The Hidden Factory

Problem: The report says the process takes four days. Everyone who does the work says two weeks. Both are correct, and the gap between them is where your capacity is going.

Quick Win: The missing time lives in the hidden factory, a plain name for the work your company does every day that appears in no process document and no system report: the chasing, the re-keying, the corrections, the private spreadsheets, the favours. Armand Feigenbaum named it in the 1970s. It is "that part of your organization that exists to do bad work," and he put "anywhere from about 20% to 40% of total capacity of many American companies" inside it (IndustryWeek). You cannot report your way to it. You have to ask. Here are the fourteen questions, the order, and the test for when a step is captured.

Why The Dashboard Cannot See This Work

This is not a data quality problem you fix with a better report. It is structural. Every reporting tool can only read what a system wrote down.

Process mining, which rebuilds a process from the digital trail your software leaves behind, is the best version of this and has the same hard limit. It sees every step that touched a system and none of the steps that did not. The vendors are open about it. Task mining, which watches what people actually do on their desktops, exists precisely because process mining misses "the manual steps, workarounds, and shadow processes" (Mimica), and Celonis describes its own process mining as something you combine with task mining data, "which captures how work flows across a company's desktop applications" (Celonis).

The best automated discovery tool on the market needs a second instrument to see the work between the systems. And that work is not a rounding error. In Miro's 2025 Momentum at Work report, a survey of 6,148 knowledge workers across eight markets, 57% said they redo work in different tools or apps (HR Dive). None of that redoing leaves a record.

So the interview is not a soft add-on to the real work with the numbers. For this kind of step it is the only instrument that reaches it.

The Hidden Factory, Defined Plainly

A step belongs to the hidden factory when three things are true at once:

1. It is in no written process, checklist, or system workflow.
2. It consumes real hours from real people, every week.
3. It happens somewhere that produces no record: an inbox, a chat thread, a spreadsheet, a phone call, somebody's memory.

Two ordinary examples. The salesperson who rebuilds the quote in a spreadsheet because the official tool cannot handle the discount structure. The finance lead keeping a private list of invoices with a wrong purchase order number, because the system will not flag it until it is late.

Neither has an owner, a step number, or a line in a report. Both are load-bearing. Remove the person and the process stops.

This is also why improving what you can see so often produces nothing. The Lean Enterprise Institute defines a constraint as "any factor that limits the organization from getting more of whatever it strives for," and notes that "most organizations have very few true constraints" (Lean Enterprise Institute). If the one real constraint sits in the hidden factory, every documented step you optimise is the wrong step, which is why only one step matters.

Who To Interview, And In What Order

Five to seven people, one process, in this order. The order is the method.

1. The person who does the work. Most hours, least documented.
2. The person who approves it. They see the queue and why things come back.
3. The person who receives it next. They know what arrives broken and what they silently fix.
4. The customer-facing person who handles failures. They know the real failure rate, not the recorded one.
5. The manager accountable for the outcome. Last. Always last.

Interview leadership first and you get the documented process, and every interview after becomes a comparison against a version of reality that does not exist. Go the other way and the manager's description is evidence, not the frame.

This is old practice: the lean discipline of going to the gemba, the actual place where value-creating work occurs, summarised by Toyota chairman Fujio Cho as "Go see, ask why, show respect" (Lean Enterprise Institute).

The Fourteen Questions

Four blocks, in order. Inside a block, move freely.

Block 1: trace one real job, not a typical one.

1. Walk me through the last one you finished. Not a typical one, the actual last one.
2. Where did it come from, and how did you know it had arrived?
3. What did you have to open to do it? Name every tab, file, and system.
4. What did you have to go and find that was not already in it?

Block 2: the work between the steps.

5. Between it arriving and you finishing, where did it sit, and who was it sitting with?
6. Who did you have to ask something, and how did you ask them?
7. What did you have to decide yourself, with no rule to follow?
8. What did you type in twice?

Block 3: the exception path.

9. What do you do when the normal way does not work?
10. How often is that? Last week, how many?
11. Who do you go to when it is stuck, and what happens the week they are away?

Block 4: rework and private systems.

12. What comes back to you after you thought you were done, and why?
13. What do you keep in a spreadsheet or a notebook that is not in any system?
14. If I removed one thing from your week, what would you pick, and what would break if I did?

Block 1 exists to stop the person generalising. There is a formal method behind it: the critical incident technique, introduced by the psychologist John Flanagan in 1954, which collects specific recalled events rather than general opinions, because general questions produce summaries (Nielsen Norman Group). "How does this usually work" gets the process document read back. "Walk me through the last one" gets the two hours spent chasing a missing number.

The One Question That Finds The Most Hours

Question 9. What do you do when the normal way does not work?

Every documented process describes the good path. Nobody documents the other one, because it was invented under time pressure and never written down. The exception path is unmeasured, unowned, and usually performed by the most senior person available, which makes it the most expensive work in the building.

Then ask question 10, the count, which turns an anecdote into a number. "Sometimes we call the supplier directly" is a story. "Eleven times last week, about forty minutes each, and it is always me" is roughly seven hours a week of one salary, and now it can be ranked.

The Completeness Test: When A Step Is Actually Captured

Here is the standard we hold ourselves to, published so you can hold us to it. A step is not captured until five things are on record. Four out of five is a guess with good manners.

The breakpoint row is the one people skip, and skipping it is how a mapping exercise produces a clean diagram of a company nobody works at. A step with no breakpoint on record has not been captured. It has been imagined.

Round one usually ends with half the steps complete, and the incomplete half tells you where to go back. A test everything passes is not a test.

When The Interviews And The Systems Disagree

They will disagree. The disagreement is the finding, not a problem. Four rules:

Systems win on timing. People win on what and why. A timestamp beats memory on when something moved. No system tells you why it sat six days, and the why is what you can change.

If the system says four days and people say two weeks, the clock started before the record did. The most common gap by far. Work arrives somewhere unofficial first, then gets entered later. Everything before that entry is invisible, and it is usually the largest block of elapsed time.

If people describe a step and the system has no trace of it, the step is real. Count it. Absence of a record is not absence of work. That is the hidden factory by definition.

If the system shows a step nobody mentioned, treat it as suspect. Either it is automatic and genuinely invisible, or it is a ghost: an approval nobody reads, a field nobody uses. Ghosts are free capacity.

Once numbers and interviews agree, you can cost each gap, which is the revenue leak audit.

Where This Breaks: The Person Who Describes The Process They Wish They Had

The most reliable failure here has a name in safety research. Erik Hollnagel's distinction between work-as-imagined and work-as-done separates what managers believe happens from what actually happens under real pressure, and a gap between the two is always present in complex systems (NHS England / Hollnagel, Pennsylvania Patient Safety Advisory).

It shows up four ways, all recoverable if you expect them.

You get the process as designed, not as run. The person answers from the training document, or gives a duration that is really a hope. Refuse the general question. Go back to "the last one" and the count from last week.

The interview reads as an audit, so the workaround stays hidden. People conceal the shortcut because it is technically against policy. Say in the first minute that the workarounds are what you want, that they mean the process is wrong rather than the person, and that nothing here goes into a performance review. Then behave accordingly, because word travels between interviews.

One person's exceptions become everyone's process. Fourteen good answers from one confident person is a portrait of one person. Treat anything only one person mentions as unconfirmed.

You capture the hidden factory, then automate it as-is. The expensive mistake. Some of what you find should be deleted, not automated. Ask of every hidden step: does this exist because a customer needs it, or because something upstream is broken?

Turning Fourteen Transcripts Into A Ranked Map

Transcripts are not a deliverable. Nobody acts on a folder of notes. Four moves:

1. One row per hidden step, deduplicated. A step described by three people is one row with three sources. Independent mentions are your confidence score.
2. Frequency times minutes times people. Every row gets an hours-per-month figure from question 10. No count, no row.
3. Attach the system evidence, and mark where there is none. Interview-only rows are valid, but label them, because whoever defends the status quo will attack the weakest row to dismiss the rest.
4. Rank by hours and money recovered, then name the one constraint. Not by how annoying an item feels or who complained hardest. Find the single step everything else waits on, because that is the only one whose improvement changes the total.

The output is one page: ranked hidden steps, hours attached, evidence named, one circled constraint, and what happens to it on Monday. Same discipline as drawing the map, applied to the steps the map could not see. After an acquisition it runs on a clock: the first 100 days.

Frequently Asked Questions

Can I send the fourteen questions as a survey?

No, and it is the shortcut that reliably kills the exercise. Every question in Block 1 depends on a follow-up you cannot script, because the hidden step is usually in the answer after the answer. A survey collects the process as imagined. Faster, and worthless.

What if leadership insists the documented process is accurate?

Show them one completed row where all five fields are on record and the step appears in no document. One is enough. The argument stops being about whether the hidden factory exists and becomes about how big it is. That is the conversation you wanted.

---

Most companies already suspect where their time goes. What they lack is a ranked page they can defend in a leadership meeting, built from their own numbers and their own people rather than an outsider's best-practice slides. That is what a bottleneck diagnosis produces: every hidden step counted, evidenced, costed, ordered, ending with the one thing to fix first. Here is what we install.