The Businesses That Got Safer
The Businesses That Got Safer
Every bit of the rise in English and Welsh insolvency over a decade came from companies past their fourth year. Nothing in the economics explains that.
Something in the official insolvency statistics does not fit the story being told about them.
The story is familiar and largely correct: energy costs, interest rates after 2022, employer National Insurance, HMRC returning to collection after years of forbearance, Bounce Back Loan repayments falling due on companies that borrowed to survive something else. Business insolvency rates in England and Wales were 116 per 10,000 in 2025, up from 88 in 2019.
That is a real deterioration, and it has real causes.
But the Insolvency Service publishes the same figures broken down by the age of the business, and a sentence in the 2015-to-2025 release is one nobody seems to have picked up.
“The increase in the overall insolvency rate since 2015 was driven by businesses more than four years old.”
All of it. Every bit of the decade-long rise is accounted for by companies past their fourth year.
And the same passage continues: businesses aged seven to ten saw the largest increase of any age band, 88 insolvencies per 10,000 higher than in 2015. The largest decrease, the only group that improved, was businesses aged two to four years, down 9 per 10,000.
Young businesses got safer while everything else got worse.
The cross-section says the same thing
The single-year figures point in the same direction, and they are more startling than the trend.
In 2025, businesses less than two years old had the lowest insolvency rate of any age band: 37 per 10,000. The all-business rate was 116. The highest, at 170 per 10,000, belonged to businesses aged seven to ten.
Before 2025, the peak was four to seven years, every year from 2015 to 2024.
So the youngest companies, least capitalised, least experienced, no track record, no banking relationship, no reserves, fail least often, by a factor of more than four. Companies several years in, past the point of proving anything, with customers and revenue and history behind them, fail most.
And two other variables agree
This is where the pattern becomes difficult to dismiss as an artefact of how age is measured.
By headcount, the highest insolvency rate in 2025 was for businesses with 20 to 49 employees: 269 per 10,000. Businesses with 0 to 1 employee had the lowest, at 82. Medium-sized businesses – 50 to 249 people – came second at 220.
By turnover, the highest rate was for businesses between £2 million and £5 million: 189 per 10,000. Businesses under £250,000 were below average, as were businesses over £50 million.
Three independent measures: age, headcount, revenue, and all three peak in the same place. Not the smallest, not the largest. The band in the middle: a few years in, twenty to fifty people, a few million in revenue.
What the economics cannot account for
Energy prices do not discriminate by company age. Interest rates do not. An employer National Insurance increase hits a two-year-old business and an eight-year-old business the same way. HMRC does not enforce more aggressively against companies with thirty employees than against companies with three.
Every pressure driving the insolvency level is distributed evenly across these bands. The failures are not.
Which means the conditions determine how many companies come under pressure, and something else determines which of them break. That second thing is what the demographic data points to, and it isn’t visible in any of the explanations usually offered.
Coherence by accident, and what replaces it
A company in its first two years is coherent without anyone designing it to be.
Everybody knows what everybody else is doing, because there are six of them. Decisions get made by whoever is nearest the problem, and a wrong one is corrected by lunchtime. Nobody holds information somebody else needs, because it has nowhere to sit. Authority is undefined, and it does not matter, because the founder is in the room.
That is not good architecture. It is the absence of any need for one, and it is why the youngest companies fail least. Not resilience, not energy, not the founder’s commitment. There is simply no gap yet between what the organisation needs to decide and what it is arranged to decide.
Then it grows. Thirty people, four functions, two sites, customers expecting consistency. And none of the arrangements that produced coherence at six survive, because they were never arrangements. They were a property of smallness.
What replaces them is usually nothing. The structure isn’t redesigned; it’s extended. More people are added to a shape built for fewer. Decisions that used to be made where the information sat now route to someone senior who no longer has it.
Approvals accumulate one at a time, each introduced for a reason and none ever removed because adding a control is defensible and removing one is exposed. Accountability lands on people who have not been given the authority to discharge it, and they escalate everything, which is the correct response to answering for something you cannot determine.
Nobody decides any of this. Each step is reasonable. The aggregate is a company whose structure answers a question it stopped being asked three years ago.
Why it doesn’t look like a problem
Here is what makes that band dangerous rather than merely untidy.
An organisation in this condition performs adequately. It performs adequately because capable people absorb the difference: staying late, chasing what shouldn’t need chasing, holding together connections the structure no longer makes, carrying in their heads the coordination that used to happen by proximity.
That absorption is invisible, and invisible for a structural reason rather than an accidental one. It appears on no report. It has no budget line. The person doing it is rewarded for it, because the thing gets done, and the only visible fact is that it got done.
So the founder, who can still see most of the organisation, experiences a company that works. The board sees results. Absorption does exactly what absorption does: it converts a structural fault into personal effort, and personal effort doesn’t show up anywhere a structural fault would.
The gap between what the organisation can decide and what it needs to decide therefore grows, unobserved, and by design, for several years.
What the capacity actually is
Consider what a company must do when an input cost rises thirty per cent. Reprice. Renegotiate supply. Cut what can be cut. Decide which customers to keep at lower margin and which to let go.
Each of those is a decision. Response speed is how quickly decisions are made and implemented, and nothing else.
Where decision rights are clear, that is weeks. Someone is authorised to decide, knows they are, and holds the information needed to decide well.
Where authority has drifted, each decision travels. It goes up because nobody lower is sure they can make it, and being wrong about that is expensive. It waits for someone managing four other things. It returns with a question. The information needed to answer it sits with somebody who wasn’t asked, because the escalation route and the information route are different. Three months to decide something that needed deciding in three weeks.
The conditions did not cause that failure. They removed a margin for error that the incoherence had been consuming for years.
Which is the thing worth stating plainly:
A company’s capacity to survive adversity is not a quality of its people or its product. It is a property of how quickly it can decide and act, and that is designed, whether or not anyone designed it.
I want to be careful, because this is an explanation fitted to a pattern rather than a finding established by research.
The data shows where insolvency concentrates. It does not say why. The insolvency profession’s account is that cash flow is the proximate cause in almost every case, and that is right: cash flow is what runs out. The question is what made a particular company less able to protect it.
So the claim is narrower than it may appear. Not that architecture causes insolvency. Conditions cause insolvency, and for three years the conditions have been severe.
The claim is that architecture determines how much adversity a company can absorb before conditions become fatal and that the published rates are consistent with that, because the band where structure has fallen furthest behind complexity is the band that fails most, on three separate measures.
Consistent with. Not proven by.
One movement not to read too much into
The age peak shifted in 2025, for the first time in a decade, from four-to-seven years to seven-to-ten.
The Insolvency Service offers its own explanation, and I think it is the right one: businesses aged four to seven in 2025 were formed between 2018 and 2021, so that cohort’s composition may itself be a pandemic effect. Fewer formations, different formations, and a band with unusual membership.
Worth saying, because reading structural meaning into every movement is how a correlation becomes an explanation.
What follows
Three things, and none of them is a program.
Growth is the risk, not the reward.
The period in which a company is most likely to fail is the period in which it is succeeding: adding customers, staff, sites, complexity. Nothing about that feels like danger, which is precisely why the structural lag goes unexamined. A struggling business looks at itself. A growing business does not, because the evidence says it is working.
The absorption has to be made visible before anyone will act on it.
As long as structural incoherence presents as effort rather than as a fault, there is nothing to act on. What makes it visible is not a survey. It is the question of how long decisions take and who makes them, because that converts an invisible cost into a measurable one.
And that is measurable now, by anyone, without assistance.
Take three decisions from the last quarter that mattered. Trace each: who raised it, where it went, how many people touched it, how long it took, and whether the person who finally decided held more information about it than the person who raised it.
If the answer to that last question is no – and in this band it usually is – the organisation has a routing problem rather than a judgement problem. It will show up as slowness, exactly when slowness is expensive.
In conditions like these, that’s the only margin that matters.