Factory dashboards that drive decisions, not vanity: a practical pattern

[Aior]

The wall of dashboards everyone has[/HEADING>
Walk into any factory in 2026 and there's a wall of monitors. Throughput. OEE. Quality first-pass yield. Energy consumption. Inventory levels. Most of those dashboards are looked at twice a day at best, and the decisions they're meant to support are made by gut feel anyway.

A useful factory reporting layer has fewer dashboards, sharper metrics, and a clear consumer for each one. Below is what we ship.

Three audiences, three layers​

• Operators — what's happening right now. One-shift horizon. Glanceable. Calls out attention with colour, not text.
• Supervisors — last shift to last week. Comparisons, trends, deviations from plan. Decision focus: what to act on this week.
• Plant management — last week to last quarter. Strategic comparisons across lines, products, shifts. Decision focus: where to invest.

Each audience needs its own dashboard. Mixing operator and supervisor views creates a screen useful to neither.

Operator dashboard — the live wall​

Visible from the line, big screen, no login. Shows:

• Current throughput vs target (single number, colour-coded)
• Last hour throughput trend
• Active alarms count + topmost alarm
• Time to next planned event (changeover, break, end of shift)

That's it. Anything else clutters the glanceable use case.

Supervisor dashboard — the morning meeting​

Used in the morning standup. Shows:

• Yesterday's OEE with breakdown (availability, performance, quality)
• Top 3 loss categories yesterday with minutes/parts attributed
• 7-day OEE trend
• Quality reject Pareto for the last week
• Active CAPA / countermeasures with owner and due date
• Anomalies vs the same period last week (e.g. "20 % more downtime than last Monday")

The dashboard is built to support a 15-minute meeting. If the meeting needs more than the dashboard, the dashboard isn't done.

Plant management dashboard — the strategic view​

Updated daily, reviewed weekly. Shows:

• OEE by line, last 30 days, with target line
• Cost per unit by line/product, trended
• Energy intensity per output (kWh / unit), trended
• Top 5 assets by maintenance cost, last 90 days
• Forecast vs actual production for the period
• Quality cost (rework + scrap) as a % of production cost

Plant management dashboards are not for live operation. They're for direction-setting. Don't try to combine the two.

The technology stack we converge on​

• Time-series store — TimescaleDB or InfluxDB. PostgreSQL with a time-series extension is fine at most factory scales.
• Aggregation layer — pre-computed continuous aggregates (TimescaleDB) or scheduled materialised views. Don't query raw 1-second data for an hourly dashboard.
• Visualisation — Grafana for engineering dashboards, Apache Superset / Metabase for management dashboards. Avoid hand-rolling unless you have a specific reason.
• Alerting — separate from dashboards. Alerts are the ones that page; dashboards are the ones that inform.

The reporting cadence that produces results​

• Real-time — operator dashboard, live alarms
• Daily — supervisor dashboard, morning standup
• Weekly — plant management review
• Monthly — strategic review with corporate, if applicable

Each cadence has a meeting attached. Without a meeting, the data doesn't drive a decision; it accumulates.

One pattern we'd avoid​

"Dashboard sprawl". Every requested KPI gets its own dashboard, until the wall is unreadable. Curate aggressively. Retire dashboards that nobody opened in 30 days.

One pattern that works​

Annotations on the time-series. When a major event happens (line stop, product changeover, CAPA implemented), annotate the time-series. The trend graph in 6 months shows the events alongside the data, and the patterns become legible.

What's your reporting stack? And — controversial — has anyone successfully replaced PowerBI for plant reporting with Apache Superset?​