An architecture group wants one metric on the company's executive dashboard, next to revenue and churn. Which one earns its place?
Show the full answer Hide the answer
The deciding property
The metric must be one the audience already uses to make decisions, denominated in their currency. Executives allocate money and set prices. Cost per transaction connects architecture to both: it moves when the architecture improves, it is comparable across periods and business lines, and it feeds directly into pricing and margin conversations that are already happening.
It also has the property that makes a metric survive: it can get worse. A number that only goes up is decoration. Cost per transaction rising after a launch is a genuine signal, and being the group that reports it builds the credibility the architecture function needs.
Why the other options fail
- Deployment frequency is an excellent engineering metric and an unreadable executive one. The audience cannot tell whether 40 deploys a day is good, and no decision they own changes because of it. Its right home is the engineering leadership dashboard, where it belongs next to change failure rate.
- p99 latency matters enormously and is the wrong altitude. It answers a question executives have not asked, and the honest translation — conversion impact of latency — is the metric that would earn the slot instead.
- Architecture decision records published measures activity, not outcome. It is the clearest case of the general failure: a metric the team controls directly and the business does not care about, which invites gaming and teaches the audience to ignore the dashboard.
What would flip the decision
| If the business's constraint is | Publish | Because |
|---|---|---|
| Growth, not margin | time from idea to customer, for a defined class of change | speed is the currency being spent |
| A reliability crisis with customers | error budget consumed against the commitment | it maps directly to contractual and reputational exposure |
| Regulatory exposure | proportion of the estate meeting the mandated control | the board is accountable for exactly this number |
| An acquisition or divestment | separability of the affected capability | it is the fact that decides deal cost |
What a strong answer adds
One metric plus the two-line explanation of what moves it. A number without a mechanism produces demands to make the number better, with no theory of how. "Cost per transaction fell 12% because the batch pipeline moved to spot capacity; it will rise next quarter when the second region comes online" is a sentence that teaches the audience how architecture behaves, which is the actual objective.
When not to publish anything
When the number would be unreliable for the first two quarters. A metric introduced to an executive audience and then corrected downward twice does more damage than absence. Instrument it privately, watch it for a quarter, and publish when you would defend it.