Business process management is not at the center of an IoT strategy

Solution CTO, IBM

I was in a debate recently on LinkedIn regarding whether or not business process management (BPM) was the single most important (and hardest) part of leveraging the burgeoning Internet of Things (IoT). The person I was debating worked for a BPM specialist and was arguing that BPM was the center of an IoT strategy, as well as the hardest part of pursuing an IoT strategy. While it is easy to understand why he would have that point of view, the problem is that, in my experience, that stance is simply wrong. The reality, which still holds an important role for BPM, is a bit more complicated and broader than that.

Before I go any further, let me explicitly say that I am not deprecating the importance of BPM in a modern enterprise—far from it. BPM engines and the related technologies have a key role to play in how modern enterprises run, they are critical in the day-to-day (often minute-to-minute or even second-to-second) decisions that keep the lights on. What is happening broadly, however, is that we are seeing a wholesale change in how those decisions are made because we are seeing an overall change in firms moving to utilize data far more holistically and aggressively than ever before. As firms transition to data-driven decision-making, a far greater number of sources than ever before are being considered, and firms are struggling with how to move from reporting to business decisions based on predictive analytics. In addition to all of that being a huge amount of work, that shift to predictive analytics fundamentally changes the whole lifecycle of how you make decisions and, thus, the role and scope of  how a BPM engine fits into the overall picture.

To better explain, let's take a look at a day in the life of the sensor and what is needed before you can make smart business decisions based on the data it produces:

  1. Whatever you want to instrument needs to be designed and built including the sensors, and, of course, it needs to have a mechanism for the sensors to send their data home
  2. You then need to catch and accurately record the data that the sensor is sending, which often means creating new and highly scalable distributed environments and platforms especially when you want to make low latency decisions

tom d bpm iot.jpg

Ideally, as the data is being written out, you need to understand if there's anything in the sensor output that requires immediate action, but how does one know if action is required? This requires an understanding of what is normal, what is trending towards not normal and what data is irrelevant; knowing that requires a holistic understanding of many operating factors, and, in turn, requires robust statistical modeling to understand all the data involved as well as the data’s relationship to the operating environment and desired outcomes. In other words, you need to do non-trivial data science to figure out how to best collect and respond to the senor data before anything is stored.

Once immediate action is determined, you need to write the data out for storage, making sure it is secured and prepare it for deeper analytics, which often means combining it with multiple other sources; this requires competency in handling the data and related metadata property so you get a good build and correctly build a superset of data that has all the features you need for your data science.

Simply having a good data build isn’t the end of the challenge; you now need to make sure what you have put together is not subject to bias issues that would taint your models. Then, you have to start building models, and that often requires data science skills which are not abundant—you may even find that your firm doesn't have these resources available at all.

After the building, you then have models that need to be tested. Validating your data science requires rigorous and iterative testing by flowing real world events through them. Only then can you step back and look at how the implemented models (which may be executed by a BPM engine) are actually performing, and if it is an improvement over how things were being done previously. If it is an improvement you can then implement it, but if it isn't, you can back up through all the process laid out above and have another go at it.

It is also important to keep in mind that the real world is not static, and that means that this whole process needs to be frequently reevaluated and results challenged to make sure we are learning and evolving as quickly as the world around us. 

So, is BPM important? You bet. But, it isn’t at the center of the IoT.