Simulators and emulators make IoT testing fast and cheap to scale, which is why teams lean on them early. The question is where that stops being enough, because some behaviors only show up on physical hardware.
Application logic, API integration, protocol handling, and most functional flows can be tested against a simulated device without touching hardware at all. This is also where most of the test volume should live, since it is far faster to run and far easier to automate.
Battery drain under real workloads, behavior on an unreliable or intermittent network connection, sensor accuracy in real-world conditions, and how a device performs after physical wear are all things a simulator does not reproduce. These need real devices, in realistic conditions, not a lab bench with ideal signal strength.
A device that talks to other devices, hubs, or third-party ecosystems needs to be tested against the actual hardware and firmware versions it will meet in the field. Two devices that both pass their own simulated test suites can still fail to talk to each other correctly.
Use simulators for the bulk of functional and regression testing, where speed and scale matter most. Reserve a smaller, targeted set of real-device tests for power behavior, connectivity edge cases, sensor accuracy, and interoperability with the actual ecosystem the product ships into.
Talk to our QA team about IoT testing.