One of the big promises of IoT is supposed to be insight. The idea is that, by collecting all kinds of data from a myriad of connected sensors, both businesses and consumers will be able to learn more about the systems, devices, and environments around them.
The key component to bridge the gap between data and insight is, of course, analytics. Big data analytics has been a major buzzword in the IT world for the last five years or more, and it’s the key to generating the sort of knowledge we’re all hoping IoT can enable.
The problem, or challenge, is analytics is a complex topic few people really understand. (frankly, it’s a topic that can, and does, mean a variety of different things to different people.) Analytics for IoT is the sort of vague software “magic” that has many people salivating over the potential of what it can do, without necessarily looking at the reality of what it has actually achieved.
The theory is that you pump IoT-generated data into the black box of an analytics engine—most likely hidden on some unknown server in the cloud—and you’ll get a continuous stream of insights fed back to you.[pullquote]Analytics for IoT is the sort of vague software ‘magic’ that has many people salivating over the potential of what it can do, without necessarily looking at the reality of what it has actually achieved.”[/pullquote]
While there may be a system or two that comes close to this ideal, it seems that, at present, this is more the exception than the rule. Instead, there are a fair number of cases in which a significant amount of sensor-generated data gets fed into some sort of pattern or rule matching tool and, at best, the output is only modestly useful data points.
Part of the problem may be inaccurate assumptions or expectations about what’s really possible. For one thing, I think many people assume analytics projects are essentially never-ending–if you keep feeding data in, the results will keep coming out. In reality however, many are finding that, while analysis of IoT-generated data can create some solid insights, it’s essentially a one-trick pony.
For example, in the widely discussed story of connected cows—where female cows were fitted with pedometers and Fujitsu researchers discovered that, when they go into heat, they exhibit a particular walking pattern—the result was an extremely positive increase in insemination rates. It’s a great insight, but once the analysis was done, all they had to do was look for that pattern and then take appropriate action. Mission accomplished.
Similarly, the kinds of automated HVAC systems that have been integrated into “smart buildings” for some time can track the movement and density of people in a building and adjust settings accordingly. It’s practical and useful, but not necessarily the profound outcome many people seem to associate with the analytics of IoT.
The example of the connected cows also highlights other common misconceptions around analytics and IoT. For one thing, it’s not always big data; it can be little data—as in the footsteps of a herd of cows. Therefore, it’s possible the “analytics” can be done directly on an endpoint device and don’t necessarily have to be done with big server hardware somewhere in the cloud.
One could argue wearables with integrated sensors could perform these kinds of actions themselves. Yes, they could compare their own data to a set of data retrieved from the cloud, but they also could be built as a closed-loop environment. While there are certainly arguments to be made to keep things like wearables more open, there’s no denying the reduced security risks in a closed loop versus an open one.
Analytics in the IoT world is still evolving and I look forward to the interesting applications created over the next several years. Nevertheless, I think it’s critical to keep expectations in check because there’s no guarantee analyzing IoT data is going to generate useful, much less earth-shattering, information on a regular basis. In fact, we will likely see many more data dead ends than insightful vistas in the years to come.