G10 - If Your AI Dashboard Is Always Green, It's Measuring the Wrong Things.
AI dashboards fail in two ways: dishonestly green — thin data displayed as health, metrics recovering while causes stay open — and mutually contradictory, where the board pack says fine and the pager says fire. Both destroy the only asset a dashboard has: trust.
Where this gets hard
- Insufficient data gets rendered as green, because grey looks like an admission.
- Metrics recover while the underlying finding is still open — and the indicator flips back to green as if nothing happened.
- Different tools compute status differently, so the board chip and the operational view disagree. Then both get ignored.
- Averages hide the one red that matters inside a sea of healthy systems.
- A dashboard nobody has ever seen go red isn't reassurance; it's a measurement problem wearing a good outfit.
Where to start
- Compute status once, server-side, and let every surface — board pack, dashboard, pager, audit log — display that same verdict.
- Show honest grey for thin or stale data. Make ‘insufficient evidence’ a first-class state.
- Bind indicators to findings: no return to green while the finding is open, whatever the metric does.
- Never average across systems — one red is red. Escalate the worst state, not the mean.
- Put the monitoring pipeline's own health on the page, and force red when it fails.
The companion consulting document on our website includes the dashboard design rules and the status state-machine specification.
Part of RMAT's 12-part series on AI governance — Governing AI with Evidence. The companion consulting document — detailed checklists, a risk table, a maturity self-assessment and a 90-day action roadmap — is available on our website.
When did your AI dashboard last show red — and did anyone check whether that's because nothing was wrong?