The silent machine: how long do you go before you notice?

Most operators find out a machine is down the same way: the location manager calls. Sometimes it is a customer who tried three times with a card that would not read. By then the machine has been dead for days, the location has decided your service is unreliable, and you are driving out on an unplanned trip to fix something you could have caught on a Tuesday morning.

The problem is not that you are not paying attention. The problem is that silence does not announce itself.

A dead machine looks exactly like a slow one

When a coil jams, you get a fault code. When a bill validator chokes, you get a fault code. Those are events, and events are easy to catch, because something happened and the machine said so.

A machine that goes silent does the opposite. Nothing happens. No read arrives, no error arrives, nothing lands in your dashboard at all. If you are watching a sales number, a silent machine and a slow Monday look identical: both are zero.

That is why “check the dashboard every morning” does not work at scale. Zero is not suspicious on its own. You need something that knows what that specific machine’s zero is supposed to look like.

Silence is judged by how the machine talks

The naive rule is “alert me if a machine has not reported in 24 hours.” It fails immediately: a dead machine sits most of a day before anything is said, and every carrier hiccup on a healthy one cries wolf. You learn to ignore it, which is worse than not having it.

So the rule has to match the transport, because your machines do not all talk the same way. VendGogh watches two:

DEX machines report on a sweep. The expectation is a read every hour, and every read is counted against that schedule. The Machine metrics view we shipped earlier this month shows the count directly: a healthy machine reads “26 of 24 reads, 24h” — slightly ahead of expectation. (That figure and the ones below come from our demo fixtures, not from a customer’s fleet.) When the count falls behind, you have a signal long before the day is over.

Cashless machines do not sweep — the card reader is the heartbeat. For MDB machines, liveness comes from the terminal itself: the payment provider tells us the moment a reader drops offline, and that event is the signal. No waiting on a schedule at all, and — just as important — a working card machine is never called dead simply because its data arrives by a different road.

Three missed reads, then say something

One missed read is a network blip. Two might be weather, or a tower, or a device rebooting. Three in a row is a pattern, and patterns are what you want to drive on.

For a DEX machine on the hourly schedule, that means a dead machine tells you in about three hours — not the next morning, and not after a location manager calls. For a cashless machine, it can be faster still, because the provider’s offline event arrives the moment the terminal drops.

It also means the verdict is honest. “Not reporting” says the machine has missed its expected reads three times running. It does not say the machine is broken, and it does not say the machine is fine. It says: this one stopped talking, and it has been long enough that it is not a fluke.

Rank by money, not by age

Here is where most alert lists fall apart. Sort by age and the top of the list never changes. Our fixture fleet has a machine, Thunderbird Molding #1, that has not reported in 75 days at a location that is effectively closed. Sorted by time, it sits at the top forever, and the machine that went quiet at 9am this morning at a busy site is somewhere below the fold.

Sorted by dollars at stake per week, the list reorders itself into your route. The 75-day machine carries $0 a week, because it has no sales to lose. A depletion alert on a machine that actually moves product carries $110 a week and rises to the top. You do not have to triage the list. The list already did it.

The alerts rebuild going on now extends this: tracking which alerts cleared themselves and the median time to clear, so you can see whether the ones you drove out for were worth the trip, and a set-aside action for the ones you have decided to live with.

Do this today

• Pick your five highest-revenue machines. Look up the last read time on each, right now.
• For each one, write down how often it should report. If you do not know, that is the first gap to close.
• Check whether anything in your current setup would tell you if that machine stopped reporting, and how long it would take.
• Find your oldest silent machine. Decide whether it is a service call or a machine that should come off the fleet.
• Sort your open alerts by weekly revenue at risk instead of by date, and see whether the order changes. It usually does.
• Agree on the escalation rule with whoever drives the route, so the alert lands on a person, not a screen.

The point of all this is not more notifications. It is that a machine losing you money every hour should reach you before your customer does.

See how ranked alerts and per-machine reporting intervals work on our alerts feature page.