How to find out how a business actually runs
Sit in the office for a week and count. What to count, what it tells you, and why it beats the data room on the questions that decide the outcome.
You find out how a business really works by sitting in it for a week with a notebook and counting things. Not by reading the process documentation, which describes an intention, and not by asking the owner, who will describe the business as they designed it rather than as it currently behaves. The gap between those two descriptions and the actual one is where every operating decision you are about to make lives.
Why the documents cannot tell you
Every company of any size has some account of how it works: a folder of procedures, a diagram somebody drew during a certification audit, a description in a diligence pack. All of it is a snapshot of what somebody intended at a moment, and none of it has been updated for the four workarounds invented since.
The owner's version is more useful and still not the thing. They know the business better than anyone, and they will tell you what it does in the general case. What they cannot tell you, because nobody can, is what proportion of the week is consumed by the exceptions, because the exceptions feel like interruptions rather than like the job.
The technique is not novel. Manufacturing has used direct observation on the floor for decades, precisely because the map and the territory diverge. What is unusual is applying it to the office half of a service business, where the work is invisible, nothing is timed, and everyone assumes it is basically fine.
What to count
Counting rather than watching is the discipline that makes this work. Watching produces impressions and impressions confirm whatever you already believed. Counting produces numbers you can be surprised by.
Touches per job
How many separate times does a human handle one job from first contact to paid. Every phone call, every re-keying into another system, every message asking where something is. In an unimproved company the number is usually somewhere between eight and fifteen, and almost nobody guesses it correctly in advance.
Re-entry
How many times the same information gets typed into a second place. Address into the diary, then into the job sheet, then into the invoice. This is the single clearest signal of what can be removed, and it is entirely visible from a chair.
Waiting
Not how long work takes, but how long it sits. A quote that takes twenty minutes to produce and four days to leave the building is not a twenty minute problem. Log the timestamp on both ends of each stage.
Interruptions to the one person
Every time somebody has to ask a specific individual something only they can answer. Tally them by person. The tally is a map of your key-person dependency drawn in real data rather than in opinion.
Exceptions
What proportion of jobs go through the normal route versus needing a special arrangement. When exceptions are above a fifth, the standard process is not the process, and any system built for the documented version will fail in month two.
Skilled hours spent on unskilled work
The one that matters most. How much of a qualified person's week goes on scheduling, paperwork, chasing parts and finding information. This is the number that tells you what the operating upside is actually worth.
Six tallies, a week, and a notebook. No software, no consultant, no interviews. Skilled hours spent on unskilled work usually produces the largest surprise, and it is the figure that converts a general belief that things could be tidier into a specific case for changing something.
Since the whole method is a sheet of paper, here is the sheet. The right-hand column is not a benchmark from anywhere, it is the point at which I stop assuming the number is fine and go and look at what is behind it.
| Count | How to record it | What makes me look harder |
|---|---|---|
| Touches per job | One tally mark per human handling, on a card that travels with the job | Above about ten, or any job you cannot reconstruct afterwards |
| Re-entry | Count the systems the same address or job detail is typed into | More than two. Three or more is almost always removable this quarter |
| Waiting | Timestamp in and out of each stage, not duration of work | Waiting longer than the work itself at any single stage |
| Interruptions per person | A five-bar gate under each name, all week | One name carrying most of the marks, whoever it is |
| Exception rate | Normal route or special arrangement, one mark per job | Above a fifth, at which point the documented process is fiction |
| Skilled hours on unskilled work | Note the task each time a qualified person is not doing qualified work | Anything above a quarter of their week |
The last row is the one to convert into money afterwards, and it converts steeply. A quarter of a skilled week recovered even by half is not a marginal gain, it is roughly a seventeen percent lift in revenue per person with nobody hired and nobody working faster. That is why this row is worth a week of somebody's attention on its own.
Nobody in the business can tell you these numbers, because from inside it none of this is an anomaly. It is just Tuesday.
How to do it without changing what you are measuring
The observation problem is real. People behave differently when watched, particularly when the watcher is a new owner or a prospective one, and the version they perform is the tidy one.
- Say plainly what you are doing and that it is about the process, not about anybody's performance, then repeat it on day two when nobody believed you the first time
- Sit where the work happens rather than in an office, and stay long enough that people stop narrating for your benefit
- Count, do not comment. The moment you suggest an improvement, everybody starts showing you improvements
- Do a full week including the bad day, since a Tuesday tells you about Tuesdays and the exceptions cluster
- Ask what happened, never why, because why invites a defence and what invites a description
- Write the tally the same day, not from memory at the end of the week
Counting without commenting is the hardest of these. You will see something obviously wrong on the first morning and the urge to fix it is strong. Resist it for the week. Once you have started making suggestions you are no longer observing a business, you are observing a business that is trying to look good for the person making suggestions, and the remaining four days are worthless.
It also explains why the week has to be a week rather than two good days, which is the compromise people reach for when the diary is full. The performance wears off on a schedule:
Two days of observation therefore does not give you two fifths of the answer. It gives you the brochure, with a number attached to make it feel measured, which is worse than not counting at all because it is persuasive. If the week is genuinely impossible, take three days and treat only the third as evidence.
What it is good for, and what it is not
Two distinct uses, and they need different framing.
As an owner, this is how you decide what to change and in what order. The touch count and the skilled-hours number together tell you where the recoverable money is, and the interruption tally tells you which dependency to break first. It also gives you the baseline you will need later to prove that a change worked, which is the thing most improvement efforts lack and then argue about for two years.
As a prospective buyer, it answers the question that legal and financial diligence structurally cannot: does this business run, or does one person run it. Accounts show you what the company earned. A week in the office shows you whether it will keep earning it once the person who has been holding it together is gone. That is the question owner dependence is really about, and it is not in the data room.
What it does not do is tell you anything about demand, pricing power, customer concentration or the durability of the market. It is a read on the inside of the business only. Somebody could run a beautifully organised company in a dying sector, and a week of counting would come back clean.
What comes out of it
The output is not a report, it is a short list with numbers attached: the three places the most time disappears, the one person the business cannot do without, and the proportion of skilled hours going on work that does not need a skilled person.
From that, the first change picks itself, and it is almost never the one you would have chosen from the outside. The improvement that looked obvious in the diligence memo turns out to save forty minutes a week, while the thing nobody mentioned, usually a piece of information that gets re-keyed four times, turns out to consume a day. That inversion is the normal result rather than the exception, and it is the reason the week is worth spending.
It is also the honest test of the whole operating thesis. If you sit in a business for a week and cannot find substantial time going on work that should not require a person, then the upside you were counting on is not there, and you have learned that for the price of a week rather than the price of the company. What the upside looks like when it is there is in the product cannot be disrupted, the operations can, and how it fits into the wider screen is good company, bad ops. The conditions I want satisfied before spending a week anywhere are on the buy page.
The short version
- Documented procedures describe an intention and the owner describes a design. Neither is the business as it currently behaves.
- Count six things for a week: touches per job, re-entry of the same information, waiting time between stages, interruptions to each key person, the exception rate, and skilled hours spent on unskilled work.
- Count, do not comment. The first suggestion you make ends the observation and turns the rest of the week into a performance.
- The performance wears off on day three. Two days of watching is not two fifths of the answer, it is the brochure with a number attached, which is worse than not counting because it persuades.
- An exception rate above a fifth means the documented process is not the process, and any system built for it will fail quickly.
- It answers the one question diligence cannot: whether the business runs, or whether one person runs it.