Learning to Work with Machines That Learn What the Experts Know

It is no secret that future generations of managers are being asked to learn all about the different processes in the end-to-end supply chain, and be able to come together with others in a fast action plan when a disruption strikes. This capability in turn requires older generations of managers, often labeled “subject matter experts” who are willing to serve as key informants to educate these younger folk, and who are willing to transcend typical functional barriers between sales, operations, and procurement, to offer insights into the right solutions to different types of business problems.

Today, such experts often exist in a vacuum in a different geography or location within large organizations. And people are often hesitant to bother these experts, or in many cases don’t know who they are, nor how to ask the right question that helps to access the specific gray matter between the ears of these experts for help with a specific problem! In an ideal world, an individual would “ask the system” how to solve a problem that is new to them, and be directed to the learning materials or individual to advise them on on how to deal with that situation. In reality, this situation poses some unique challenges. One of the interesting observations from experience is that older managers are occasionally more hesitant to share information, as they worry if they give it all away, they will be made “redundant”. Younger generations have been observed to “share faster”, but often don’t have that much knowledge to share to begin with.

Systems such as IBM’s Watson are being designed to help capture knowledge through resolution room activities, that can lead to a learning capability around situational intelligence. Over time, as the system learns, it can be used to better train young professionals based on the expert work of more experienced supply chain professionals. For instance, Watson might be able to make a recommendation based on an observation that “The expert looked at this report, this set of data feeds, and requested these inputs when they encountered a similar situation.” In this manner, such cognitive learning systems can emulate the footprint of a company’s best and most knowledgeable experts, and capture them in a system.

A critical skill that future generations of supply chain managers will need to embrace is the ability to interact with a cognitive learning system such as Watson. The popular press acknowledges that in order to drive to a truly transparent, real-time, and cognitive environment, individuals will need to fully comprehend the need for an end-to-end view of the supply chain. In the past, individuals have been trained to optimize in their own business function, leading to “silo’ed” views of the function. Optimized decisions in manufacturing would lead to inventory shortages or surpluses in other parts of the chain. Sales forecasts that were designed to avoid “stocking out” created inconsistent signals for others in the chain (the well-known “bull-whip” effect). But moving to a broader view also requires that individuals be able to absorb much more information from a diversity of sources. End to end views produce a plethora of data that is overwhelming!

Clearly, cognitive learning systems are not just about creating a data science experiment. Rather, they involve enabling a dynamic workforce to embrace change and complexity, and be able to quickly learn from both machines and other humans in the supply chain. The machine-human interaction will become even more critical, as humans learn how to scale and work with technologies like Watson.

Because cognitive learning systems have the ability to learn and learn faster than humans, technologies such as the Watson “knowledge studio” can create domain-specific models that feed into a Natural Language environment. These studios will benefit by having “super-users” interact with them, forming “playbooks” that provide guidelines on decision-making in the face of uncertainty. These playbooks reflect common processes that might arise under different conditions in specific industries. Examples include order to cash, facility turnarounds, sales and operations planning, missing a delivery promise date, distribution to customer transportation planning, customer demand planning, logistics optimization, and other “foundational playbooks”. But IBM is also conscious of the fact that every client will have their own unique supply chain business model, constructs, and culture. So while such playbooks will provide a starting point for use, a module known as “Watson conversation” will customize this to learn each individual’s context when he or she is interacting with the system. If an individual is working from a different role, whether in shipping, logistics, or procurement, they will be coming at the system from a very different frame of reference.

This evolution of the human-machine interaction has the potential to be very powerful, as Watson can eventually become an advisor to individuals in the supply chain. This is about having the platform “wrap around” the individual, and learning how they interact with the system, creating cohort mapping systems, identifying data sources, and making that individual more effective. This can lead to improved productivity, with individuals no longer having to comb through mountains of data to make a decision, as well as helping new individuals to adopt to a new role. The ability of systems to emulate at both the organization and individual level will require a new type of manager, one who is proficient in working with machines, and able to adopt their queries and thinking to exploit the data present in the ecosystem.