One central bench, many companies

Some capabilities get genuinely cheaper when several companies share them, and some only look like they do. Telling the two apart before you build anything is the difference between a holding company that compounds and a head office that has to be paid for out of somebody else's margin. The test is not whether the function is similar across businesses. It is whether the work itself is identical.

The claim, and why it is usually wrong

The standard argument for a group is that five companies do not need five of everything. Five bookkeepers, five marketing arrangements, five software subscriptions, five people working the same thing out separately. Consolidate and you pay once.

That argument is correct roughly a third of the time and it is applied universally, which is how head offices end up costing more than they save. The reason it fails is that most functions in a small company are not one thing. They are a generic core wrapped in a large amount of local knowledge, and consolidation captures the core while destroying the wrapper.

So the useful question is narrower than "could this be shared". It is: if I did this once for all five companies, would the output be identical to what each of them needs, or would it need to be adjusted per company by somebody who knows that company. If the answer is the second, sharing it adds a coordination cost and removes the knowledge, twice over.

The test

Three questions, and a candidate has to pass all three rather than score well on average.

The second question is the one that does most of the work and gets asked least. A capability worth centralising is usually one that each business needs rarely and badly. A small company negotiates a lease every seven years, so it is bad at it, every time, forever. Do that across five companies and somebody at the centre has done it fifteen times and is genuinely good at it. That is a real gain and it is not available any other way.

Contrast with something each business does daily. Scheduling happens hundreds of times a week in each company, so each company is already competent at it, and the centre would be worse because it lacks the site knowledge. High frequency locally is a strong argument against centralising, and it is the opposite of the intuition that high volume means economies of scale.

What passes

1. The things each business does once in a decade

   Property leases, insurance renewals, financing, acquisitions, disputes. Rare locally, frequent centrally, and the expertise gap between somebody doing it for the fifteenth time and somebody doing it for the second is very large.

2. Systems, built once and installed

   The capture, quoting, dispatch and billing layer, configured once and rolled out rather than rebuilt in each company. This is the single largest one for an operating group and the reason the shared bench is worth having at all.

3. Buying identical inputs

   Genuinely identical, from the same supplier, on the same specification. Insurance, fuel, vehicles, finance, common consumables. Not trade-specific parts, where the local relationship and the availability matter more than the unit price.

4. Knowing what worked somewhere else

   The cheapest and most neglected. When one company solves something, the others should not solve it again from scratch. This needs almost no structure, just a habit and somebody whose job includes noticing.

5. Recruiting for the roles every business needs

   A pipeline for qualified technicians is worth building once, because each company hires too infrequently to be good at it and the shortage means the pipeline is the constraint.

Systems built once and installed change the economics most, and it is worth saying precisely why. Designing an operating layer properly is expensive, mostly in attention rather than money: somebody has to work out the sequence, migrate the records, and hold the line through the weeks when it is slower. Doing that once and installing it five times is a completely different proposition from doing it five times, and it is the closest thing to a genuine structural advantage a small group has.

The bench is not a head office. It is the place where an expensive thing gets built once and then arrives cheaply for the fifth company.

What fails, and looks like it should not

The list of plausible-sounding failures is longer and more important than the list above.

Run all of them through the three questions at once and the test does the sorting for you, which is more useful than the two lists, because the next proposal will not be on either.

One row is an honest exception and worth flagging rather than hiding. Buying identical inputs fails the frequency question and shares anyway, because the mechanism there is negotiating leverage rather than accumulated expertise. It is the only item on the list where volume rather than rarity is doing the work, which is why it is also the one most often used to justify centralising things that have neither.

• Customer service and dispatch, because both depend on knowledge of specific sites and specific people
• Pricing, which is entangled with local competition and individual relationships nobody at the centre can see
• Sales, since the reason these businesses win work is that a named person is known locally
• Trade-specific purchasing, where availability tomorrow beats a better unit price next week
• Day-to-day bookkeeping, as distinct from the reporting standard, because the queries are all local
• Anything that adds a reporting requirement without removing a task, which is the definition of overhead

The first is the one groups get wrong most expensively, usually with a shared call centre. It saves an obvious amount and it moves the first conversation a customer has away from anybody who knows their building, their equipment or their history. In a business where the entire proposition is that you know the site, that is not a cost saving, it is a product change.

That final line is the general test, and it is worth applying to every proposal. If the centre takes on something and the operating company still has to do the same work plus provide information upward, nothing was shared. A cost was added and given a name that sounds like efficiency.

How to build it, in the right order

The sequencing rule is that the bench should always lag the companies rather than lead them. Build capability in response to a problem two businesses have actually had, not in anticipation of a group you intend to assemble.

Concretely that means the first company gets no bench at all. The second one reveals which things you just did twice, and those are your candidates. By the third you know which of them genuinely transferred and which needed rebuilding anyway. A centre designed before the second acquisition is designed from theory, and theory in this area is consistently wrong about which functions are local.

It also means the bench should be uncomfortably small for longer than feels right. A centre with spare capacity finds work, and the work it finds is inside the operating companies, which is precisely the thing the conglomerate discount is measuring. Understaffing the centre is a structural constraint on your own future interference, and it holds better than a policy does.

One last observation on what this is really for. A shared operating layer is the mechanism by which the improvement you make to the first company becomes free for the fifth, and that compounding is the actual argument for owning several rather than one. Without it a group is five separate businesses with common shareholders and a coordination cost, which is the version the market has been discounting for thirty years. The improvement itself, and why it is available in these businesses at all, is the product cannot be disrupted, the operations can, and the discipline that decides which company gets the money next is where the cash goes after a good year.