The single-system fallacy

A central team had produced the official version of a form: legally correct, several pages long, written in the language the centre uses about itself. A local leader downloaded it, decided the people in front of them would never get past the first page and rebuilt it onto a single side of paper. Same legal purpose, plainer words, even some of the details already filled in.

They weren’t being difficult. They had tried the official version once, watched people give up early and drawn the obvious conclusion. Nobody could have made them use the long one.

There is a sentence I have heard in more than one boardroom, and it always sounds like the responsible thing to say. “Let’s just build one system that everyone uses.”

It sounds like efficiency, and like finally getting a grip on the mess. In the kind of organisation I spend my working life inside it is very often the wrong call, and the expense rarely shows up until a couple of years and a lot of goodwill have already gone.

That original idea is worth scrutinising. Shared systems aren’t the problem. The instinct behind “one system for everyone” assumes a kind of organisation that plenty of us don’t work in.

What a federated organisation is

A federated organisation is made up of parts that are legally or operationally independent, held together by a shared purpose rather than a single chain of command. The centre can persuade but cannot instruct. Scouting movements work this way. So does the National Trust across its regions, a co-operative, a trade union, a professional membership body and large parts of the public sector.

This is a more common shape than the label suggests. Most organisations built like this have no shared name for the pattern they are in, so each one tends to assume its situation is unique and works out the same hard lessons on its own.

Most change advice assumes a corporate hierarchy instead: head office decides, the branches comply and the rollout is a logistics problem. Put that assumption to work in a federation and one of two things tends to happen. Either you build the polished single system and discover that nobody was ever obliged to use it, so plenty don’t. Or you push hard for adoption, and the pushing itself, the sense of the centre reaching in and taking a decision away, hardens the local resistance you most needed to avoid. Either way the money has gone and the next proposal starts further back.

Both failures come from the same mistake, which is treating the federated structure as an obstacle to route around when it is the brief you have been given.

This is also the shape the AI conversation keeps taking in organisations like these. “Let’s pick one AI system and roll it out everywhere” is the same sentence with a different subject, and it deserves the same suspicion, for the same reasons.

A different question to ask

The more useful move is to swap “how do we get everyone onto one system” for “what is the smallest shared architecture that lets independent parts work together without asking them to become the same.” That’s a genuinely different brief. It means working out what truly needs to be common, and being honest that almost everything else can stay local.

In practice this tends to mean a few specific choices, not one big one:

• a small shared core where it counts: one identity system, a single agreed data standard
• a lot of local freedom around everything else
• standards offered as something that makes someone’s life easier, not a mandate that makes it harder
• a centre that earns adoption by being useful again and again, not by requiring it once and assuming the job is done.

The clearest public example I know of is the UK Scouts replacing Compass, their single membership system for adult volunteers. Ten counties and areas went first as an early adopter cohort from 2022, the rest of the UK was promised for early 2024, testing raised concerns about moving that much data, and ordinary counties went live county by county through November 2024. Judged against a corporate timetable that looks like failure. It only is if you thought the centre could instruct everyone at once.

The form on one side of paper is the same limit at the smallest possible scale, and it is what the single-system instinct keeps missing. In a federation, a rule the people at the sharp end do not follow does not become compliance because it exists on paper.

Where I’m less sure

I don’t think the federated frame holds forever. Some organisations look federated on the chart but behave as networks, where there isn’t even a centre to do the persuading, and the advice has to change again. Where that line sits is something I am still working out, mostly by writing about it in public.

What I’m fairly sure of is the first part. In a federation, “let’s just build one system everyone uses” deserves suspicion however efficient it sounds, and that includes when the system in question is an AI one.

So the question worth putting to any proposal that starts there is a narrower one: what has to be common here, and what has only been assumed to be?