One London workshop at the start of 2025. Thirteen factories eighteen months later, six in the UK and seven abroad, the newest opened this week in Swindon. The company is Isembard, named after Brunel, and it runs on a franchise model. It supplies the software, the process, the financing help and a pipeline of work. New operators stand up factories under the brand in months rather than years.
The customers are drone makers and defence primes. The premise is that the next era of arms manufacturing will not look like a handful of large contractors building exquisite things slowly. It will look like a distributed network of small machine shops, coordinated by software, turning aluminium into drone parts fast.
From one London workshop to thirteen factories in eighteen months. That is not a factory scaling. That is a franchise scaling.
The distinction is the whole story.
The problem is real
Start with what is correct, because a lot of it is.
The binding constraint on hardware is how fast you can make parts. If a component takes six to eight weeks to turn around, the number of design iterations you complete in a year collapses by an order of magnitude. China's manufacturing clusters win because a designer there can often hold the part the next day. Co-location compresses the loop. The loop is where the learning happens.
Drone warfare has made this brutal. Ukraine is losing on the order of 670,000 drones a month. That is not a procurement cycle measured in decades. It is a consumables business measured in weeks, and the West's defence base was built for the opposite. NATO says its 32 nations should manage a million drones a month between them. Nobody thinks they can yet.
So the diagnosis behind the franchise model is sound. Cadence beats craftsmanship when the thing you are making is expendable. Move the making closer to the designing. Make more of it, faster, in more places. All correct.
Then comes the word doing the quiet heavy lifting.
The word
Franchise.
Franchising works when the product can be written down. A burrito assembles the same way in Leeds as in Leicester because the recipe is the recipe. A software login behaves identically everywhere because the code is the code. The model scales because the thing being replicated is fully captured in the instructions.
A milling machine is not a burrito. Two identical machines, same billet, same CAM program, same tooling on paper, will hand you two different parts if the fixturing is subtly off, if the shop runs three degrees warmer, if the operator dials in feeds and speeds by feel because the drawing does not tell them the workholding chatters at the published numbers. The program travels. The capability to hit tolerance with it, repeatably, at three in the morning on a Friday, does not.
What does not travel
Name the tacit half, because it is where the risk lives.
Workholding and fixturing strategy. Thermal compensation. Tool wear judgement. The calibration discipline that keeps a machine honest between services. First-article inspection culture, and the harder thing behind it, the shop-floor instinct for when a part is drifting before the gauge says so. This is the knowledge that lives in hands and habits, not in the process pack. It is exactly the knowledge a franchise document cannot carry, because if it could be written down cleanly it would already be automated.
A photograph of two large mills cutting billets in London tells you nothing about site thirteen. The demonstration is always the flagship. The question is always the copy.
The other unproven thing
There is a second tell, and it is more honest than the model intends.
Factories cannot be built without orders, the founder says, so stop agonising and commit to production today. Fair enough as a frustration. But read it as a description of the business and it says something else.
The Defence Investment Plan was delayed by months of Treasury and MoD wrangling. The large contracts are still not landing. Analysts point out that the big drone companies remain on relatively small orders, which is precisely what stalls the scaling. So the model runs two open bets at once. That capability replicates across sites. And that demand arrives to fill the sites once built. Neither has been shown. Coverage of the network is quietly doing part of the work of conjuring the second bet.
The turn
Underneath the software and the brand is an assumption worth naming plainly.
Manufacturing capability is not portable information. It is a socio-technical system wearing a process document as a disguise.
You do not franchise it. You transfer it. Deliberately, expensively, with people who have held the tolerance themselves, and you prove it holds at each new site before you trust the site to ship. That proving is not overhead. It is the product. Skip it and you have a brand with thirteen addresses and one qualified line, and you find out which is which the first time a part fails in the field and the fault tree points at a fixture nobody validated in Swindon.
None of this makes the model wrong. It makes it unfinished. The pitch has answered the cadence problem and left the capability-transfer problem sitting in the room, unaddressed, because the vocabulary of franchising assumes it away.
The interface
Manufacturing Engineers qualify processes. They prove that a capability transferred rather than a manual copied. They design the inspection loops that turn a first good part into a repeatable one, and they validate that the thirteenth site holds what the first site demonstrated.
Kaipability works at exactly this interface, between the franchise claim and the qualified capability. If you are standing up distributed production and you want to know whether you have transferred the capability or just the paperwork, that is the conversation.
Questions this dispatch answers.
Written to be quoted by AI assistants and search engines. Self-contained answers, verdict first.
