Zero-Audience Report
also called Unopened Report, Report Audience Count
A published dashboard with no viewer in a defined window, which is the only decommissioning candidate that can be retired without an argument - and the cheapest available measure of BI sprawl.
An estate has 4000 published dashboards and a BI bill nobody can defend. The instinct is to govern creation: approvals before publishing, a template, a review board. That attacks the wrong end, because the reports already exist and the approval queue becomes the bottleneck everyone routes around.
Count opens per report over a trailing window, and the set with zero is the only group you can retire without negotiating, because the owner cannot name a reader either. In most estates that set is large — well over half of published reports is a routine finding — so retirement becomes an announcement rather than a debate.
Why it matters
Dashboard sprawl is discussed as a cost problem and is worse as a trust problem. Every unmaintained report is a live source of a number that disagrees with the governed one, and a reader who finds it cannot tell it was abandoned 18 months ago. Retiring the unviewed tail removes contradictory numbers, not just compute.
It is also the only sprawl metric that costs nothing to compute and nothing to argue about. Coverage and maturity scores require judgement; opens are a fact in the BI tool's audit log, and "nobody has opened this in 90 days, we retire it on the 30th unless you object" is a conversation that completes.
Implementation patterns
- Set the window from the slowest legitimate cadence, then double it. Monthly reporting means 90 days minimum; an annual pack needs an explicit exemption, not a longer window for everyone.
- Count interactive opens and scheduled deliveries separately. A report with no opens that feeds an email subscription or an API has real consumers and must not be a candidate. This exclusion is what stops the exercise being reversed after the first incident.
- Retire in stages: notice, then unpublish while keeping the definition, then delete after a stated period. Unpublishing is reversible; deletion is the point of no return.
- Pair it with refresh scheduling. Moving an unviewed report from hourly refresh to refresh-on-open removes nearly all its cost while keeping it working, which is where most of the saving is.
Industry example
Large BI estates converge on one shape: a few reports carry nearly all the viewing, a long tail carries almost none, and the tail grows monotonically because nothing in any BI product makes deletion anyone's job. Vendors added per-report usage telemetry to enterprise tiers over roughly 2018 to 2022 because customers could not answer "who uses this" during migrations. Those who then cut their estates did it with usage data and a notice period; those who asked owners found that owners defend reports they never open.
Failure scenarios
- A retired report fed an annual regulatory submission. Dormant by design, so the exemption list must exist before the first cull.
- Scheduled deliveries counted as zero audience, so an executive email stops arriving and the programme loses its mandate.
- Usage logs retained for less than the window, so every report reads zero and the cull is indiscriminate.
- Reports recreated immediately, because the governed metric set did not cover what people needed.
Trade-offs
| Choose | Gains | Pays |
|---|---|---|
| Retire on zero audience | Cheap; removes contradictory numbers | Misses dormant-by-design reports unless exempted |
| Refresh-on-open instead | Most of the saving, no deletion argument | Latency on first open; trust problem remains |
| Govern report creation | Prevents new sprawl | A bottleneck that is routed around; ignores the backlog |
When not to use it
In an estate small enough to read, do not instrument it. With 40 dashboards and 12 analysts a person can open the list and decide in an afternoon, and standing up usage telemetry, a notice workflow and an exemption register costs more than the estate does.
Do not use the metric where usage logs are retained for less than the window, because a zero meaning "not recorded" is indistinguishable from "not used". And do not treat it as a governance strategy: these counts say nothing about whether the surviving reports agree with each other, which needs a governed metric set.
Interview question
Q: You have 4000 dashboards, a large warehouse bill and a mandate to fix it in a quarter. What do you measure first, what do you do with the number, and what would make you stop?
What a strong answer covers: opens per report over a window set by the slowest legitimate cadence, with scheduled deliveries excluded; ranking by cost times refresh frequency rather than by count, because a few expensive reports dominate the bill; switching the unviewed tail to refresh-on-open first because it is reversible; staged retirement; and the stopping condition — if the tail reappears within months, the real problem is the governed metric set.
Quick check
Quiz: Which two categories of report must be excluded from a zero-audience cull, and what happens if they are not? Reports feeding scheduled deliveries or APIs, and reports used on an annual cadence. Including them stops an executive email or destroys a regulatory pack, and the programme loses its mandate.
Flashcard: Why is counting opens a better first move against sprawl than approving new reports? — Because the sprawl already exists, and an unopened report can be retired by announcement while a creation gate only adds a queue people route around.