. A task that takes 90 minutes a week pays you back in under two. Start with the big ones.

After it passes

A passing task still needs three things.

Sort by what a mistake costs. A scheduling slip is small. A wrong contract filing can cost you a client’s trust. Both pass the test. Big ones get a named person’s approval before anything leaves the building, as I covered in the first post in this series.

Give the system an owner. Every handoff leaves one written line: what moved, who got it, and when. The owner reads a small sample each week. That is how a quiet failure gets caught early.

Expect shapes to change. A client switches formats, or a contract template gets updated. When that happens, the rule needs a fresh look. A task that fails today may pass next quarter after you have done it three times. I wrote about the difference between a timer and a system that adjusts in A Schedule Is Not Staff.

Picture the day where the status updates are already sent when you sit down. You find out when a client replies to one.

The Profit Leak Scorecard shows where those Monday tasks are still waiting on you.

---

It’s Handled™ | Operational architecture for founders who built the business, and are ready to stop being the infrastructure.