At 2:40 on a Wednesday afternoon, the outdoor LED sign above a fuel station in Springfield, Missouri goes dark mid-cycle. Nobody calls. A driver pulls in, looks at the blank panel, assumes the station is closed for remodeling, and leaves. On Thursday at 9:15 the dealer's weekly schedule report still shows the board as scheduled and playing. On Friday at 8:05 the advertiser, who bought four weeks of morning drive time, asks what exactly they paid for.

The fault took ninety seconds to find once someone opened the console. Power supply three on cabinet twelve had reached its 80 °C cutout, protected itself and dropped to zero volts, and nine modules downstream of it had nothing to run on. The other two supplies in that cabinet were fine. The screen was not broken. One part was.

Nothing in that sequence is unusual. What is unusual is that the dealer had a camera pointed at the board the whole time and still lost two days.

What a camera can and cannot see

A camera proves one thing: at the moment of the frame, the panel was not showing a picture. It cannot say why, and it cannot say since when. Those are the two facts a service call is built on.

In practice it also proves several things that are not true. At dusk the panel is brighter than everything around it, so a metered exposure turns the screen into a white rectangle and the alert fires on a board that is working. Fog does the same thing. A dirty lens does it for a week. A driver who parks in front of the board overnight does it until he leaves. If the site has a second network drop to spare, the camera also competes with the player for bandwidth on a link that was sized for one device.

None of that makes cameras useless. A frame after a fault is good evidence for an advertiser dispute, and a picture of a vandalised panel is worth more than a paragraph describing it. But a camera is a recording device that happens to be pointed at the problem. Alerting needs instruments.

Five readings that predict a truck roll

An outdoor LED sign is a small electrical plant, and health monitoring is simply reading that plant on a schedule: supply output, cabinet temperature and receive-card heartbeat, each compared against a threshold or against the sign's own history. The values below move before a customer notices anything, in the order they usually move.

Thresholds reflect the cabinets in our own deployments: 4.2 V module rails, a 60 °C cabinet alert and an 80 °C supply cutout, mostly on 6.35 mm and 10 mm outdoor product. Another manufacturer's numbers will sit somewhere else. What does not change between them is which reading moves first.

Two of these five are worth a text message to a technician at 3 a.m. A sagging rail that has not collapsed yet, and a cabinet that has crossed its 60 °C alert twice in a week. Those are the cases where a fifteen-minute phone call replaces a half-day drive. The other two are worth a ticket in the morning queue.

Four faults, four different alerts

A camera sees one symptom: no light. The platform sees four different signatures, and each one points at a different part. These four produce most of the calls we see, and they are the ones worth putting in front of a dealer before a demo.

Not one of those four reads as a camera alert. Each is a number crossing a threshold, and that is what lets the platform name a part instead of naming a screen.

Alerting is a filtering problem, not a detection problem

Detection is the easy half. Every receiver card made in the last decade reports its own heartbeat, and every decent power supply exposes a rail. The hard half is what to do with 41 messages when one power supply dies and nine modules behind it each report a fault of their own.

A platform that sends all 41 is a platform whose notifications get muted in the second week. The technician does not become faster. He learns to swipe left, and then he misses the one message that mattered.

The version that survives contact with a real fleet does two things well. It groups by root cause, so one dead supply becomes one incident with nine affected units attached. Then it waits before it speaks, long enough that a two-second brownout during a mains switch does not turn into a phone alert at 3 a.m.

The rest is sorting. A brightness drift can wait for the morning report. A supply sitting at 0 V cannot.

An alert that arrives nine times is an alert that gets muted. The value of monitoring is not that you hear about the fault first. It is that the phone call comes to you instead of to your customer.

Once a day, before anyone calls

Alerts fire when something crosses a line, which makes them blind to everything that crosses a line slowly. A fan that is 40% blocked will not trip a threshold this month. A rail that has drifted 0.2 V since spring reads as normal on any single sample. A cabinet running 7 °C hotter than the one beside it stays inside its limit every afternoon, and will not stay inside it in August.

So once a day, at an hour the dealer picks, the platform walks the whole fleet and writes a report: heartbeat history, rail average and minimum, cabinet peak against ambient, the modules that failed the frame check, and the playlist version against what was published. None of those rows needs a fault to be present.

The comparisons are the useful part. Two cabinets on the same board, same sun, same load, 7 °C apart is a fan problem with a date on it. That becomes a work order for a scheduled visit, instead of a call at 8:05 on a Friday.

It also produces the least glamorous document in the business: one page saying the sign was checked every day last month. When an advertiser asks whether anyone was actually watching their slot, that page answers the question without a meeting.

What the absence costs, in the units a dealer actually counts

Digital is 36.3% of a $9.46 billion US out-of-home market, and every point of that share sits on hardware someone has to keep lit. A service visit on a board an hour and a half out is not one cost. It is fuel and a truck, four hours of a technician's day that cannot be sold twice, and the ad slot that was dark while he drove. On a contract with a stated availability number, it is also the credit at the end of the month.

Set that against a fault that can be identified remotely in nine minutes. When the answer is a power supply, the visit still has to happen. But it happens once, with the right part in the truck, in daylight, on a day the slot is not sold. That is a different job from the one that starts with a phone call at 8:05 on a Friday.

How much that is worth depends on how many of your faults are remote-answerable. In our own data it is most of them: power supplies, card links, brightness settings, and scheduling errors that look like hardware failures until someone reads the log. If six in ten calls stop being a guess, you have bought back roughly a day a week of technician time, and that is before anyone counts the credits you did not pay.

When LED sign monitoring is not worth the money

There are deployments where we would tell a dealer to skip it, and this is the part most vendors leave out.

• One board, visible from the office window.Someone looks at it forty times a day. An uptime ping and the camera you already have will cover the rest, and the camera was cheap.
• Self-owned signs with no advertising contract.A church board or a school marquee has no slot to credit and no advertiser to answer. The cost of a dark day is that it looks bad, which is a smaller number than a monitoring subscription in most cases.
• Under ten screens, all inside thirty minutes.Below that fleet size the technician is the monitoring system, and he is already driving past two of them today.

The crossover is not screen count. It is whether a dark panel costs somebody money they will ask you to give back. Once a sign carries a paid slot, a stated availability commitment, or a customer who calls at 8:05 on a Friday, the arithmetic flips, and it flips hard.

Five questions to ask before buying LED sign monitoring

Feature comparison sheets from five vendors will look almost identical. These five questions separate them faster.

• 01Does one failed part generate one alert or nine? Ask to see the message log from a real power supply failure, not a demo with a planned outage.
• 02Is there a confirmation window before a notification goes out? A platform with no window will page someone for a two-second mains dip.
• 03Can it show rail voltage, or only online and offline? Online/offline is a camera with an API. It tells you the same thing a minute later.
• 04Do remote actions leave a timestamp? When a customer asks whether you already tried a reboot, the log should answer without a second call.
• 05What happens to the data when the site loses its link? A cabinet that buffers four hours locally and backfills on reconnect is a different product from one that goes blind.

Questions about LED sign monitoring

LEDPLAY Cloud is a monitoring and diagnostics platform for commercial LED displays, deployed on AWS us-west-2 and supported for sign companies and their end-customer sub-accounts. Field readings quoted here come from cabinets in our own deployments, not from a vendor specification sheet. Editorial questions: news@ledplaycloud.com.

Related: why US DOOH growth puts the pressure on service, not sales. The 2025 revenue figures and the digital share trend behind this article are laid out there.