Choosing IoT Connectivity: Wi-Fi, BLE, LoRaWAN, NB-IoT, or Cellular

The single fastest way to derail an IoT product is to pick the wrong radio. A wrong choice on the connectivity layer cascades into firmware redesigns, certification cost, runtime battery, and the cloud bill — all of which surface late and hurt expensively.

We approach this decision in five questions. Answering them honestly takes about an hour. Skipping them takes about a year.

How much data, how often?

Estimate your worst-case payload size and reporting cadence. A vibration sensor pushing 4 KB every five minutes is in a different universe from a glucose monitor uploading 60 bytes every fifteen. Add 30% for retries and protocol overhead. The result anchors every other choice.

If your honest answer is “less than 1 KB per device per hour,” LoRaWAN, NB-IoT, and even Sigfox stay on the table. Once you cross 100 KB per hour or need real-time streaming, you are deciding between Wi-Fi and cellular.

Where will it live?

Range and obstruction matter more than spec sheets suggest. Wi-Fi inside a steel-walled industrial enclosure is not Wi-Fi at the office. Cellular in a basement utility closet is a coverage map you cannot read from your laptop. LoRa over a tree line is not LoRa over an urban street.

Survey, do not assume. A field walk with a signal-strength tool catches problems that two months of engineering will not solve.

What is the power budget?

If the device plugs in, the radio is rarely the bottleneck. If it runs on a coin cell or a single 18650 for years, the radio is the only thing that matters. BLE in advertising mode and LoRa with infrequent uplinks both deliver multi-year lifetimes. Wi-Fi without aggressive sleep handling will not survive a season.

Put a number on it. “Two-year battery on two AA cells” is a budget. “Long battery life” is not.

What is the cost model per device per year?

Cellular has SIM and data plan cost — and SIM management cost, which is often invisible until you have a thousand devices and need to swap a carrier. NB-IoT is cheaper per byte than LTE-M but has narrower coverage. LoRaWAN public networks bill by message; private networks have gateway capex but no recurring per-device fee. Wi-Fi and BLE piggyback on the user’s network and cost zero per device, with the trade that you inherit their network problems.

A rough rule: if connectivity per device per year is more than 10% of bill of materials, you have picked the wrong layer or you have not negotiated.

Who owns the gateway?

If your product needs Wi-Fi, you own the onboarding flow and the support tickets when the user changes routers. If your product is BLE, you own a phone-app pairing flow forever. If you go LoRa, you own the gateway hardware on customer premises or pay a network operator. If you go cellular, you own the SIM relationship and the international roaming spreadsheet.

There is no free lunch. Each layer trades a hardware problem for an operations problem. Pick the one your team is ready to own.

A short decision table

These rules of thumb are right more often than not:

• Indoor, mains-powered, high-data, consumer or office context: Wi-Fi.
• Phone-tethered, low-data, consumer device, paired once: BLE.
• Outdoor, low-data, multi-year battery, predictable site density: LoRaWAN.
• Mobile asset tracking, low-to-medium data, global roaming: LTE-M / NB-IoT.
• High-data, mobile, real-time: LTE / 5G cellular.
• Mesh of devices in a building, low-data, no Wi-Fi dependency: Thread or Zigbee.

What we recommend at the start of a project

Spend a week building a connectivity prototype before locking the architecture. Two boards, two radios, your real environment, your real payload. Measure: throughput under load, power draw, gateway-to-cloud latency, and what happens when the link drops.

Then commit. The teams that ship reliable IoT products almost always made the radio decision deliberately and early. The teams that struggle made it implicitly, by picking a development board.

If you would like a second opinion on a connectivity choice you are wrestling with, book a call — we have probably already shipped it once.