Published on 03/10/2017 | Strategy
If the classic movie The Graduate were written today, instead of "plastics" as the key to future success, our young protagonist would hear one word: "sensors." It has become gospel that sensors will soon be at the heart of everything, with Business Intelligence sitting above analyzing their data. Perhaps. New technology can certainly be wonderful, but it’s most critical to have a deep understanding of one's industry. The past two decades have seen a significant investment in sensors and analytics, made on the promise of clean and actionable data that is relevant to specific business processes. The sensor ecosystem is far from a homogenous market-space, which explains the significant confusion from authoritative sources in whitepapers, conferences and blog posts. There are horizontal and vertical plays being mixed together and IoT adds another layer of confusion. On top of that, we must distinguish between IoT, iot, IoE and IIoT.
Looking at a specific example, the Chemical Processing & Manufacturing market has been utilizing sophisticated sensors since the late 80s. There have been frequent attempts to layer the same approach over Upstream Oil & Gas without any consideration as to whether it's appropriate for that context. Upstream O&G has a much different return curve associated with its assets than do Chemical & Manufacturing plants. Upstream has a drastically declining curve, which means the capex and operating model which are typically based on lease-operating expenses (LOE) must be kept low. Companies that have poured millions of dollars into SCADA systems for monitoring assets trying to follow the Chemical & Manufacturing approach have seen very little comparative value in return because of the decline curves associated with wells. The SCADA assets stick out like sore thumbs as the Price per Barrel of Oil Equivalent (BOE) drops over time.
This has prompted some organizations to try a different approach. We spoke to a mid-market company that prototyped a Machine Learning solution in conjunction with experienced field personnel and front-line engineers to improve uptime, predict maintenance challenges and increase throughput for a client's field of Electric Submersible Pumps (ESPs). The prototype's outcomes changed the way the ESPs were operated, reducing costs and significantly increasing each well's production by optimizing operating parameters using the existing control systems. The vendor was able to price based on a cost per ESP instead of a standard licensing deal. Their client quit entertaining new Analytics vendors with the latest interactive graphs to focus on supporting their existing systems, which were now being fed smarter information based on predictive models developed with Subject Matter Experts. [Note: This vendor solution was provided by OspreyData – and in full disclosure, I’m on the Board of Directors].
Why haven't more IoT companies combined sensor data with machine learning and existing human expertise under an economic LOE model to match their target industries? The reason is that sophistication in Data Collection and Data Science and the application of that information to industrial data science is incredibly weak in the market today. These missing technologies are what is needed if IoT and especially Industrial IoT is to flourish. Everyone talks blithely about a $30B, or even $40B IIoT market by 2020, but without a new-paradigm solution, what we are really talking about could be little more than marketing hype and hyperbole instead of profitable reality.
There is already massive demand for M2M intelligence and its ability to connect Physical Systems and share real-time data among users. This is not terribly new (though check out SWARM Engineering for a novel approach to this) but more critically, what is done with the data has not changed for many years. The 1990s saw the emergence of Business Intelligence tools like Cognos, Crystal Decisions, BusinessObjects and others. These gave IT a post-mortem view of the data and its historical relevance. In the years that followed, we saw an explosion of technologies to enable business users to format, analyze, forecast and plan. We began to see the deployment of vast computing power to end-users, a trend that has gathered pace recently. But are these tools capable of tapping into the value of the astronomically Big Data physical sensors will provide?
My belief is that Business Intelligence must evolve into something I call Physical Systems Intelligence (PSI). PSI combines Machine Learning done on physical sensor data with human intelligence and intuition to allow organizations to learn from their Subject Matter Experts and field operators along with their industrial devices and machines. The augmentation of expert human knowledge with embedded machine-learning systems is where IoT systems must go. This initiative follows legitimate research around AI, Machine-Learning, Deep Learning, Reinforcement Learning, Expert Skill Acquisition, Intuition vs Rationalization (Hubert & Stuart Dreyfus, et.al.) Interestingly, the list of buzzwords surrounding this topic seems much longer than the list of first-rate and validated tools that move us from BI to PSI.
Optimizing industrial systems requires the creation and maintenance of many complex predictive models, and modern AI has only recently caught up with the explosion of sensor data. Traditional techniques, using general purpose modeling tools, or trying to convert BI tools and programming languages to solve the IIoT challenges, simply does not work. As I write this, TV is awash with ads for the movie "Deepwater Horizon", a sobering reminder of what can happen if you have the sensors and data but not enough real-time and actionable information that has learned the consequences of certain states and conditions.
No matter what industry you are in, IIoT will be a powerful reality. The edge - the sustainable competitive advantage - will go to those organizations and individuals who can adopt Physical Systems Intelligence successfully, which means they must:
1. Integrate expert human knowledge with big data analytics. Firms that can do this are grabbing control of niches (e.g., Proskriptive in healthcare).
2. Capture and use deep industry knowledge. What are industry-specific economics? What are the triggers that may alter the environment? What semantic meaning does this data possess?
3. Execute effectively in the field. The DevOps movement tells us that there is often a surprisingly steep learning curve for implementing tools that integrate across complex elements. In the next blog post, I will look at how DevOps is evolving, and what you can do to accelerate this.
Yes, the one word for the future may be "sensors," but the race will be won by the organizations and individuals who can move from Basic Monitoring Systems to Physical Systems Intelligence, where applications can prescribe and predict humans to act in a deliberate manner to resolve and optimize operations and performance and embrace the human-focused actions in the three points above.
Please follow me on Twitter! I would love to hear your related #IIOTchallenges in the comments and on Twitter @mahoneymiles.
This article was originally posted on Linkedin.
