Dow, Inc., one of the largest material science companies in the world, has been at the front of a digital transformation before the term was in common usage. The company had a RFID center of excellence (COE) in 2006, which expanded into a push to get good end to end visibility of their supply chain. The COE concept was also useful for building an innovation culture around these digital technologies.
I thought what Dow was doing back then was ahead of the curve. But John Wassick, a Supply Chain Technology Fellow at Dow Chemical, told an audience at ARC’s Digital Transformation Forum in February that “traditionally R&D spend focused on product and process innovation. But I’m happy to report now that leadership is looking for innovation in the supply chain space to the point that it is disruptive and that it produces competitive advantaged intellectual property. We have a vision to approach process innovation in our logistics assets like our commercial businesses innovate their assets.”
Applying Process Automation Tools to Order Fulfillment
Dr. Wassick is working to apply process automation to the order fulfillment process. In a conversation a few weeks ago I dived into that topic with Dr. Wassick in more detail than was provided at ARC’s forum. The quotes below come from both the speech and my conversation with Dr. Wassick.
Dow, like many companies has started the journey to implement robotic process automation (RPA) for business processes. RPA is process automation software that is used to automate high volume, repeatable tasks. A robotic process automation solution can be used to automate an end to end process that requires the use of multiple systems, internal and external websites, and portals with real-time data acquisition and integration. RPA automates tasks by mimicking what a human does with a user interface, performing the same computer keystrokes and opening the same applications that humans do. Alternatively, an RPA can automate a task by writing data directly into an application’s database.
Dr. Wassick points out, however, that the process automation Dow is investigating and developing extends to higher level decisions for which robotic process automation is not well suited. “So, RPA is only one component of our automation toolset,” Mr. Wassick explained.
At its core, Dr. Wassick points out, a supply chain is “focused on material flows – material into plants, to our warehouses, to customers’ sites. But behind this is a transactional process with many decision makers passing information back and forth.” The transactional process also must work for material flows to be efficient and high levels of customer service to be achieved.
Before achieving full automation, Dow needs to understand the complexity surrounding their order fulfillment transactions. Dow used a process mining tool to “bring to life in graphic fashion how our transactional processes behave.” The tool extracts data from their SAP solutions to examine the paths an order can take from the time it is created until goods are loaded on a carrier’s vehicle. In the diagram below, you can see the different paths an order can take. “If things work well, we get an order, create a delivery note, and issue goods.” But many things can happen that cause nonstandard transactional flows surrounding the order. These nonstandard transactions can “cause delays and customers end up being less than happy with us.”
Applying Batch Chemical Principles to Order Flows
In the diagram above, on the left you see the order paths as documented by the Celonis visualization tool. But Dr. Wassick noticed that the order flows resembled the batch processes in chemical plants. An example of a batch process can be seen on the right in the diagram above.
Manufacturing processes can be classified as continuous, discrete, or batch. Continuous processes are classified as processes which have a continuous outflow; for example, oil produced at a refinery. In discrete processes it is possible to trace how individual components are combined to produce a finished good. Examples of discrete processes would be the manufacture of cars or furniture. With batch manufacturing, parts of the process are discrete and parts are continuous. Dow Chemical and other chemical companies had been very successful in automating batch processes.
Dr. Wassick’s insight is that the methods they used to automate batch manufacturing could be applied to order fulfillment transactions. All that experience Dow has in batch control can be useful. By applying that methodology, Dow can approach transactional process automation “in a more structured manner.”
In the early 90s a standard – ISA S88 – was developed that established a process, terminology, and concepts for controlling batch processes. One part of the standard is focused on models and terminology. This portion of the standard has four key concepts:
1. How to model the plant (referred to as the physical model). For example, we could have a physical model of a house. A house has a certain number of rooms. A room might be a kitchen, office, or a bedroom. A room contains furniture. Furniture might be a table, a sofa, or a chair.
2. How to define what you would like to accomplish in the plant (referred to as the recipe). This is analogous to the recipes we use to create food that specify the materials and procedures to be used.
3. What actions can be performed in the plant (referred to as the equipment logic). The steps defined by the recipe are executed with equipment. If baking a cake, the logic is simple. Set the stove to 300 degrees, wait for the oven to reach that temperature, and then bake the cake for twenty minutes. In chemical plants, the equipment logic is much more complex.
4. And finally, the standard defines how the three pieces mentioned above need to work together to achieve successful automation.
“The chemical engineering community developed very comprehensive and formal methods to model a process, and then design a rule-based hierarchy to achieve real-time control of the process. All those methods can be applied to transactional processes,” Dr. Wassick explained. “In our plants, operators are in a control room managing the overall process. They are not looking at the position of valves or the temperature in a piece of equipment. We want supply chain planners managing the overall supply chain network, not just individual pieces of the process.”
Further, “within process engineering, there is something called ‘process intensification.’” Process intensification seeks to combine multiple process steps into a single, more comprehensive step that can execute all the subtasks simultaneously. Used for transactional processes, “This reduces manual interventions and handoffs because there are fewer steps to manage.”
There is also mathematical rigor associated with describing the behavior of a chemical process. “Chemical engineers use physical differential equations and other math constructs to model physical chemical processes.” These mathematical constructs won’t work for modeling transactions. “But there are obvious paradigms to draw on.” Standard deviations around how long a process step should take can be developed. “If transaction A and B occur, C could happen and that would be really bad.” Markov chains and Bayesian inference could be used to describe such behavior.
“When an order comes into the company, and we understand the attributes of that order, and how the attributes fits (flow path) probabilities, we can predict how long it will take to fulfill the order. Then we can take action ahead of time to mitigate potential problems.” Dow will also look to use AI and machine learning to help automate processes.
In conclusion, the RPA community might have a lot they can learn to improve their solutions by drawing on the methodologies developed by the chemical engineering community. It is also possible Dow will develop some valuable intellectual property in their quest to automate their own processes.