Data in Action: Capturing It, Structuring It, Modeling It, and Putting Data to Work for You

(Source: nmedia/Shutterstock.com)

Aggregating data from various sources is prologue to building data models and simulating conditions. Such simulations help process and learn data, and is at the root of artificial intelligence and other tools. Such tools drives decisions and builds efficiencies in unprecedented ways by putting data to work for us.

Data may be very available. For example, a courier service has much structured data available to it: weather, delivery time, damaged goods, geography of a parcel delivery, and so on.

Data is more available when we own it. If data is also proprietary it takes on a new and unique value. It is ours to use and capitalize on however we wish.

There’s also data in the public domain. For example, visit www.data.gov and one can find hundreds of databases. You can find data about electric current and how it is delivered throughout the world; state-by-state government tax collections; a detailed listing of operating nuclear power reactors in the U.S.; and hundreds of other data sets. This data is open source.  Programmers and data scientists may compute the open source data, and transform it into models, dashboards, and other formats to create unique and proprietary intellectual property.

To effectively put data to work for us, we must properly structure it as with the aforementioned data extraction used in contact tracing. A suitable data architecture is required to interact with other data elements and programming.    

The process by which data interacts with programs and other data is at the core of building a data model. A model is systematic and follows a specific framework.  The model is driven by programming commands, guidelines, and rules. Data frameworks are the basis for data modeling and allow us to apply analytics, and predictions, so we may be best prepared for likely scenarios.

For instance, Siemens (who makes control systems for industrial applications) describes a scenario where an automotive manufacturer finds that a power window system for one of their vehicle product lines is defective and does not stop if a hand is in the way of its closure (Figure 1). The problem is complex and includes data from numerous functional disciplines such as electrical and mechanical engineering, product management, quality assurance, and software development. All functional areas have a stake. But they also have different data and processes.

Figure 1: An example of a manufacturing simulation from Siemens. (Source: Siemens)

A data framework may be created to understand how they all interact so as to reverse engineer the defect to find where it’s failing. So a data framework, in this case, isn’t only used to find an optimal solution. It may also be used to find the root cause of a defect, through reverse engineering, by working backward through the data model.

Data models are used to process data found in data frameworks and structure. They help support business processes, iterations, and simulate an endless number of real-life scenarios. Therefore, the process of data modeling involves professional data modelers working closely with business stakeholders, as well as potential users of the models.

Creating a data model begins with a conceptual outline of the specifications and objectives sought from the model, sometimes referred to as a conceptual model. Once established, the conceptual model may be further refined by representing a “domain” of data and is sometimes referred to as the logical or domain model. It represents the domain of data in an abstract structure.

So, the logical model builds out the conceptual model with a specific data structure. Examples of data structure include tables that indicate relationships between one data element and another. This may be demographic information such as a person’s age, weight, nationality, etc. The data may also be XML tagged, learned, and used in programming applications.

The next step entails bringing all data together into a design or physical model. A structure may be established–for example, as a table with rows and –columns—to organize the data. The relationship between data within the tables may be further established.  One table may link to another table.  The physical data model comes together as the actual data design.  It may use relational database management systems (RDBMS) to compute and relate the data to complete the design and produce the desired output of the model.

To illustrate: when a mostly online apparel retailer utilized data, the owner was able to open up a temporary “brick and mortar” store, pinpointed to a data-driven target market. The Wall Street Journal reported in 2019 that she “used a combination of data, including ZIP Codes and sales data, to determine which section of San Francisco was the most popular with shoppers. She analyzed sales data of top-performing styles, colors and complete outfits to know what to put on the shelves—more luxe wool mid-length coats, pullover sweaters and silk collar tops, for instance.” The model allowed the proprietor to create a seasonal, physical store, based on data artifacts, to supplement the online store.  The model did the thinking and drove the decision of what to sell off-line, when, and how.

Decisions like these in retail, as well as in many other industries, may be greatly aided by modeling and simulation. Modeling and simulation use the logical representation of a system and its accompanying data. Representations include systems, machines, processes, and other entities through computing. Computing includes the application of different software to create outcomes and visualizations, based on programmed instruction as well as iterations.

The advantages of modeling and simulation are evident in so many different applications. For example, a manufacturing assembly line, using robotics, may be 3-D modeled using data to simulate a manufacturing cell. This simulated factory model may be prototyped before it’s implemented to work out kinks in the system, perform “what if” scenarios, and gain useful intelligence that will ultimately make the manufacturing cell more efficient.

The manufacturing cell, once implemented, may capture data such as a machine’s operation, the quality of the products it produces, the speed of output, and other information to monitor production and detect problems before they occur. This allows manufacturers to make decisions, bring equipment on and offline, and change things like speed and inputs to affect quality and output.

Data models may also be used to monitor live operations, aided by the Internet of Things (IoT). IoT denotes the transformation of data via the internet, from connected devices such as cameras and sensors—in a manufacturing setting or elsewhere—and continuously feeds such data into a data model that may be used to prototype systems and measure and monitor performance.

The evolution of using data to build models is still in early adoption.  Modeling is improving due to better data, more available data, and more experience in testing, migrating data, and better algorithms.  Better modeling means better performance and more solutions.

In conclusion, our digital world is progressing through smart and intelligent usage of data. Such models may be combined with predictive analytics and artificial intelligence methods to improve performance and build efficiency. Data, from a variety of sources, can be used to incrementally create a physical data model in so many different ways.  Simply stated, data through its many applications, creates new possibilities and solutions like never before.