Radical decentralisation, and where it stops
The evidence says a corporate centre subtracts value. So the burden of proof sits on centralising, and the short list of exceptions is the whole doctrine.
The default assumption should be that a corporate centre makes a group of companies worth less, not more, because that is what the market has priced for thirty years. Which means every single thing the centre decides to control has to justify itself against that default, and the useful work is naming the short list that survives.
Start from the discount
The finance literature on this is old, large and consistent in one direction. Lang and Stulz in 1994 found that multi-segment firms carried lower valuations than a portfolio of comparable single-segment firms would. Berger and Ofek in 1995 put the mean discount at roughly 15 percent, with diversified firms worth something like 13 to 15 percent less than the sum of their parts, and attributed it to investment distortion, agency problems and the cross-subsidy of weaker segments.
That result has been replicated many times. It is the empirical starting point and it points the wrong way for anyone building a group.
The honest caveat matters and it is the more interesting part of the literature. Whether the discount is evidence of value being destroyed is genuinely contested. Campa and Kedia, and Villalonga, showed that firms which choose to diversify were frequently worth less to begin with, so a large part of the gap may be selection rather than causation. Companies in trouble diversify, and then they trade at a discount, and the arrow may run in either direction.
Either way the practical conclusion is the same, which is why the debate does not need resolving here. As a market fact, the discount is real. As a management instruction, it says the burden of proof sits on the centre, always, for every function it proposes to take on.
That burden is quantifiable, which makes it more useful than a principle. Put the measured discount on a group of five:
| Value | |
|---|---|
| Five companies, each worth 100 standalone | 500 |
| The same five, grouped, at the measured discount | 425 |
| What the centre has to create before it has added anything at all | 75, or 15 per company |
The last row is the number to hold a head office to. Not "does the centre do useful things", which it always does, but whether the useful things exceed 15 percent of each company's value plus whatever the centre costs to run. Most functions a centre proposes to take on do not clear that, and the ones that do are a short list rather than a department.
And the selection debate above does not soften it. If the discount is partly selection rather than causation, then some of that 75 was never yours to lose, which is a comfort about the past. It changes nothing about the decision in front of you, because you still have to be right about which functions genuinely share, and being wrong still costs the full amount.
Why cross-subsidy is the specific poison
Of the three mechanisms Berger and Ofek name, one is worth isolating because it is the one a small holding company will actually commit.
Cross-subsidy means a weak business gets funded out of a strong one's cash, on the strength of an argument about strategy, turnaround, or fairness. In a single company that money would have to come from a lender who would ask questions. Inside a group it comes from the centre, which is emotionally attached and has no comparable discipline.
The strong business gets less capital than it earned and the weak one gets more than it deserved, and it happens because both are in the same family rather than because either case was tested.
The defence is procedural and it is the reason a hurdle rate exists. Every business inside the group competes for capital against the same threshold and against the outside alternatives, and a business that cannot clear it does not get funded because it is one of ours. That discipline is the whole of allocation and it is the first thing to go once relationships form.
The list that survives
So what should a centre actually hold? Constellation Software is the most instructive published example, running aggressively decentralised with acquired teams keeping real autonomy, and Mark Leonard has said plainly that this autonomy is a large part of why owner-managers are willing to sell to them. Berkshire runs a similar doctrine at greater extremity.
What both keep at the centre is short, and it deserves naming rather than a gesture at culture.
Where the cash goes
All of it, without exception. Surplus cash leaves the operating company and gets allocated centrally against one threshold. This is the entire reason the structure exists, and a group where each business reinvests its own cash is not a holding company, it is a shared address.
Who runs each business
The hiring and removal of the person in charge, and how they are paid. Not the layer below, not the hiring plan, not the org chart. One appointment per company, and the incentives attached to it.
A small number of non-negotiables
Safety, legal compliance, how the numbers are reported and on what calendar, and anything that could damage the other companies. Short enough to fit on a page, absolute where it applies, and silent everywhere else.
Whether to buy or sell anything
Acquisitions and disposals are allocation decisions wearing operational clothing, and they belong with allocation. An operating manager proposing to buy a competitor is proposing to spend group capital.
That is four items. Everything else, pricing, hiring, suppliers, marketing, how the work gets scheduled, which software is used, stays with the people who know the customers. Not as a delegation that could be withdrawn, as a boundary.
The non-negotiables list is the one that quietly grows. A list of non-negotiables that starts at four and reaches fourteen over three years has become a head office by accretion, and nobody will have made that decision. It is worth counting the list annually and requiring anything new to displace something.
Where decentralisation genuinely stops
Taken as an absolute, autonomy fails, and a doctrine worth following names the failures rather than pretending they do not happen.
- When the numbers stop arriving on time or stop being reliable, because allocation cannot function on figures you do not trust
- When a business is going to lose money and the manager has not said so, since the concealment is the problem rather than the loss
- When something is unsafe or unlawful, where there is no local judgement to respect
- When one company's behaviour damages another's standing, because the shared name creates a shared exposure
- When the appointed manager turns out to be the wrong person, which is the centre's own decision and its own responsibility to correct
- When two companies are unknowingly competing for the same customer, which nobody local can see
The second is the one that determines whether the whole structure holds. A manager who reports a bad quarter early keeps their autonomy. A manager who conceals one has broken the only thing the arrangement actually requires of them, which is that the centre is not surprised. Bad results are survivable and surprises are not, and saying that out loud in advance is what makes it true.
The version that fails
The failure mode is not a dramatic power grab. It is a sequence of reasonable steps, each individually defensible.
Somebody notices two companies buying the same materials at different prices, so purchasing gets centralised. Then the reporting is inconsistent, so a group finance function appears. Then it seems wasteful for three businesses to run separate marketing, so that consolidates too. Each step saves money on paper. Collectively they have moved every decision further from the customer, added a layer that has to be paid for, and given the operating managers a reason to explain results by reference to a constraint imposed on them.
What makes this hard to resist is that the first two steps usually do work. Genuine shared purchasing on identical inputs is real money, and inconsistent reporting genuinely does prevent allocation. The problem is that success on those two is then used as the argument for the next five, and the next five are where the discount comes from. Which functions genuinely share and which only appear to is the subject of one central bench, many companies.
The other half of the discipline is size. A centre small enough to be unable to interfere is more reliable than a centre that has decided not to, because capacity creates its own justification. A head office with spare people will find work for them, and that work will be inside the operating companies. Keeping it small is not frugality, it is a structural constraint on your own future behaviour, and it is easier to hold than a policy. Why the structure exists at all, and what it changes about what you can buy, is in permanent capital versus the fund clock, and why consolidation so often fails regardless is in why roll-ups fail. What has to be true before I would buy anything is on the buy page.
The short version
- The burden of proof is a number. Five companies worth 100 each are 500 apart and about 425 together, so the centre must create 75, or 15 per company, before it has added anything. Most proposed functions do not clear that.
- Berger and Ofek put the conglomerate discount at roughly 15 percent, with diversified firms worth 13 to 15 percent less than the sum of their parts. The measured default is that a centre subtracts value.
- Whether that is destruction or selection is genuinely contested, and it does not change the instruction: the burden of proof sits on the centre for every function it takes on.
- Cross-subsidy is the specific poison, and a hurdle rate applied without exception is the defence.
- Four things belong at the centre: where the cash goes, who runs each business and how they are paid, a page of non-negotiables, and whether to buy or sell anything.
- Autonomy stops on declared conditions, and the one that matters is concealment. Bad results are survivable, surprises are not.
Questions I get on this
Do conglomerates destroy value?
What should a holding company control and what should it leave alone?
When should a holding company intervene in an operating business?
Diversification discount figures from Larry Lang and René Stulz (1994) and Philip Berger and Eli Ofek, "Diversification's effect on firm value" (1995). The selection-effect critique is José Manuel Campa and Simi Kedia, "Explaining the Diversification Discount", and Belén Villalonga, "Does Diversification Cause the Diversification Discount?". Decentralisation practice from Mark Leonard's shareholder letters for Constellation Software.