Vending

Beyond remote monitoring: Turning equipment data into restocking, merchandising and service decisions

Connected vending machines generate valuable data, but visibility alone doesn't improve profitability. Nathan Yin of NexMart Smart explains how operators can use telemetry to optimize restocking schedules, product selection and technician dispatch while reducing unnecessary service visits and operating costs.

Image: Adobe stock

October 21, 2026 | Nathan Yin, Director of Operations, NexMart Smart Inc.

Summarize with

For many operators using connected equipment, the pattern is familiar. At first, the new visibility is genuinely useful. Over time, the dashboard becomes a tab somebody checks on Fridays, and the route still runs on the schedule it ran on before.

Connected equipment makes stock levels, sales and machine faults easier to monitor. The harder part is deciding what to do with that information. If a low-stock alert does not change a restocking decision, and a fault alert does not help anyone decide when to send a technician, the system has added visibility without changing much else.

That leaves three decisions: when to restock, which products deserve more space, and when a fault warrants a technician's visit. Each needs its own rules, and those rules have to reflect the economics of the operation.

Restocking: From a calendar to written thresholds

Moving from scheduled to on-demand servicing gets described as a technology change. A large part of the operational work is setting clear thresholds and making sure people follow them. A calendar tells a driver where to be on Tuesday; a threshold decides whether that trip should happen at all.

Some useful starting figures have circulated for years. In a 2011 VendingMarketWatch article, Glenn Butler — a long-time vending software executive — described servicing a snack machine once two products or three columns are sold out. Visit economics have their own conventions: roughly 100 vends per visit has been suggested as the point at which a stop is worth making, while Scott Meskin of Black Tie Services in Boston aimed to collect $100 to $150 per visit.

Those two measure different things — units sold and money collected — and neither is a universal standard. Both predate widespread connected equipment. The more useful exercise is to compare them against your own sales, service costs and stockout patterns, and then set thresholds that fit.

One structural detail is easy to miss. The decision unit is the location rather than the machine: if one machine at a site forces a visit, every other machine there is serviced early, and that cost is spread across all of them. So the question is not whether one machine justifies a trip, but whether its signal justifies moving the whole stop forward.

A practical test is whether the drink machine can hold until the snack machine's next scheduled visit. If it can, avoid adding a stop; if it cannot, service both together.

Merchandising: Two questions slot data can answer

Slot-level data cannot tell an operator what assortment to carry, but it can help answer two practical questions — provided they are asked separately.

Did it sell out, or did it not sell? In a total-sales report, those outcomes look identical, but they call for different responses. Sold out means demand exceeded supply, which points at depth or facings rather than the product. Steady slow movement at a location with traffic points at the product, or at where it sits.

Second, is the bottom of the range earning its space? Meskin used a rough screen: over the period being measured, the bottom seller should be at least 20 percent of the top seller. It is a rule of thumb rather than a threshold, and it moves with price, margin, facings, replenishment cost and the length of the observation window.

Often the cheaper fix is space rather than product. Taking facings from weak sellers and doubling up stronger ones can extend the service interval without losing sales. The same article illustrates it with a machine-level example: allocating an extra column to a cola was projected to save three deliveries a year and stretch the interval from 27 days to 36 days. That is a modeled example from a software report, not a result measured in the author's own operation.

There is a demand-side reason to revisit planograms as well. The NAMA Foundation's 2024–25 State of Convenience Services census, conducted with Technomic, found that 65 percent of operators cited client requests for healthier product mixes. Those requests are a reason to review assortments, while the selections themselves should still reflect the needs of each location.

Dispatch or defer: The decision that gets skipped

Much of the published advice about telemetry covers restocking. Less covers whether to send a technician today, which is where money is lost more quietly.

A simple three-tier framework can help operators prioritize service calls. Same-day: the fault stops vending or payment, or threatens perishable stock. Roll into the next scheduled visit: the machine still vends but something is degraded. Defer and watch: a reporting fault with no revenue impact, or one that has already cleared.

Two checks should come before anyone is dispatched. The first is to rule out a shared cause. If several machines fail at once, the cause is more likely to be a shared network, platform or payment-processing problem than several simultaneous hardware failures — so start with the services those machines have in common.

The second is to price the two kinds of error separately. A false alarm costs a trip. A missed refrigeration failure costs stock, a customer relationship, sometimes the account. Those consequences are not symmetrical, so alert thresholds should reflect the cost of each type of error rather than treating them as equally serious. For an operator running a large fleet, this is one of the most consequential settings to review.

Whether any of it worked shows up in first-time fix rate and repeat calls rather than in alert volume. Did the visit end the problem, or reset a symptom?

The machines that go quiet

Low stock is a visible problem. Machines that have stopped reporting are a quieter one, and the two get confused in practice.

Reporting frequency varies by device and platform. Card readers are generally expected to stay connected, though many can authorize offline for a period, while some telemetry devices are designed to report only every four to twelve hours. A machine can therefore be out of contact for much of a day without looking unusual.

The operational risk is what happens next. If a machine stops reporting, its stock data may no longer be current, and low-stock alerts may be delayed or missed — so the dashboard can look reassuring at exactly the machines that are least visible, and the route falls back to the calendar for those.

The remedy is an operating rule rather than a technical one: make stale data visible, confirm last-contact time, treat a missing heartbeat as its own alert class with a named owner, and show the last-contact timestamp beside the stock level.

Where telemetry earns the least

Return scales with variability. At a location with a stable population — the same people, the same pattern, the same preferences — data tends to confirm what the route driver already knows. That is a reason to manage expectations, not to leave those machines unconnected.

Low-volume sites add a second problem. When a machine sells a handful of units a week, one sale moves the depletion reading and any product ranking is mostly noise, so product decisions there need a longer review period. Restocking intervals and par levels still have to answer to shelf life, temperature requirements, stockout risk and the cost of an extra visit.

What telemetry will not do

It does not create demand. A weak location stays weak however well it is instrumented, and data tells you which stops still need a visit rather than which stops should exist.

More sales do not automatically mean more money. Products differ in the margin they carry, the space they occupy, and the cost of keeping them in stock, and those differences can matter more than popularity. A fast-selling item that consistently runs low can increase service frequency for the whole machine.

A full machine is not necessarily the most profitable. Higher par levels can reduce stockouts but tie up more inventory and can increase waste; lower levels can mean more frequent visits, or a customer finding their usual product missing. The right balance depends on product margins, shelf life, demand variability, and the cost of servicing the location.

Servicing against thresholds also takes product and route discretion away from the driver. That is the intent, but it is an organizational change with a compensation conversation attached rather than an installation.

The bottom line

The test of a deployment is not how many signals it collects. It is whether you can write, on one page, a rule for each signal: what triggers it, who owns it, what action follows, and what it takes to override it.

Remote monitoring answers what is happening. The return comes from the next question — what are we going to do about it, and is it worth the trip? None of that requires new hardware, which is probably why it is the part that gets skipped.

Nathan Yin is Director of Operations at NexMart Smart Inc., the company behind XMAI AI-vision refrigerated smart coolers. He works with vending and micro market operators in the United States.