From 2011 the UK's Government Digital Service required central approval for departmental digital spending above a threshold (about £100,000, and zero for some categories such as public-facing websites) alongside a published service standard and assessments. By 2026 the GDS-run controls had been retired in favour of departmental ownership with a much higher pipeline threshold. What did controlling the money achieve that a standards document could not, and where would copying it be a mistake?

The situation they were in

Central government had no authority to tell departments how to build systems. Departments are separate organisations with their own ministers, budgets and accounting officers, and a central team publishing guidance had exactly the influence that guidance has, which is very little against a signed supplier contract.

What the centre did control was approval of spend. That is a different kind of authority: it does not require agreement about architecture, only a point in time at which money cannot move without a conversation.

What they chose

The combination matters more than either half:

• A threshold that catches the decision early, before a large procurement is committed rather than after. A control that bites at contract signature is too late to change anything.
• A zero threshold for specific categories — public-facing websites, identity for citizens — where the centre had a strong view and the cost of divergence was high.
• A published standard and assessments, so the conversation at the control point had criteria rather than opinions, and teams knew in advance what would be asked.
• Common components (publishing, payments, notifications) that made the compliant answer cheaper than building it again, which is the paved-road half of the mechanism.

Why it fit their constraints

A pure standards document has no moment of contact. A pure spend control has contact and no criteria, so it degenerates into negotiation. Together, the control creates the conversation and the standard decides what the conversation is about, and the common components mean the answer is not merely "no".

It also put the centre in the path of the decisions that actually have long consequences — multi-year supplier commitments — rather than in the path of everyday engineering, which is where central review boards usually end up and where they are least welcome and least useful.

What it cost them

Delay and friction, on things that did not always deserve it. A spend control is a queue, and a queue in front of a department's delivery is politically expensive; the central team spends credibility every time it holds something up for a reason a department finds unpersuasive. The 2026 change — GDS-run controls retired, departments owning their own approvals, the pipeline threshold raised substantially — is the visible consequence of that accumulated cost, and of the judgement that departments had matured enough to own it.

The second cost is scope. Assessments that catch the design of a service do not catch its operation three years later, and a control at the funding point sees new spend rather than the estate that already exists.

Where copying it would be a mistake

• Where the centre has no money lever, which is most commercial organisations below the capital-approval line. Imposing an approval gate without either budget authority or a genuinely better alternative produces the routed-around review board.
• Without the components. Spend control plus a standard and nothing to use is a tax. The paved road is what converts the gate from an obstacle into a shortcut, and organisations copying this reliably copy the gate and not the road.
• At small scale. For a company with one delivery organisation, the control point is the planning conversation that already exists, and formalising it adds ceremony.
• As a permanent arrangement. The GDS example is most instructive for the ending: the control was a mechanism for a period in which capability did not exist locally, and holding it past that point converts a useful constraint into an obstacle. Design the exit condition when you introduce the gate, or someone will design it for you later and less kindly.

Common weak answers

• "Central approval fixes inconsistent architecture." Approval without an alternative fails the same way everywhere: teams route around it, or comply on paper. Roughly the entire value here came from pairing the gate with components that made compliance the cheap path.
• "Publish a standard with executive backing." A standard with no moment of contact changes nothing, which is precisely the situation the spend control was invented to escape.
• "Put the review earlier in the design process." Prefer the point where commitment becomes irreversible, which for large organisations is procurement rather than design. A design review can be revisited in an afternoon; a six-year supplier contract cannot.

What a strong answer adds

That this is the general form of architecture governance with real authority: find the decision point where a commitment becomes hard to reverse, and put the conversation there. In most organisations that point is not a design review, it is a procurement, a hiring plan, or a multi-year licence. An EA function that reviews designs and never sees the contract is governing the reversible decisions and missing the irreversible ones.