Bus factor 1: when only one person knows how it really works
One employee is absent and half the company stands still. Why the bus factor is the most underrated metric in mid-sized companies, and how you reduce it.
In short
The bus factor says how many people must be absent for a process to stop. If the number is one, part of your company depends on a single person. In mid-sized companies that is the norm, not the exception. Introducing AI without knowing this number merely shifts the dependency from a person to a system that even fewer people understand.
Auf Deutsch lesen: deutsche Fassung
In almost every company there is one person whom everybody asks. Not the managing director. Not the one with the biggest office. The one who knows why the machine has to be set differently for this one material, which customer absolutely needs its delivery papers in duplicate, and where the file is that carries half the month-end closing. As long as this person is there, things run. They even run conspicuously well.
Then he goes on parental leave. Or he falls ill. Or he resigns because someone has poached him. And suddenly the company realises that an entire workflow was never written down anywhere, but lived in one head. I have seen this in audits in five industries, from aviation to precision engineering, and I see it in my own company. This is not a failure of leadership. It is the normal state when nobody asks about it.
The bus factor is not a metaphor but a number
The term comes from software development and sounds more cynical than it is meant. The question behind it is: how many people would have to be absent for this process to come to a standstill? If the answer is one, you have a bus factor of one. And it does not take a bus. Flu is enough. Sometimes a holiday is enough.
What I like about this metric: it is brutally simple. You need no maturity model and no consultancy to determine it. You go through your most important workflows and write a name behind each one. If the same name appears everywhere, you know where you stand. If there is no name at all behind a workflow, it is usually even worse, because then nobody knows that anyone is responsible.
The mistake I see most often is confusing a deputy with the bus factor. On paper almost everyone has a deputy. There is a second name in the holiday plan. Except that this second name has never done the task alone, only ever watched. A deputy who has never actually stepped in is not a deputy. It is a declaration of intent.
A deputy who has never done the task alone is not a deputy. It is a declaration of intent.
Where it really sits in the company
Most managing directors look for the bus factor in the wrong place. They think of the foreman or the head of sales, in other words people with a title. But the critical points are almost always one level lower, where someone has built a home-made solution over the years because the official system did not provide it.
| Where it sits | How you recognise it | What fails |
|---|---|---|
| The spreadsheet that grew over time | A file that has been maintained for years, with formulas nobody explains any more. | Costing, scheduling or tracking, often without it being noticed immediately. |
| The home-made helper | A script or macro that runs under the personal account of the person who built it. | The run stops, and nobody knows where it starts or what it touches. |
| The relationship with the supplier | A supplier who only talks to one particular person. | Escalations and special cases, because the connection is missing, not the contract. |
| The unwritten special knowledge | Sentences such as: with that customer we always do it differently. | Quality and on-time delivery, usually first on the awkward orders. |
| The only access | An administrator account whose password exactly one person knows. | In an incident, everything at once, and precisely when it is urgent. |
The last row is the one that interests me most as an auditor. An access that only one person has is not a security gain but an outage risk with a coat of security paint. I have seen a company need three weeks after a change of personnel to get back into its own systems. Not because anyone was malicious, but because nobody had ever written it down.
AI does not lower the bus factor. It moves it.
Now comes the part that is almost never discussed in AI projects. The usual story goes like this: we automate the task, then we no longer depend on that one person. That sounds logical and is still rarely true. Because the dependency does not disappear, it migrates.
Three examples from practice in which I have seen this:
- An AI in HR pre-selection pre-sorts applications. Before, three people in the HR department could explain why someone was sorted out. Afterwards exactly one can, namely the one who connected the model and set the thresholds. For a decision about people, that is the worst place for a bus factor of one.
- A vision inspection sorts parts by surface defects. The system runs, the scrap rate falls, everyone is satisfied. Except that after six months nobody knows any more which sample images were used for training and when it was last retrained. When the batch then changes and detection gets worse, there is no basis at all for troubleshooting.
- An AI is attached to the ERP and suggests order quantities. Purchasing adopts them because they are usually right. After a while nobody can say how a quantity comes about. The expertise used to be spread across several heads and now sits in a system that two people in the company understand.
In all three cases the company is more efficient after the introduction and at the same time more vulnerable. That is not an argument against AI. I build such solutions myself and consider them right. It is an argument for measuring the bus factor beforehand and again afterwards. Anyone who counts only the hours saved and not the newly created dependency is doing half the calculation.
Automation makes the company faster. Whether it also makes it more robust depends on how many people understand the solution.
What the standards require on this
The bus factor is not in any standard. The principle behind it is in several, just under other names, and that is practical for you: if you are heading towards a management system anyway, this work can be used twice.
In ISO/IEC 27001:2022 you find the topic in several places in Annex A. A.5.2 requires information security roles and responsibilities to be defined and allocated, in other words named in writing. A.6.5 governs what has to happen on termination or change of employment. A.8.2 requires privileged access rights to be restricted and managed, which hits one-man administration directly. And A.5.30 requires ICT readiness to be aligned with business continuity requirements.
ISO/IEC 42001:2023 adds the AI perspective. Clause 5.3 requires assigned roles and authorities, clause 7.2 the necessary competence of the people who operate the system. Annex A requires documented objectives for responsible use in A.9.3. The specific implementation guidance on this, including on human oversight, is in Annex B under B.9.3. The order matters: the standard first requires that someone is named, and then that this person is also able to do it.
This is my assessment as an auditor and managing director, not legal advice. Anyone who needs a binding assessment for their specific case should obtain one.
Four measures that really hold up in mid-sized companies
I think nothing of manuals that nobody reads. What works in practice is smaller and less spectacular:
- The deputy really steps in. Once a quarter the deputy does the task alone while the regular person is in the building and only helps if things come to a real standstill. That costs half a day and is the only test that proves anything. Everything else is a claim in the holiday plan.
- One page instead of a manual. One page per critical workflow: what does the task do, with what, which special cases are there, whom do you call when it gets stuck. Anyone who asks for twenty pages gets none. Anyone who asks for one gets one.
- Access belongs to the company, not to the person. Service accounts instead of personal accounts for everything that runs automatically. A deposited emergency access, sealed, with logged release. That is not a declaration of mistrust, it is the ability to operate.
- Build up knowledge under the four-eyes principle, not in an emergency. Two people in every critical workflow, from the start. On paper that costs productivity, and it saves it at the moment when it would otherwise collapse completely.
And one thing that costs nothing: in the next appraisal, do not only ask what someone does, but what would be left undone if they were away for four weeks. The answers are usually more precise than any process map, because people themselves know best where they are irreplaceable.
How I would start
Not with a survey across all departments. That peters out. I take the three workflows where a standstill hurts fastest, usually order intake, delivery and invoicing, and write down for each one who could carry it through alone today. Not who is responsible. Who could carry it through alone.
This list is uncomfortable, and that is exactly why it is useful. In most companies there is then only one name in at least one place, and everyone in the room knew it beforehand, nobody had said it out loud. From then on it is a management decision and no longer a coincidence.
The bus factor is such a good question because it does not look for blame. It only asks what happens when someone is not there. That is not a criticism of the person. It is the opposite: it is the recognition that they have become too important to be left alone.
Be honest: how many of your most important workflows have only one name behind them today?
Primary sources
Lowering the bus factor in your company, in six steps
Select critical workflows
Take the three to five workflows where a standstill costs money or customers fastest. Not all processes at once, otherwise the survey peters out in a map that nobody uses.
Write names behind them
For each workflow, note who could carry it through alone today. What counts is the could, not the formal responsibility. If there is only one name, you have found the bus factor of one.
Track down home-made solutions
Ask specifically about spreadsheets that have grown over time, macros and scripts, and about which account they run under. Personal accounts for automatic runs are the most common hidden single risk in mid-sized companies.
Write one page per workflow
Have the responsible person write one page themselves: purpose, tools, special cases, contact persons for faults. One page, not twenty. Short gets written and read, long gets neither.
Let the deputy really step in
Once a quarter the deputy does the task alone while the regular person is reachable and only intervenes if things come to a real standstill. Whatever gets stuck belongs on the page from step four.
Detach access from the person
Switch automatic runs to service accounts and deposit a sealed emergency access with logged release. Check once a year whether it still works, not only in an incident.
Frequently asked questions
What exactly does bus factor mean?+
The bus factor indicates how many people would have to be absent for a process to come to a standstill. A bus factor of one means: if this one person is absent, the workflow stops. The term comes from software development, but the principle applies to every operational workflow, from order intake to the maintenance of a machine.
How do I determine the bus factor without a big project?+
Take the three workflows whose standstill hurts fastest, and write behind each the names of the people who could carry it through alone today. Not who is responsible, but who actually could. That takes an hour and shows you more than most process surveys.
Does automation lower the bus factor?+
Not automatically. Often the dependency merely moves from the clerk to the one person who built the solution and is the only one who understands it. So recalculate after every automation: how many people in the company could operate, check and adapt the solution in an incident?
Is the bus factor a topic in an ISO audit?+
Not under that name, but in substance yes. ISO/IEC 27001:2022 requires allocated roles and responsibilities in A.5.2, managed privileged access rights in A.8.2 and ICT readiness aligned with business continuity in A.5.30. ISO/IEC 42001:2023 requires named roles and the matching competence behind them in clauses 5.3 and 7.2.
Author & expert review: Lars Zimmermann · ISO/IEC 42001 Senior Lead Auditor & Senior Lead Implementer · ISO/IEC 27001 Lead Auditor & Lead Implementer (PECB)
Last updated: 05 August 2026. Researched and reviewed to the best of our knowledge; not a substitute for individual legal advice.
Questions about your own case?
In a free 15-minute intro call we assess where you stand on ISO 42001, ISO 27001 and the EU AI Act, honestly and without a sales pitch.