A temperature sensor in a warehouse sends a reading every minute. That fact alone is not the Internet of Things. The Internet of Things begins when that reading is identified, transmitted, stored, compared with a threshold, and used to open a vent or alert a manager, without a human typing anything. The value of connected objects is not the data they produce. It is the closed loop that turns data into action. A great deal of confusion in the field comes from treating the sensor as the product, when the sensor is only the cheapest and least interesting component of a much larger system.

What the term means, and what it does not

Several overlapping terms describe adjacent ideas, and confusing them leads to bad design. Machine to machine communication is the older idea of two devices exchanging data without human intervention, usually over a closed network. The Industrial Internet of Things and Industry 4.0 apply connected sensing and control to factories and supply chains, with reliability and safety as dominant constraints. Cyber physical systems emphasize the tight coupling between computation and physical processes, where a decision in software changes the physical world within a bounded time. The Internet of Things is the broader umbrella: everyday objects embedded with sensing, identification, and connectivity, joined to an internetwork and to applications that act on what they measure. The distinctions matter because a consumer smart bulb, a factory robot, and a medical implant share almost no design constraints despite all being called IoT.

Three derived qualities separate a genuine IoT system from a gadget with a Wi-Fi chip. Context awareness means the system knows where it is, what is around it, and what state it is in. Adaptability means it changes behaviour as that context changes. Proactivity means it acts before a human asks, within defined limits. A thermostat that turns on when you walk through the door exhibits all three. A thermometer that only displays a number exhibits none of them.

The five layers that every real system contains

A robust IoT system is organised as five layers, each with its own engineering concerns. Understanding them is what prevents a project from collapsing when it moves from a pilot to a deployment.

  • The perception layer touches the physical world. Sensors convert a physical quantity into an electrical signal, actuators convert a signal back into physical action, and identification technologies, chiefly RFID, give each object a unique name. Passive devices draw no power of their own, semipassive devices monitor but do not initiate, and active devices both sense and transmit.
  • The network layer moves data. It covers identification schemes, the communication protocols, security at the link level, and the routing that moves a packet across a lossy, low power mesh. This is where most field failures originate, because radio links in real environments are neither symmetric nor reliable.
  • The middleware layer sits between devices and applications. It handles service discovery, so a new sensor can announce itself, data exchange formats, so a device from one vendor can be read by software from another, and computation offloading, so that simple devices can borrow processing power elsewhere.
  • The application layer delivers the actual service. It includes identity services, information aggregation that turns many readings into a picture, collaborative services where multiple devices coordinate, and ubiquitous services available from anywhere.
  • The business layer is where the data becomes a decision. Semantics give the readings meaning, and analytics turn them into forecasts, optimisations, and alarms. This is the layer that justifies the investment in the other four.

The enabling technologies, and where each one matters

Sensors and actuators have been transformed by miniaturisation. Micro electromechanical systems put a complete accelerometer, gyroscope, or pressure sensor on a chip costing less than a dollar, and nanoscale and soft actuators extend sensing into flexible and wearable form factors. A sensor is only as good as its calibration, however, and drift over time is the silent failure mode that no dashboard catches until the data is already wrong.

Identification, most commonly RFID, gives every object a unique, machine readable identity. It is the foundation of traceability, from digital product passports in supply chains to asset tracking in hospitals. The trend is to fuse identification with sensing, so that a tag reports not only what it is but also its condition. Connectivity divides into two families. Short range radios such as Bluetooth Low Energy and Zigbee suit high data rates over metres. Low power wide area networks such as LoRa and NB IoT suit tiny messages over kilometres, at the cost of low data rates and duty cycle limits. Choosing the wrong radio is the most common and most expensive mistake in IoT design, because it is baked into hardware and cannot be fixed in software.

Cloud, fog, and edge computing answer the question of where the data is processed. The cloud offers unlimited scale but high latency and bandwidth cost. Edge computing places processing on the device or gateway, so that only alarms and summaries travel upstream. Fog computing sits between them, in the local network. The rule of thumb is to process as close to the data as the latency and accuracy requirements allow, because every byte sent to the cloud costs money and leaks a little privacy.

Where the value lives, and where the risks hide

The economic value of IoT rarely comes from the readings themselves. It comes from the three kinds of innovation the data enables. Product innovation turns a static object into a service that improves over time. Process innovation replaces periodic manual inspection with continuous monitoring. Business model innovation changes what is sold, from a machine to the outcome the machine produces. Each of these requires the full five layer stack to work. A pilot that demonstrates only the perception and network layers proves nothing about the business case.

The risks are equally structural. Security and privacy are not add ons. A device with a default password, deployed in the thousands, becomes a botnet participant the day it ships. Interoperability fails when vendors lock data into proprietary formats, and the cost shows up years later when the system must be extended. Power and lifetime constrain every battery operated device, and a ten year deployment requires a ten year power budget. Sustainability and ethics are now first order concerns: billions of short lived connected devices create e waste and surveillance risks that the technology alone cannot resolve. A credible deployment names these risks in the design phase, not in a postmortem.

Further notes on applied technology are published on the blog, and technical enquiries can be sent through the contact page. The source reference is Qusay F. Hassan, editor, Internet of Things A to Z: Technologies and Applications, second edition, IEEE Press and Wiley, 2026.