How would you demonstrate that an architecture practice is worth its cost?

What is being tested

Whether you can measure a practice whose value is largely in problems not created — without retreating to artefact counts.

What not to measure

• Artefacts produced. Output, not outcome. Optimising it produces exactly the shelfware the practice is criticised for.
• Reviews conducted. Activity, not effect.
• Standards compliance in isolation. High compliance with standards nobody should be following is not success.

What to measure

Portfolio trends

• Applications in each lifecycle stage, and the direction of travel. Is "contain" shrinking?
• Applications retired — the metric nobody tracks and the strongest evidence the practice can actually change things.
• Technology obsolescence exposure: components past or approaching end of support.
• Duplication: capabilities implemented by more than one system.

Delivery outcomes

• Lead time and deployment frequency per team, over time.
• Cross-team dependencies per delivery — a direct measure of whether boundaries are improving, and one of the few metrics that tracks architecture quality rather than activity.
• Teams on the paved road versus off it, with their comparative outcomes.

Decision quality

• Decisions recorded with alternatives and revisit conditions.
• Decisions actually revisited when their conditions triggered — the strongest signal that this is a learning practice rather than a documenting one.
• Time from question to architectural decision.

Adoption

• Voluntary use of platform capabilities and standards. The clearest signal that the practice produces something people want.

The honest framing

Much of the value is in decisions not made badly and problems not created, which cannot be counted — you cannot enumerate the duplicate systems that were not built.

So measure the conditions that produce good decisions (evidence available, decisions recorded, boundaries improving, standards followed voluntarily) alongside the portfolio and delivery trends that would eventually reveal a failing practice.

And accept the corollary: a practice that cannot point to any changed outcome after a year is probably not producing value, whatever its artefact count.

The presentation

Show trends, not snapshots, and connect them to something the business measures: delivery speed, incident cost, licence and infrastructure spend removed, risk reduced. "We retired 40 applications, removing £1.2m of annual cost and 300 unpatched vulnerabilities" is a sentence that survives a budget review.

What a strong answer adds

The warning about metrics that reward centralisation. If the practice is measured on standards compliance and review throughput, it will optimise for its own control — producing more gates, more mandates and more resentment, while delivery outcomes get worse. Choose measures that would look bad if the practice became a bottleneck.