The four-blank test

It's 4pm on a Tuesday and somebody has said it again in a meeting: we should automate the job sheets. Everyone agrees, the meeting moves on, nothing is written down. Three months later the same sentence comes back, still with no idea what better would look like. That gap is the loop this post: jobs nobody can describe the win for, talked about automating for years.

The test takes five minutes and a pen. Fill in four blanks for the job you have in mind:

• The bucket. Is this job there to win more customers, to earn more per customer, or to lower cost? One of those, written down.
• The number. The one measure this job moves.
• The baseline. Where that number sits today, in units, not adjectives.
• The target. Where it should be in 60 days.

Any blank left standing, and the job is not ready to automate. Not because AI can't do it. Because you can't describe "better" yet, so you would never know whether the machine moved anything. A tool that can't be judged gets kept out of habit or dropped in a huff. Either way, the money is gone.

Try it on job sheets for the shape. Bucket: lower cost. Number: minutes of paperwork per job. Baseline: blank. Most owners have never timed the paperwork, which is exactly why it keeps eating evenings. Fill that blank and, in 60 days, the minutes either moved or they didn't.

The blanks cost more than any tool. Unmeasured work can't be improved; it just repeats quietly for years.

The maths (a sketch, not a client figure)

Take job paperwork as the shape. Suppose 15 minutes per job, 10 jobs a week:

• 15 minutes × 10 jobs = 150 minutes a week of paperwork
• × £30 an hour = £75 a week
• × 46 weeks = £3,450 a year of typing nobody chose.

The prize depends on the baseline you haven't measured. If the real figure is 15 minutes and a fix cuts it to 5, the sketch says £2,300 a year back. If the paperwork turns out to be 5 minutes a job, the prize is a third of that. Either way you only know once the number exists; a tool bought before it exists can't be judged against anything. (A sketch, not a client figure: the figure that matters is yours.)

Score it: the three checks

The four blanks ask whether a job is worth handing over. The three checks ask whether AI can hold it. Take job-sheet paperwork through both:

• Hours. 150 minutes a week in the sketch (3).
• Repeat. Fill the fields from the job; a new starter could follow it from one page (4).
• Risk-if-wrong. A slip is a wrong field, catchable on review before anything leaves (3).

Two checks at 4 or more, so the verdict is start narrow: automate the typing of the job sheet, not the scheduling around it, with a person approving each one. Run the blanks first: without the baseline you can't prove the typing got faster.

Try one thing this week (free)

Pick one job and fill the four blanks on one side of paper this week. Five minutes, no purchase. The blank you stall on is the finding. Most owners stall on the baseline, the blank that decides everything else: timing the paperwork for one week costs nothing and gives you the only number in this post that is real. Judge it on Friday.

Where a person stays

The four blanks are filled and signed by a person. Software can draft a guess at the baseline from your data, but the owner signs the number, because a baseline nobody checked is a wish. The same rule runs through the work: AI drafts, a person approves, nothing sends unchecked. The bucket, the measure and the definition of better are decisions, and decisions keep a human name on them.

FAQ