Connected devices in year one: what pays off, what can wait

A young company operating premises, a workshop, a fleet or equipment quickly wonders whether to "add sensors". The answer depends on one question, to be asked before opening any catalogue: which decision will this measurement change, and how often? Data that triggers no decision only produces a chart.

Three cases that pay off early

• Energy, when it is a visible cost line. Sub-metering by zone or by machine reveals out-of-use consumption within weeks — equipment running overnight, heating fighting ventilation. The decision is immediate and quantifiable.
• Cold chain and regulatory thresholds. Here measurement is not comfort, it is evidence. A timestamped, retained record beats a manual reading forgotten on a Sunday, and it protects you in a dispute.
• Presence and access, for shared or isolated sites. Knowing a door stayed open, a room was entered out of hours, or a machine started without an operator triggers action the same day.

Three cases that can wait

• Predictive maintenance without history. A predictive model needs past failures to learn from. The first months build the dataset; prediction comes later. Start with measurement and a simple threshold, not with the model.
• The real-time dashboard nobody opens. If no one has time to look at a screen each morning, you need an alert, not a dashboard.
• Full rollout before a pilot. Ten sensors for one month on the costliest case teach more than a hundred sensors everywhere.

The cost nobody budgets

The sensor price is rarely the issue. The real cost sits in connectivity, power and batteries, mounting, maintenance — and above all in integration with the information system: as long as the data stays inside the manufacturer's app, it triggers nothing and does not really belong to the company. That is what a gateway is for: capture locally, normalise, and hand the data to the rest of the system, including the automations and agents that will act on it.

Security: the two minimum rules

A connected device is IT equipment. Two rules remove most of the early risk: no default password left in place, and no device exposed directly to the internet — access goes through a controlled gateway. Updates, segmentation and logging come next, but those two are not negotiable.

The order that works

Before sensors, know which decisions the company makes and on which data. In other words: architecture first, then the digital foundation that will receive the data, then field measurement — and finally who acts on the alert, human or agent. A sensor alerting a mailbox nobody reads protects nothing. To set that sequence without spending a quarter on it, the minimal loop is described in TEAF Light. And if your company is less than a year old, five places are open in the 5 Founders programme.