When Your Best Operator Leaves, What Leaves With Them?

There is a version of this that every founder recognises. Someone senior asks for a quiet word, and by the end of the afternoon you are working out who covers what. The formal machinery starts up straight away. A handover document gets scheduled, the files are all on the shared drive, the CRM is up to date, and on paper the position looks manageable.

Then they go, and for the next six months the work that used to be fine is not fine. Quotes take longer and land less often. Enquiries that would have been dismissed in a sentence now get chased for a fortnight. A supplier problem that used to get solved with one phone call becomes a three-day thing. Nothing broke, exactly. It all just got worse in a way that is difficult to point at.

The documents stay. The judgement goes.

The reason the handover felt adequate is that it was measuring the wrong thing. Almost everything a departing person can hand over is the part that was already written down somewhere, and the part that was already written down somewhere was never the part you were relying on.

What you were relying on was judgement, and judgement is not a document. It is a long series of small decisions made repeatedly under conditions the process never covered: which quotes to price aggressively because the client will come back with volume and which to price properly because they will not, which enquiries are real and which are somebody filling in a form, which supplier to call when the first one lets you down and which to avoid even though their price looks better. Whether a particular customer means it when they say the deadline is fixed sits in the same category, and somebody in your business knows the answer without ever having been taught it.

None of those decisions is written anywhere, because none of them ever felt like a decision. They felt like doing the job. That is precisely what makes them invisible until the person making them stops turning up.

This is a different problem from the general drift of a business losing track of what it knows. The aggregate cost of a business forgetting accumulates slowly across everybody, and you can live with it for years. A key person leaving is the concentrated version: one person, one notice period, and a clock that started the moment they told you.

Replacing them takes longer than the notice period allows

The instinctive response is to recruit, and the recruitment market is where the plan usually meets its first real obstacle.

The Department for Education's Employer Skills Survey covers 22,712 UK employers, and its finding on this is blunt. Just over a quarter (27%) of all vacancies in 2024 were skill-shortage vacancies, down from over a third (36%) in 2022, but higher than in 2017 (22%). A skill-shortage vacancy is one that is hard to fill because the applicants do not have the skills, qualifications or experience. So a little over one in four of the roles being advertised in this country are not straightforwardly fillable, and the roles that carry the most accumulated judgement are disproportionately in that group.

The same survey puts a number on the other half of the position. In 2024, 12% of employers reported skills gaps (at least one member of staff judged to lack full proficiency). Even where the recruitment works, the person arriving is frequently not yet able to do what the person leaving could do, and that gap is closed by time rather than by hiring.

Put those together and the honest timeline is not a notice period. It is a notice period, plus a recruitment process, plus however long it takes somebody new to rebuild a judgement that took years to accumulate, in a business that will not be pausing while they do it.

Buying a business rather than losing a person compresses the same problem into one event with a deadline attached, which is what an acquisition quietly fails to transfer.

The trap that makes it worse

There is a second-order effect here that is worth naming, because it quietly deepens the dependency rather than easing it.

Skills England's 2026 annual report describes a pattern specific to smaller employers. For firms with 10 to 50 employees in particular, it notes that limited promotion routes can make the link between investing in workforce skills, staff progression, and business performance less visible, and that this, coupled with concerns about trained employees moving on, can mean smaller business owners are less willing to invest in the skills of their staff.

Read that carefully, because it describes a loop. The fear that trained people will leave makes owners less willing to develop people. Fewer developed people means the capability stays concentrated in whoever already has it. And a capability concentrated in one person is exactly the exposure that made the fear reasonable in the first place.

The national picture bears it out. Total employer expenditure on training and development over the previous 12 months was £53.0 billion in the same Employer Skills Survey, which represents a 10.2% decrease in real terms on the 2022 figure of £59.0 billion. Businesses are buying capability back less than they used to, at a point when replacing it from outside has got harder rather than easier.

Your automations already contain their judgement

Here is the part that has changed in the last few years, and it is the part most handover conversations still miss entirely.

If your business has automated anything meaningful, somebody made a set of judgement calls to build it: what counts as a qualified enquiry, which cases route to a person and which get handled without one, where the thresholds sit and why those numbers rather than slightly different ones. Every one of those calls is now living in your business as configuration, and configuration records what was decided while saying nothing at all about why.

We built an RFQ workflow for a commercial flooring business where handling time for an inbound enquiry dropped from around 20 minutes to under 90 seconds, giving back more than 12 hours of capacity a week. That system works because the extraction and routing rules encode how the people who had been doing it manually actually thought about an enquiry. The hours came from the automation. The accuracy came from the judgement that was poured into it.

Which means the automation is a genuine asset and a genuine liability at once. It holds the judgement, so the judgement survives the departure. It holds it in a form nobody can read, so the moment conditions change, no one can say whether the rule still makes sense. In practice teams do the rational thing and leave it alone, and a rule nobody dares to touch slowly stops matching the business it was built for. This is the same failure that stalls AI projects generally, where the system is only ever as good as the direction it was given and the direction was never recorded.

What reduces the exposure

The useful answer is not a better handover template, and it is not a document repository that nobody opens.

It is that the reasoning behind repeated decisions should exist somewhere current, findable and separate from the person who holds it. Not the process, which was probably documented already and was never the thing at risk, but the reasons behind it: why the pricing works the way it does, what makes an enquiry worth pursuing, which conditions turn a standing rule into an exception. That layer is what an intelligence system actually holds, and it is the difference between a business that can absorb a departure and one that has to relearn its own operating decisions from scratch.

Senior leaders have broadly worked out that this is where the constraint sits. IBM put the question to 2,000 chief executives. Surveyed CEOs say roughly one-third (31%) of the workforce will require retraining and/or reskilling over the next three years, while 65% say their organization will use automation to address skill gaps. Automation only closes a skill gap if the thing being automated has been written down well enough for a system to act on it, which puts the writing before the tooling rather than after it.

The economics are usually kinder than the recruitment route as well. On one partner platform we built, automating the onboarding research handed back 25 hours a month against an alternative that was a £20,000-plus annual hire. The point is not that a system replaces a good operator, because it does not. It is that the capacity you free up buys you the time your remaining people need to develop, which is the thing the Skills England loop above says most smaller businesses never quite get round to funding.

None of this needs to start as a programme. It starts as writing down what the business knows, in a sensible order, beginning with the two or three areas where you already know the answer to the fortnight question is uncomfortable. And it is worth being clear-eyed that if that reasoning only ever accumulates inside the tools you rent, you do not really hold it either.

Practical takeaways

1. Run the fortnight test. For each significant part of your commercial operation, ask what happens if one named person is unreachable for two weeks. Where the answer is that things get done but not as well, you have found a dependency.
2. Write down reasons, not processes. The process was probably already documented and was never what you were relying on. Capture why repeated calls get made the way they do, and the conditions that make something an exception.
3. Treat every automation as an undocumented decision. Whoever built it made judgement calls that now exist only as configuration. Record the reasoning next to the workflow while the person who made those calls is still there to explain them.
4. Do not wait for a resignation to start. A notice period is a negotiation with a clock, and you only get answers to the questions somebody thinks to ask. This work is far cheaper done calmly.
5. Break the training loop. The fear that trained people leave is what keeps capability concentrated in one head. Freeing up capacity is what makes developing the rest of the team affordable.

Nobody plans for their best operator to leave, and no system stops it hurting. What a business can decide, well before the conversation happens, is whether the judgement that person built up over years exists anywhere other than in their head. That decision is made months in advance, quietly, or it is made in a fortnight under pressure.

Frequently asked questions

What is key person dependency?

It is the state where a specific part of your business only works properly because one particular person is available. Not a formal role or a documented process, but an accumulation of judgement that person built up over years and nobody ever wrote down. The dependency is invisible while they are there, because the work simply gets done and nobody has cause to ask how. It becomes visible in the fortnight after they resign, when the same work starts taking longer and coming out worse, and nobody can quite explain why.

How do you know if your business has a key person dependency?

Ask what happens to a specific piece of work if a specific person is unreachable for a fortnight. Not whether it stops, because it rarely stops, but how much slower and worse it gets. The tell is not that nobody else can do the task. It is that everyone else does it differently and less well, and nobody can say why the difference exists. If the honest answer for any part of your commercial operation is that things get done but the quality is not the same, that is a dependency, and it is worth naming now rather than during a notice period.

Can you capture an operator's judgement, not just their process?

Partly, and the part you can capture is the part that matters most. Process is easy to write down and rarely the thing you are missing. Judgement looks like a series of small decisions made repeatedly under conditions the process does not cover: which enquiries are worth pursuing, when to hold a price, which supplier to call when the first one lets you down. What makes those decisions capturable is that they have reasons, and the reasons are stable even when the specific cases are not. Recording the reasoning behind repeated calls, with the conditions that shaped them, gets you most of the way. It will not reproduce twenty years of instinct, but it means the next person starts from your business's accumulated position rather than from nothing.

What happens to our automations when the person who set them up leaves?

They keep running, which is the problem. An automation encodes a set of judgements: which enquiries count as qualified, what the routing rules are, which cases get escalated and which get handled quietly. Those decisions were made by someone who understood why, and they are now expressed as configuration rather than as reasoning. The workflow carries on behaving exactly as it was told to, and nobody notices anything is wrong until the rule stops matching reality. By then the person who could explain the rule has gone, and the choice is between leaving it alone because nobody dares change it and rebuilding it from guesswork.

What should we do before someone resigns?

Treat it as a standing job rather than an event you respond to. Pick the two or three parts of the business where the honest answer to the fortnight question is uncomfortable, and start writing down the reasoning behind the repeated calls made there: decisions with their reasons, standards with the thinking behind them, and the conditions under which something becomes an exception. It does not need a project or a system to begin with, it needs somebody to write down what they know while they still work for you. The moment a resignation arrives you are negotiating with a clock, and the answers you get in a notice period are the ones somebody thought to ask for.