In a nutshell: An IoT system is best understood as four layers — device, connectivity, platform, application. The device layer turns physical events into data; the connectivity layer ships the data (Wi-Fi, cellular, LoRaWAN); the platform layer stores, processes and secures it; the application layer turns it into something a human or business system acts on. The hard parts are rarely the radios — they're integrating with existing systems, securing fleets of devices, and proving ROI past the first deployment.
What "IoT" actually means in business
Strip away the marketing and an IoT system is: physical things that report data and/or accept commands over a network. That covers everything from a building's chiller plant streaming temperatures to a logistics company tracking pallets across the region to a hospital's infusion pumps pushing telemetry to a central system.
What makes "IoT" different from regular IT isn't the technology — most of it is well-understood. It's the operational model:
- Lots of cheap endpoints instead of a few expensive servers. Thousands or hundreds of thousands of devices, deployed across messy physical environments.
- Devices live for years. Often deployed and forgotten. Firmware updates need to be reliable and remote.
- Constrained. Limited CPU, memory, battery, bandwidth. You can't run a normal antivirus or pull down a 200 MB OS patch.
- Physical access matters. Devices get stolen, tampered with, soaked, dropped. Security has to assume hostile environments.
The four-layer IoT stack
| Layer | What it does | Examples |
|---|---|---|
| Device | Sense or actuate, run firmware | Sensors, actuators, gateways, controllers (PLC), cameras |
| Connectivity | Move data between device and platform | Wi-Fi, 4G/5G, NB-IoT, LoRaWAN, Ethernet, BLE |
| Platform | Ingest, store, process, secure, manage devices | AWS IoT Core, Azure IoT Hub, Google Cloud IoT, ThingsBoard, Particle |
| Application | Turn data into insight, action, automation | Dashboards, alerts, ERP integration, analytics, control loops |
Device layer — sensors, actuators, gateways
Three main device types:
- Sensors measure the physical world — temperature, humidity, vibration, position, presence, current, water level, air quality, image. They convert a physical quantity into digital data.
- Actuators change the physical world — open a valve, switch a relay, move a motor, sound an alarm. Less common than sensors but more impactful when they go wrong.
- Gateways aggregate many local devices (often using a short-range radio like BLE, Zigbee, or wired Modbus) and bridge them to a long-range backhaul (Wi-Fi, cellular, LoRaWAN). Gateways are often the single most expensive and security-critical part of a deployment.
Many industrial IoT projects start by adding network connectivity to existing equipment — PLCs (programmable logic controllers), CCTV cameras, building management systems — rather than greenfield new sensors. That's where most of the practical wins are: data that was already being generated locally is now visible centrally.
Connectivity layer — pick the right radio
The right wireless technology depends on where the devices are, how much data they send, and how much power is available. See The Modern Wireless Stack for the full comparison. The short version for IoT specifically:
- Wired (Ethernet / serial / Modbus). Industrial controllers, fixed sensors in panel-mounted equipment. Most reliable; least flexible.
- Wi-Fi. Anything indoors with mains power and within range of existing APs — smart displays, kiosks, IP cameras, voice assistants.
- Bluetooth / BLE. Short-range, very low power. Often used between a sensor and a nearby phone or gateway (beacons, wearables, asset tags).
- Cellular (4G / 5G). Mobile or remote devices where Wi-Fi isn't available — vehicles, vending machines, kiosks, outdoor cameras.
- NB-IoT / LTE-M. Battery-powered cellular IoT — smart meters, parking sensors, asset trackers.
- LoRaWAN. Battery-powered, long-range, low-bandwidth — water meters, environmental sensors, agriculture, building automation.
Platform layer — where the data lands
The IoT platform is the cloud (or sometimes on-prem) service that:
- Ingests data from millions of devices reliably, at scale.
- Authenticates each device and enforces what it's allowed to do.
- Manages devices — provisioning, firmware updates, decommissioning, monitoring health.
- Stores time-series data efficiently (most IoT data is timestamped sensor readings).
- Processes data — rules engines, stream processing, alerting.
- Exposes data to upstream applications via APIs and integrations.
The big three hyperscaler offerings — AWS IoT Core, Azure IoT Hub, Google Cloud IoT (note Google has scaled back its standalone offering) — dominate the platform layer for greenfield projects. Independent platforms (ThingsBoard, EMQX, Particle, PTC ThingWorx) compete on flexibility, on-prem deployment, or vertical specialisation. Many vendors also offer turnkey platforms bundled with their devices.
Application layer — where IoT pays for itself
The platform layer is plumbing. The application layer is what justifies the project. Three patterns:
- Visualisation & alerting. Dashboards showing real-time and historical sensor data; SMS / email / push alerts when thresholds are crossed. Often the first thing built.
- Integration with existing systems. Pushing IoT data into the CMMS (maintenance system), ERP (asset register), BMS (building management), or warehouse management system so it gets used inside business processes that already exist.
- Analytics & control. Detecting patterns (a chiller showing early signs of failure, a route consistently slow on Tuesdays) or closing control loops automatically (dimming lights when no occupancy is detected).
Protocols you'll hear about
- MQTT. The dominant messaging protocol for IoT. Lightweight publish/subscribe over TCP, with quality-of-service levels. Almost every IoT platform speaks MQTT.
- CoAP. A web-like protocol (REST over UDP) for constrained devices. Used in some smart-home and constrained-network deployments; less common in business IoT than MQTT.
- HTTP / HTTPS. Still used heavily for less constrained devices (camera APIs, gateway uplinks). Heavier than MQTT but universally supported.
- LoRaWAN. The networking protocol on top of LoRa radio. Defines device classes, encryption, network/application server split.
- Modbus, BACnet, OPC UA. Industrial / building protocols you'll encounter when bridging existing equipment into a modern IoT platform. Gateways typically translate these to MQTT or HTTP upstream.
- Zigbee, Z-Wave, Matter. Short-range smart-home / building protocols. Matter (an Apple/Google/Amazon-backed standard) is gaining ground for cross-vendor interoperability.
Singapore use cases that actually pay off
- Smart buildings. Occupancy and air-quality sensors driving HVAC and lighting; common in BCA Green Mark Platinum certified buildings. Often delivered using LoRaWAN or Wi-Fi sensors integrated into the BMS.
- Cold-chain & warehouse monitoring. Continuous temperature/humidity tracking for pharma, food, biotech. Required by HSA for many regulated products.
- Predictive maintenance. Vibration and current sensors on motors, pumps, and lifts; analytics flag developing failures days or weeks early. Adoption growing in manufacturing and FM.
- Asset tracking. BLE beacons indoors, GPS+cellular trackers outdoors. Used heavily in logistics, hospitals (medical equipment), and construction.
- Utility metering. Smart water and gas meters using LoRaWAN or NB-IoT; PUB has been a long-standing reference deployment.
- Smart-nation infrastructure. Public lamp-post sensors, environmental monitoring (NEA), traffic, parking. The Smart Nation Sensor Platform programme is the institutional backdrop.
Common pitfalls
- Pilot-purgatory. Beautiful POC with 50 devices that never scales because the unit economics don't work past a few hundred.
- "We'll integrate it later." Data in a dashboard that no business system consumes — interesting, not useful.
- Connectivity assumptions. Wi-Fi coverage in a warehouse that turns out to be patchy; LoRaWAN gateway placement that doesn't reach a basement; cellular signal that fails in a steel structure.
- Security afterthought. Devices with default passwords, unencrypted MQTT, no firmware update mechanism — and now there are 10,000 of them in the field.
- Vendor lock-in. Choosing devices that only work with one platform, then finding the platform is end-of-life four years in.
Where to go next
- Wireless deep-dive: The Modern Wireless Stack for choosing between Wi-Fi, 5G, LoRa and NB-IoT.
- Networking foundations: LAN vs WAN Basics for the connectivity context.
- Pick a partner: our system integrator buyer's guide.
Related articles:
Browse IoT vendors in Singapore
Looking for an IoT integrator, sensor specialist, or platform partner?