We mapped the work that has to happen for an order to ship at a manufacturer last month. Quoting, contract review, purchasing, scheduling, shipping, invoicing — thirty-four distinct steps across six areas of the business.
Then we asked one question of each step: who else can do this today, without being walked through it?
Thirty-three of the thirty-four depended on a single person.
Not "primarily one person." One person. If they were out for two weeks, the work would either stop or land on whoever was willing to improvise.
Why nobody notices
The risk is invisible while the person is still there. The work gets done, the orders ship, the phone gets answered. Nobody experiences their own dependency day to day, and the person carrying it usually doesn't think of it as unusual — they've simply always done it.
What makes it dangerous is the timing. This kind of knowledge doesn't leave gradually. It leaves on a Friday afternoon, all at once, usually without warning. The two-week notice is the entire handover.
And it isn't only the tasks that walk out. It's the exceptions. Every business runs on a documented process and a hundred undocumented workarounds: what to do when the system count disagrees with the shelf, which customer gets priority when capacity is tight, which form still has to be faxed because one account manager insists on it.
The process lives in a binder. The judgment lives in a person.
Why it can't be automated away
This is where a lot of AI initiatives quietly fail. You can't hand a workflow to a tool if nobody has written the workflow down.
AI accelerates what's documented. It has no way to infer the logic someone carries in their head, and it certainly can't learn the exceptions you don't know you have. Why process comes before AI isn't a slogan — it's the mechanical reason most automation projects stall in month three.
Automating an undocumented process doesn't remove the dependency. It hides it behind software that will look like it's working right up until the day it isn't.
A one-hour version of the exercise
You don't need to map your whole business. Pick the process whose failure would hurt most — usually quoting, order entry, or shipping — and break it into its actual steps, not the version in the SOP.
For each step, name everyone who can do it today without help. Count them. Any step with one name, or none, is a single point of failure. That's the list. Everything else is noise.
Then, in order:
-
1. Say it out loud.
Not "we should cross-train." Name the three steps that would actually stop the business, and who owns them.
-
2. Capture the exceptions.
The normal path is probably written down somewhere. The exceptions aren't. That's the hour that matters.
-
3. Train one backup on the step closest to revenue.
Not the easiest one, not the most senior person's. The one that costs the most to drop.
-
4. Then automate.
Once a step is documented and two people can do it, automation makes it faster — and you'll know which processes are actually worth automating. Before that, it just makes the risk harder to see.
The question worth asking this week
If the person who owns your most critical process gave notice on Friday, what happens on Monday?
If the honest answer is "we'd figure it out," the follow-up is harder: do you know how — or would you be reconstructing it from memory, under pressure, with customers waiting?
You don't need a matrix with thirty-four rows. You need that one answer.