Shop floor dashboard design — AWcode

AWcode is an international studio in Pattaya, Thailand. We build SaaS startups, custom web apps, and factory software for clients worldwide since 2014.

Shop floor dashboard design

A shop floor dashboard works when it serves one role, reads clearly from the distance people actually stand at, holds to a handful of elements, and tells the truth about how old its data is. Validate it on the floor, under working light, before you build. Three reference documents are worth knowing: ISA-101 for interface design, ISA-18.2 and EEMUA 191 for alarms, ISO 22400-2 for KPI definitions.

Key takeaways

Why most shop floor dashboards go unused

Stand where the operator stands. Look at the nearest display. Can you tell within a few seconds whether the line is running to plan? Most dashboards fail that test on day one, and nobody says so, because nobody wants to tell the plant manager that the new screen is wallpaper.

The metrics are usually fine. The failures are physical and organizational.

The first is conference room bias. The interface gets designed on a high-resolution, color-calibrated laptop in an air-conditioned office, then deployed under high-bay lighting with glare, dust, and wide viewing angles. What looked crisp on a desk is unreadable at the line.

The second is the newspaper problem. Data refreshed on a 15 to 30 minute lag gets presented as live. An operator only needs to catch that once. After that, the paper log comes back.

The third is one screen serving three audiences. The operator has no use for a monthly OEE trend. The plant manager has no use for a live part counter. Put both on the same screen and neither person finds what they came for.

Decide who the screen is for before deciding what it shows

Shop floor interfaces fall into three functional tiers. The table below synthesizes the interface levels described in ISA-101 with the operational escalation models published by practitioners including SYMESTIC and TeepTrak.

Tier Who Device Refresh What it shows
1 Line operator Rugged kiosk, panel PC, or large andon board Real time Machine status, target vs actual count, active stoppage reason
2 Shift leader Tablet or station PC Hourly, current shift Bottlenecks, scrap Pareto, alarm history, micro-stoppages
3 Plant manager Desktop browser Daily to monthly OEE trends, yield, planned vs unplanned downtime, drill-down

The tiers break when you mix them. A tier-one board has to be readable from across the cell. Add tier-two detail and the type shrinks, the layout fills up, and the board stops working at distance. Tier two and tier three can carry density because people sit down to read them.

The physical brief dictates the design

Office software design can ignore the room. Shop floor design starts with it.

Viewing distance sets type size. A common practitioner rule for a tier-one board is that status and counts stay legible from five meters or more, which usually means far fewer elements and far larger numerals than a designer's instinct suggests.

Lighting sets the hardware. High-bay LED and fluorescent fixtures throw reflections, so matte panels and careful mounting matter more than panel resolution. Check the position at the hours the line actually runs. A screen that reads well on nights can be washed out by morning sun through a loading door.

Input method sets the interaction design. General interface guidance from Apple, Google, and the WCAG target size criteria puts minimum touch targets around 44×44 to 48×48 pixels. That is a baseline for bare fingers on a phone. Where operators work in PPE, practitioner guidance pushes targets considerably larger, commonly 60 to 80 pixels and up.

Capacitive screens also misregister or ignore gloved, dusty, and oily hands. Plan for that in software and hardware together: enlarged targets, dedicated physical buttons, ruggedized barcode or RFID scanners, or resistive panels that respond to a gloved press.

Choosing what to show and what to leave out

Working memory is the hard limit. Nelson Cowan's research on working memory capacity points to roughly four chunks held at once, and estimates in the field vary. Treat three to five elements as a practical ceiling for a tier-one view. Load twenty metrics onto an operator screen and the whole thing reads as noise.

For the metrics themselves, use recognized definitions. ISO 22400-2 specifies how manufacturing KPIs are calculated, including Overall Equipment Effectiveness: availability multiplied by performance multiplied by quality.

A raw OEE percentage on an operator screen rarely helps. The number alone gives no indication of what to fix. What the operator needs is the loss category behind it, whether that is a micro-stop, material starvation, a tooling change, or a changeover overrun.

Tier three gets the percentage and the trend. Tier one gets the reason the line stopped.

Color, state, and the limits of red and green

A factory floor is already visually noisy. If everything on the screen is saturated, nothing reads as urgent.

ISA-101 recommends a neutral or gray baseline, reserving high-saturation color for abnormal conditions. On a mostly gray screen, one amber block pulls the eye immediately.

Never encode state in color alone. Color as the sole indicator breaks universal accessibility standards and runs against ISA-101 guidance, and it also fails in practice when glare flattens contrast on the floor. Pair muted color with shape, bold text labels, and clear iconography so the state survives a bad viewing angle.

Traffic-light schemes get overused for the same reason they are easy to build. A wall of green boxes stops carrying information, and people stop looking at it.

Distinguishing alarms from information

The test is straightforward. If a notification does not require an immediate operator action or operational response, it is a status indicator or a log entry, not an alarm.

EEMUA Publication 191 and ANSI/ISA-18.2, also published as IEC 62682, set the benchmarks. They were written for the process industries, so read them as reference points for discrete manufacturing rather than binding rules.

When everything is an alarm, operators clear them without reading them. The benchmarks exist to protect attention, which is the scarcest resource on the floor.

Show the age of your data

Lagging data dressed up as live is the fastest way to lose an operator for good.

Put a timestamp on every live figure. When a machine or gateway drops out, show the connection state explicitly instead of freezing the last known value on screen. Pick a refresh interval and publish it. If the board updates every five minutes, say five minutes rather than implying a continuous feed.

Designing for a multilingual floor

Production hubs across Thailand and the wider ASEAN region run multinational lines. Operators may read Thai, Burmese, Khmer, or Lao while the management layer and corporate reporting default to English.

That changes the build. Downtime logging works better with pictorial codes and a language toggle on the kiosk than with long text drop-downs. Status should be icon-first. Translations need checking with the operators who will use them, not with a translation tool. Shorter text helps everyone, including anyone working at speed under pressure.

AWcode builds this kind of multilingual operator kiosk as part of its factory software work.

How to validate before you spend money

Test the assumptions on the floor before anyone writes code. AWcode runs a two-day on-site discovery with machine operators for exactly this purpose. The sequence below works whether you hire that out or do it yourself.

  1. Walk the floor for a full shift, including the changeover.
  2. Watch what operators write down now, and where they write it.
  3. Print a full-size paper mockup and tape it in the intended position.
  4. Run the five-second test at the real viewing distance, under real light.
  5. Try every control while wearing the gloves your operators actually wear.
  6. Return at the worst time of day for glare and look again.
  7. Confirm which data is genuinely live and which arrives late.

Most of the expensive mistakes surface in steps three through six, on paper, for the cost of a plot print.

What a realistic build looks like

A useful floor dashboard does not require a multi-year program.

In AWcode's own practice, a working prototype on local factory Wi-Fi takes about six weeks. Full deployments typically run eight to twelve weeks, in the $10,000 to $25,000 bracket. Those are the studio's figures, not an industry average.

A Manufacturing Execution System is a different scope of project, covering scheduling, traceability, and work order management across a plant. A targeted shop floor dashboard answers a narrower question: what is happening on this line right now, and who needs to act.

Frequently asked questions

How many metrics should a shop floor dashboard show?

Three to five elements on an operator screen. A shift leader view can carry more density, but it should still hold to metrics that drive a decision within the current shift.

Should operators see OEE?

Operators should see the parts of OEE they can influence: active stoppage reason, and target versus actual count. The composite percentage is an aggregated management figure and rarely tells a line operator anything actionable.

What screen size works for an andon board?

Large-format displays of 55 inches and up are common for overhead tier-one boards, but size is a consequence, not a starting point. Fix the viewing distance first, then choose the panel and type size that stay legible from there, and confirm it with a printed mockup in position.

Do I need a full MES to get useful floor dashboards?

No. Machine telemetry, simple retrofit sensors on older equipment, and operator input at a kiosk can all feed a dashboard that earns its place long before an MES is on the table.

Next steps for your factory

A shop floor dashboard is only as good as the room it lives in. Design for one role, respect the distance and the lighting, strip out everything that does not drive an action, and you end up with a screen people use instead of ignore. If your operations have outgrown spreadsheets and paper logs, review AWcode's approach to factory software.

← All news

Machine-readable

Resources for AI agents, LLMs and integrations.

Public API — concrete examples

Markdown mirrors — concrete examples