Back to the number-one rule in IoT analytics: Never trust data you don't know.

You need to get to know your data. Just like that new guy you just met, you Google him, ask around about him, and do a criminal background check (we know you do) before agreeing to go to dinner with him.

Here is a step-by-step strategy to start to understand the source of the data:

1. Draw a picture of the device: Sketch it out or use blueprints if you have them. This is where you make friends with the design engineer and ask her some questions. Draw the key sensors in the place they are located and note the sensor type and any limitations. Ask about any environmental conditions that affect accuracy and note it.
2. Find out how the device connects to a network: Note the connectivity type on your sketch.
3. Determine what data messaging protocol the device uses to communicate: Note this on your sketch. A simple example follows:

4. If you can visit an actual site where the device is in use, do so: Take pictures. If not, have someone that travels in the field take some for you. Try to get a few different locations. The idea is to capture real-world conditions for later reference.

5. Combine the real-world pictures and the sketch into a reference document: Pin it to your cubicle wall for reference; you will probably need it.
6. Analyze your data to infer some things about it:
   1. Take a few months worth of complete data. Keep it small enough to easily work with but large enough that you have some history.
   2. Calculate the frequency between records by subtracting the time of the previous record on the same individual device from the time of the current record. This will get an interval number for each message.
   3. Randomly select at least 30 individual devices and set aside all the records for them.
   4. For each device, visually review the records. If the interval number is consistent for the most part, you now know it has a regular frequency of reporting and the length of time between reports. If it is not consistent, there is probably not a set interval and data is communicated when the device gets in range of a network (it moves) or an irregular event triggers the communication. Either way, note this on your reference document.
   5. Look for irregular increases in the interval numbers. If this happens frequently, the devices are likely in a lossy network situation. Note this on your reference document.
   6. If the interval is large, several hours or days and the message size (number of columns) is low, then your IoT device is probably constrained and dependent on a long-term battery. It is more likely to be in a lossy network situation.
   7. If the interval is small, every few seconds, or minutes, then it probably has easy access to power. It is less likely to be in a lossy network situation.
   8. Calculate the average interval by date for each device. Plot them on the same chart, each one having its own line. If the changes in the average interval happen on the exact same dates for all the devices, you have a network or data processing problem with the centralized data collection process.