A dashboard is usually discussed as though its content were the whole of it. The audience matters at least as much. The same measurement, shown to three different rooms, becomes three different objects and produces three different behaviours in the people it describes.
Diagnosis, judgement, promise
Shown to the team that owns the system, a number is a diagnosis. The question it prompts is what to change. Anomalies are interesting, and nobody is embarrassed by them.
Shown to management, the same number is a judgement. The question becomes who to ask about it. That is not illegitimate, and it changes the incentives of everyone who can influence the number.
Shown outside the organisation, it becomes a promise. The question is whether to trust you, and the reasonable response internally is caution about what gets published and when.
The data did not change. What it is for did, and what it is for determines what people do about it.
The predictable consequence
Once a measurement is used to judge, it will be improved directly. This is the mechanism set out in the article on proxies, and the audience is what triggers it.
The infrastructure version is familiar. A reliability figure used internally to decide where to spend effort is useful. The same figure used to compare teams produces fewer deployments, narrower definitions of what counts as an incident, and longer arguments about whether a degradation was really an outage. Nobody involved is behaving badly. They are responding to what the number is now for.
The workforce version is sharper because the subject is a person. An activity figure a person can see about themselves is feedback. The same figure on a manager's screen, next to nine colleagues, is a ranking, and the rational response is to produce activity.
Who can look is not who should
Most monitoring products default to broad internal access, and most organisations never revisit it. The result is that individual-level detail is available to anyone with an administrative login, including people with no role in the decision the data was collected for.
Two questions settle it, and both belong in the configuration rather than in a policy document. Which roles can see data about a named individual, as opposed to an aggregate. And whether that access is logged, so that looking is itself observable. A system where lookups leave no trace is a system where curiosity is free.
Aggregation answers most questions and exposes far less
Process questions are almost always aggregate questions. Where does this department's time go, which service is the slow one, which stage of a workflow accumulates delay: all of them are answered at the level of a group, and none of them need a name attached.
Individual-level data, conversely, answers very little about a process and a great deal about a person. Defaulting dashboards to aggregates, and requiring a specific reason to drill into an individual, is the single design decision that most reduces both the privacy exposure and the number of wrong conclusions.
Build the views deliberately
Gating one dashboard by role produces a manager looking at a diagnostic view without the context that made it diagnostic. A latency distribution means one thing to the engineer who knows a batch job runs at that hour and something else to a reader who does not.
The alternative is to design each audience its own view: what it shows, what it deliberately omits, and what decision it exists to support. Views that cannot answer that last question are the ones nobody opens, and a dashboard nobody opens is not harmless, because it is still collecting and still retaining.
Telling people is part of the design
In the workforce market, whether the subject can see their own data is a design decision with legal weight in many jurisdictions and cultural weight everywhere. A person who can see what is recorded about them can correct it, which improves the data, and can object to it, which is usually cheaper for the organisation than discovering the objection later. The jurisdictional detail is in the workforce section.
What we cannot verify
The behavioural effects described here are supported by long-standing work on measurement and incentives rather than by any measurement of our own, and the literature is argued over. Access control capabilities differ between products and are described by their vendors. Whether a given dashboard changes behaviour in a given organisation is answerable only by that organisation, and the honest way to find out is to ask the people on it.
The short version
- The same number is a diagnosis, a judgement or a promise depending on the room.
- Whatever the audience judges will be optimised directly.
- Broad internal access is usually a default nobody revisited.
- Lookups that leave no trace make curiosity free.
- Process questions are aggregate questions and need no name attached.
- Design a view per audience, including what it deliberately omits.