Andon systems in 2026: from pull cords to real-time AI alerts

If you've spent any time on a production floor, you've seen the classic andon setup: a pull cord, a stack light, maybe a horn that goes off when something stops. For decades, that system worked. An operator pulled the cord, a supervisor walked over, and the team figured it out together.

But that model was built for a world where the biggest challenge was getting someone's attention. In 2026, the challenge isn't awareness. It's speed, context, and prediction. Pull cords tell you something went wrong. They don't tell you why, how often it's happening across shifts, or whether the same issue is brewing on Line 3 right now.

The andon concept — making problems visible so teams can act — is as relevant as ever. What's changed is the technology delivering on that promise. This article walks through how real-time machine monitoring and AI-powered alerting help plants move beyond static light towers to systems that get the right information to the right person before a small issue turns into lost production time.

Understanding the basics: andon, monitoring layers, and alert types

Before diving in, let's ground some terms. Manufacturing visibility isn't one thing; it's a stack of complementary layers.

A traditional andon system lived in the "machine monitoring" row — manually triggered, locally visible. A modern andon system spans all five layers: automatically triggered, digitally distributed, and increasingly predictive.

Most plants need at least two or three of these layers working together to get a clear picture. The good news: you don't have to deploy them all at once.

The real problem isn't awareness, it's ambiguity

Here's where many operations leaders get stuck. You know downtime is a problem. You might even have a rough sense of how much you're losing. But the specifics are buried in spreadsheets, shift handoff notes, or, worst case, someone's memory.

30–50% of downtime goes unmeasured or misclassified as idle time, changeover, or "other" (Guidewheel Performance Analysis). Plants that deploy real-time production monitoring systems typically uncover 15–25 hours of invisible losses per line per week in the first few days alone.

That's the gap a modern andon system fills. Not just "Line 2 is down" but "Line 2 stopped 14 minutes ago due to a film tension anomaly that's been recurring every third shift this week, and here's the recommended fix."

Why operators sometimes ignore alerts during busy shifts

This is worth addressing directly, because it's one of the most common frustrations. If your alert system fires constantly, operators learn to tune it out. Static threshold alerts, where the system screams every time a temperature crosses 75°C regardless of context, are the biggest culprit.

The fix isn't fewer alerts. It's smarter alerts. Adaptive thresholds that learn from your baseline operating conditions and only fire when something genuinely deviates from normal. In Guidewheel's experience working with customers, moving from static to adaptive alerting has reduced false alarms by 60–80% in some facilities (Guidewheel Customer Research), which means the alerts that do come through actually get attention.

How current sensing powers the modern andon

So how does this actually work on legacy equipment, the kind where half your machines predate your ERP system?

The approach that's gained traction in brownfield plants is simple: clip a sensor onto the power feed of a machine and read its electrical current signature. A running machine draws current in a predictable pattern. When that pattern changes, something changed on the machine.

This works on everything from decades-old CNC mills to brand-new packaging lines, without touching a PLC, without pulling a single wire, and without shutting down production to install. Guidewheel's FactoryOps platform uses exactly this approach: clip-on current sensors, combined with proprietary algorithms that translate those signals into run/idle/down data, anomaly detection, and predictive maintenance alerts. It works over cellular — no plant Wi-Fi required — which means even facilities with strict IT policies can get started with no IT project required.

The output is a machine monitoring dashboard that shows every asset's status in real-time, with historical trending, downtime categorization, and alerts routed to phones or tablets.

Setting up real-time downtime alerts for supervisors

For supervisors who need to act fast, the setup is straightforward:

• Define which machines are critical (your bottleneck assets)

• Set alert thresholds based on learned baselines, not arbitrary limits

• Route alerts to mobile devices with context: what stopped, when, and suggested next steps

• Establish escalation rules so unresolved alerts automatically bump up after a set time window

The goal is to get a supervisor acting within five minutes of an event, not scrambling to figure out what happened 30 minutes later.

What the benchmarks actually show

Let's look at some data to ground this. According to Guidewheel Performance Analysis across 3,000+ tracked machines, the overall median runtime sits at roughly 32%, while the volume-weighted average is closer to 55%. That gap tells an important story: a small number of high-utilization machines pull the average up significantly, while many assets sit idle more than they run.

This chart illustrates how dramatically perceived performance shifts depending on whether you measure by machine count or by total machine-minutes. In sectors like Pharmaceuticals and Plastics, the weighted average is multiples higher than the median, meaning a handful of high-volume machines mask the underutilization of many others.

The takeaway: if you're relying on plant-level averages to assess performance, you're almost certainly missing the full picture. Machine-level visibility, the kind a real-time equipment monitoring system provides, is the only way to see where capacity is actually hiding.

These benchmarks serve as reference points. Optimal performance varies significantly by facility, product mix, shift pattern, and equipment age.

Where your biggest improvement levers hide

When plants start categorizing downtime with real data instead of estimates, a clear pattern emerges. Here are the top loss drivers from Guidewheel's dataset, each representing an equally important focus area for operations teams:

(Source: Guidewheel Performance Analysis. Percentages may not sum to 100% due to rounding.)

This scatter plot shows where modern andon systems can have the biggest operational impact. Mechanical breakdowns and operational issues cluster in the high-frequency, lower-duration zone, exactly where automated alerts and rapid response protocols deliver the fastest ROI. Staffing issues, by contrast, are less frequent but significantly longer per event, highlighting the value of remote monitoring and flexible scheduling tools.

The critical insight: while "no business/orders" dominates statistically, it's the operational, mechanical, maintenance, and staffing categories that your team can directly influence. These are the levers that respond to better visibility, faster alerts, and smarter scheduling.

Coaching supervisors to act on downtime within five minutes

The data above is only useful if it drives behavior. The protocol is simple: operators acknowledge and log a reason code within two minutes; supervisors receive an escalation if the line hasn't restarted by minute five; maintenance is automatically notified with context by minute fifteen; and plant managers receive a summary with recommended actions if the issue runs past that.

This isn't about adding pressure. It's about removing ambiguity so everyone knows what's expected and has the information they need to act.

The difference between andon lights and mobile alerts

Traditional andon lights are local, visible only to people near the line. They convey status (green/yellow/red) but no context. Mobile alerts invert that model:

In 2025, the smartest plants use both. The physical light gives operators an immediate visual cue. The digital alert ensures the right support arrives fast, with the right context, whether they're on the floor, in the office, or at another facility.

A 12-week playbook to modernize your andon system

You don't need a massive capital project to make this shift. Here's a practical rollout framework:

The pilot phase typically runs $5k–$15k and, based on Guidewheel customer data, often pays for itself within the first two to three weeks through recovered hidden downtime — though results vary by facility, asset mix, and baseline utilization. Scale and optimize phases build on proven wins, reducing organizational risk at every step.

This is the Pilot, Prove, Scale approach: start small, prove value with real numbers, then expand based on what the data tells you.

Start turning downtime data into production gains

The evolution from pull cords to AI-powered alerts isn't about replacing a simple system with a more complex one. It's about giving your teams the same situational awareness the andon cord was always meant to provide, just faster, smarter, and across every asset in your plant.

If you're still relying on spreadsheets, shift notes, or gut feel to understand where your production hours go, the first step is straightforward: pick your biggest bottleneck, clip on a sensor, and see what the data reveals. Guidewheel's FactoryOps platform makes this possible on any machine, old or new, typically within days.

We had our best month of the year, increasing production from 26k-35k/month to 46k cases in March. I attribute this to Guidewheel. Being able to see downtime data and address downtime reasons directly correlates to higher production.

Michael Palmer, VP of Operations, Direct Pack

Ready to see what your machines are actually doing? Book a Demo to start uncovering hidden capacity fast.

About the author

Lauren Dunford is the CEO and Co-Founder of Guidewheel, a FactoryOps platform that empowers factories to reach a sustainable peak of performance. A graduate of Stanford, she is a JOURNEY Fellow and World Economic Forum Tech Pioneer. Watch her TED Talk—the future isn't just coded, it's built.