Understanding Skynet: Internet of Things vs. Industrial Internet
My friend Jeff Frick (@JeffFrick of SiliconANGLE) was educating me recently on the difference between the Internet of Things (IoT) and the Industrial Internet. That conversation persuaded me to research further the difference between these two concepts and to contemplate what a supporting solution stack to enable a next generation of applications from these two trends might look like.
First, let’s start with some definitions:
- The “Industrial Internet”—or machine-to-machine (M2M) communications—describes machines, nodes or devices that use network resources to communicate with a remote application infrastructure in order to monitor and control the machine or the surrounding environment. In other words, the Industrial Internet refers to all the devices, sensors, and software that enable connectivity between machines.
- The “Internet of Things” or IoT can be thought of as the application layer of what the Industrial Internet will ultimately become. The IoT refers to the potential life- or business-changing applications that the realization of the Industrial Internet will bring.
One way to think about this is that the Internet of Things sits on top of the Industrial Internet. The Industrial Internet generates a wealth of new M2M data—massive data sources in a wide variety of formats in real time. And the Internet of Things leverages the M2M data to enable the next generation of business applications. But we’ve seen this story before—where massive amounts of new data are mined for new customer, product, and operational insight that form the foundation for a new generation of business applications. Let’s review this history lesson.
Consumer Package Goods and Retail Industry: Lessons Learned
Note: the material in this section of the blog comes from Chapter 2 of my book “Big Data: Understanding How Data Powers Big Business”. See Chapter 2 for more details on this early big data lesson.
In the 1970s and early 1980s, Consumer Package Goods (CPG) manufacturers based their marketing decisions on bimonthly Nielsen store audit data. Nielsen would send people into a sample of stores to conduct physical audits—to count how much product was on the shelf, the price of the product, product sales within that store, and other data. The results of the audits were then packaged and delivered every two months to the retailers and CPG manufacturers, usually in booklet format.
Then in the late 1980s, Information Resources Inc. (IRI) introduced their Infoscan product, which combined retail POS scanner systems with barcodes (universal product codes or UPCs) to revolutionize the CPG-to-retail value chain process. Data volumes jumped from megabytes to terabytes. Mainframe-based Executive Information Systems (EIS) broke under the data volumes, which necessitated a new generation of data processing and analytic products.
But the most interesting and relevant aspect of this POS scanner data revolution was how companies like Procter & Gamble, Frito Lay, Tesco, and Walmart were able to make use of this new source of data and new technology innovations to create completely new business applications such as demand forecasting, supply chain optimization, trade promotion effectiveness, category management, loyalty programs, etc. (see Figure 1).
It wasn’t the data or the analytics that were ultimately game changing. It was the new business applications that were enabled by leveraging this new data source and new data management and analytic capabilities.
The Internet of Things Yields New Business Applications
The real benefits of the Industrial Internal and the Internet of Things won’t be realized until leading companies develop the next generation of applications that address specific business needs from this wealth of data. It is within the Internet of Things that we’ll see a new generation of applications, such as:
- Predictive Maintenance: predict when and how a device will fail and what replacement and maintenance parts and service personnel skills will be required to preempt the failure
- Loss Prevention: monitor device and network usage to flag unusual usage situations that may be indicators of revenue loss and theft
- Asset utilization: monitor and predict asset utilization under a number of different usage scenarios in order to improve asset, device, and node utilization
- Inventory tracking: monitor inventory levels and inventory assets to minimize loss and waste and improve inventory utilization
- Disaster planning and recovery: model different disaster scenarios and likely device and network usage requirements to proactively plan for disaster situations (e.g., hurricanes, ice storms, tornados, earthquakes, Zodiac Killers)
- Downtime minimization: leverage predictive maintenance and inventory tracking to identify high probability downtime situations and ensure that the right maintenance and replacement parts are available, as well as the right skilled service personnel
- Energy Usage Optimization: optimize energy usage given current and historical device performance, historical and predicted energy costs and device performance requirements
- Device Performance Effectiveness: monitor and optimize individual device performance/through-put based upon historical performance given certain workloads and environmental conditions, coupled with a detailed profile of the performance behaviors of that device or node
- Network Performance Management: monitor and manage/fine-tune the performance of a network of devices given current load, required performance requirements (service level agreements), and forecasted performance requirements
- Capacity Utilization: reallocate device resources and jobs to optimize network and device performance given the history of device interactions and current and forecasted performance requirements
- Capacity Planning: predictive and prescriptive analytics that model product and device usage and in real time, makes resource allocations and automates the provisioning of new capabilities (turning on and off capacity as dictated by the predictive models) in order to ensure the required capacity at the optimal prices
- Demand Forecasting: leverage device behavioral models, actual usage patterns and trends and external factors (weather, traffic, events) to forecast longer-term network configurations and product and network build out
- Pricing Optimization: understand device usage patterns, coupled with demand forecasting, to optimize device and network pricing—lowering pricing when demand is low and increasing pricing when demand is higher, almost like surge pricing, but hopefully without the same customer satisfaction issues
- Yield Management: optimizing device and network usage to extract the most value out of the overall network capacity and capabilities
- Loading Balancing Optimization: balancing usage load in light of forecasted demand to ensure that all nodes are being utilized equally and to avoid performance bottlenecks
To create these and additional business applications, there must be a data management and analytics layer between the Industrial Internet (the layer that is generating all the machine-to-machine data) and the applications that are part of the Internet of Things.
Data Management and Analytics: The Business Transformation Layer
The data management and analytics layer is necessary to transform and enrich the M2M data, integrate new data sources, and build, test and continuously fine-tune the analytic models that underpin the new generation of applications (see Figure 2).
The data management and analytics layer would need the following functionality in order to build these next-generation IoT business applications:
- Data Repository: This layer, often known as the “data lake,” is the repository where all the M2M data gets dumped “as is”. No assumptions are made about how the data will be used so no data schema and no constraints need to be built prior to loading the data into the data lake. The business applications are based upon the questions the application is trying to address and the insights needed to answer those questions. The application will define the data schema “on the fly” as it accesses the data.
- Note: It is at this level that master data management (MDM), data governance and security issues need to be addressed. Privacy issues probably need to be addressed at the specific application layer. Organizations will need a cross-application “decision governance” framework that explicitly dictates and governs when and how personal information and resulting insight are used, as well as identifying situations where the information and insight are not used.
- Data Transformation and Enrichment: This is the heavy lifting set of functionality that profiles, cleanses, aligns, transforms and enriches the data. New data enrichment techniques and algorithms can be developed to create new metrics associated with frequency (how often), recency (how recent) and sequencing (in what order). Those metrics may be better predictors of device or network performance.
- Analytics: This layer would have the wide range of analytic and programmatic capabilities necessary to create the analytic models, including:
- Descriptive Analytics: the “what happened” analytics that glean insights about what happened. Descriptive analytics analyzes past performance data and look for the reasons behind past success or failure (e.g., trending, moving averages, comparisons with previous periods). Most management reporting uses this type of retrospective analysis.
- Behavioral Analytics: the “what is happening” analytics to uncover (codify) device, product and network usage characteristics, tendencies, propensities, patterns and trends. This area of analytics would include graphic analytics to understand device and node behavioral characteristics in light of the overall network relationships and interactions (e.g., understanding relationship direction, strength, frequency, and recency).
- Predictive Analytics: the “why did it happen” analytics that provide the reasons why something happened and form the foundation of predicting future performance and behaviors. Predictive analytics answers the question about what will likely happen (with some degree of confidence). Predictive analytics uses rules and algorithms to determine the probable future outcome of an event or the likelihood of a situation occurring.
- Prescriptive Analytics: the “what should I do” analytics (e.g., recommendation engines, next best action) that deliver prescriptive, actionable recommendations that lead to direct action. Prescriptive analytics leverages multiple disciplines of mathematical sciences, computational sciences, and business rules, to suggest decision options to take advantage of the predictions (recommendations). Prescriptive analytics can continually take in new data to re-predict and re-prescribe. This automatically improves prediction accuracy and makes for better decision options (recommendations).
- Application Interface Layer: This layer provides the application interfaces that applications employ to access the insights gleaned out of the analytics layer. This user interface layer could include the more traditional reports and dashboard elements, but more likely will be a standard set of Application Programming Interfaces (APIs) that the applications will use to gather and deploy the insights that will be used in an application specific to the business process. Examples include predictive maintenance or capacity planning.
There is no doubt that the Industrial Internet and the Internet of Things hold the potential to dramatically improve the operational effectiveness of many key business processes. And we’ve seen this story before, when the CPG and retail industries leveraged massive new data sources to enable a next generation of business applications. Now we’re about to see organizations leverage the wealth of new M2M data—massive data sources in a wide variety of formats available in real time—to create the next generation of operational applications that are capable of optimizing key business processes and uncovering new monetization opportunities.