In a lot of manufacturing offices, the day starts with the same spreadsheet. Someone opens the order log, scrolls past forty columns to reach the handful they use, and gets on with it. Ask them which columns they could not explain and the list is long. Ask where those columns came from and the answer is usually a person: someone who wanted a report once, got a column added, and later moved on.

Nobody deleted the column, because nobody knew whether it mattered. So it stayed, along with the tab next to it and the weekly summary that people still fill in and nobody reads.

That spreadsheet is where an owner's first AI idea tends to land: can the AI fill this in for us? It can, and that is the trouble. The work an owner most wants to hand to AI is often the work that should not be done at all.

1. Turnover leaves orphan steps behind

We call these orphan steps: a column, a tab, a report, or a sign-off that outlived the person who asked for it. Each one made sense on the day it was added. A sales manager wanted to track one customer's blanket order. A quality lead wanted a flag on one supplier's certs after a bad shipment. Then the manager left, the supplier was dropped, and the step kept going because it was on the form.

From the owner's office, orphan steps are hard to see. Leaders usually believe they know the process, but in a company that has run for decades, much of it was never written down. People do things the way they learned them, under pressure, and they get it done. The person who does the job every morning knows exactly which columns they skip. Often nobody has asked them.

Our view: orphan steps are a big part of why the same office job feels heavier every year even when order volume has not changed. Each one costs a few minutes. Nobody ever adds them up.

2. AI makes an orphan step faster and harder to kill

Point AI at the order log as it stands, and it will fill all forty columns every day without complaint. You get a faster way to do work that should not exist.

There is a second cost. Today, the person scrolling past column 31 is a small alarm, because sooner or later someone may ask why it is there. Once the AI fills it in, nobody scrolls past it, and the question never comes up again. The orphan is now built into software.

And every field costs something once AI is involved. Before an AI employee (an AI set up like a new hire, with its own login and one defined job) goes live, we test its first pass field by field against past work your team already got right. After go-live, a named person checks each draft against the source before anything leaves the building. A misread heat number is worth catching. A column that feeds a report nobody reads adds review time to every draft and gives nothing back.

3. Ask why a step exists before you cut it

The fix is plain and a little slow. Before we build anything, we sit with the people who do the job and walk through it step by step, asking why each one is there.

The opposite mistake is just as real. In 1929, the writer G.K. Chesterton described a reformer who finds a fence across a road and wants it cleared away because he cannot see its use. Chesterton's advice was to go and find out why it was put there before tearing it down. Some columns that look useless are holding up something real, like a customer requirement that only one person still remembers.

So for each step, we ask three questions:

  1. Who uses this, by name, and what do they decide with it?
  2. What would go wrong if it were blank for a month?
  3. Does the same information already live somewhere else?

If nobody uses it and nothing would go wrong, it is an orphan. If the information lives somewhere else, it is a copy, and the job should read it from one place instead of retyping it.

4. Keep what a customer or a decision depends on

Cutting is half the work. The other half is being clear about what stays, because that list becomes the AI's job description.

Here is how a review of five columns in a typical order log might go:

  • The heat number stays. It is how a part traces back to its mill cert, and a customer or an auditor can ask for it years from now.
  • The customer's required cert wording stays, and it moves out of one person's head into the customer record, where the AI can read it every time.
  • The promised ship date and the actual ship date stay, because someone decides what to expedite by comparing them.
  • The customer name typed again on a second tab goes. The job reads it from the order instead.
  • The flag for a weekly report nobody has asked for since its author left goes too.

What stays is usually shorter than people expect. It is also much clearer: a short list of fields that matter, each with a source the AI can point to and a person who checks it. That is the version of the job worth giving to an AI employee.

5. Without leadership commitment, we do not start

Deleting a column sounds small. In practice it means telling someone their report is going away, changing a form that three departments touch, and trusting that the step nobody could explain was an orphan. The person who does the job cannot make that call alone. The owner or the department head has to.

That is why we ask for leadership commitment before we start. If the owner and department heads are not willing to change how the work is done, the change does not happen, however good the software is. We do not take on a project without that commitment, because without it we would be building a faster version of the old mess.

The software side is no longer the hard part. Custom software got cheap to build, so a manufacturer no longer has to bend its work around a tool someone else designed. The tool can fit the cleaned-up job. The hard part is agreeing on what the job is.

What to do this week

  1. Sit down with the person who uses your biggest shared spreadsheet most, and mark every column they cannot explain.
  2. For each marked column, ask the three questions: who uses it, what breaks if it is blank for a month, and whether it lives somewhere else.
  3. Take the columns with the weakest answers and hide them for a month instead of deleting them. If someone asks for one, bring it back and write down why it exists.
  4. Ask your leadership team whether they are willing to change how this job is done, or only to make the current way faster.

This essay comes from Manufacturing's ChatGPT Moment Hasn't Happened Yet: Why AI Fits Your Paperwork and How to Prepare. For how a cleaned-up job becomes one digital record the AI can read, see From Paper Traveler to Traceability: Why Digital Forms Come Before AI.

Questions people ask

Should we clean up our processes before we use AI?

Clean up the job the AI will touch first. Sit with the people who do it, ask why each step exists, and cut or merge the ones nobody can explain. Then the AI drafts only what a customer or a decision depends on, and your reviewer checks fewer fields.

How do we know a step is safe to remove?

Ask who uses it and what would go wrong if it were blank for a month. If the answers are weak, hide it for a month instead of deleting it. If someone asks for it back, you have learned why it existed.

Terms in this piece

Orphan step
a column, tab, report, or sign-off that outlived the person who asked for it, and that people keep doing because it is on the form.
First pass
the draft the AI prepares, such as a cert packet or a quote, before a person checks it.
AI employee
an agent set up like a new hire, with its own login, access to specific systems, one defined job, and a named person who reviews its work.
Reviewer
the named person who checks the AI's work and approves it before it counts.
Traceability
the ability to follow any part back through every step of the job to the material and documents it came from, from mill heat to shipment.