Isn't this just a long winded way of saying that you should get your testing right?
Sensor to cloud to decision output.
This is where the current thinking with IoT fails, what if the cloud is not available? what if time taken is outside acceptable range? Lets say I have an alarm sensor on my door, if it takes half an hour to set the alarm off for whatever reason then it's useless.
Having the cloud for storing data is all well and good for future analysis but you shouldn't be relying on it for the actual function of the device.