As data science and analytics teams continue to feel pressure to deliver more value from analytics, many organizations still struggle with the processes and technology required to deploy models into production and more rapidly make data-driven decisions. When evaluating how to best undertake these activities, organizations should consider an important distinction to determine the best path forward.
Human Or Machine?
The pervasive level of analytics required to reach Level 4 or 5 on IIA’s Analytics Maturity Model requires that a significant number of analytics products are integrated into business operations. These products can be used to help humans make a better decision in the context of their workflow. For example, a call center rep might see cross-sell offers prioritized based on the context of an interaction. Or, a supply chain manager might receive an alert that the forecasted price of a key ingredient has significantly changed so that they can take action.
Analytics products can also be integrated into business processes to guide the decision making of a machine. Examples include automatically underwriting a life insurance application or managing the dynamic pricing of airline seats.
The roles, processes, and technology needed to inform a human decision maker are very different than those needed to inform a machine decision maker. Exactly who, or what, will be making a decision should be established early on to ensure the final product is designed for the correct end user.
Many data science and analytics groups develop models while focusing primarily on the data, analytic techniques, and business problem. In a pilot phase, this is sufficient. However, there are significant differences between automating a decision for a machine versus informing a decision for a human. These differences need to be accounted for throughout the analytical development and deployment process.
The Differences
While the algorithmic approach may be consistent whether informing a human or a machine to make a decision, the supporting architecture, data provisioning, and process monitoring needs to be very different.
For example, when call center employees are delivering offers from a script, they are able to identify if something has gone haywire and what they are seeing makes no sense. They may or may not deal with the situation optimally, but they will notice quickly, report the issue, and attempt to adjust.
An automated machine-based decisioning process, however, will only be able to assess problems within the exact context and parameters that are programmed into it. The process may not even realize that something has gone haywire, let alone report it or try to adjust. Care must be taken to build in more validation logic and / or to have a human monitoring a process as it runs, much like how humans monitor manufacturing processes.
There is an inherent advantage to having many people observing the actions of an analytic process on an ongoing basis. Enabling human-based decisions provides ample opportunity for the organization to find issues. A machine-based process, on the other hand, will simply continue stamping out decisions regardless of an obvious-to-a-human problem arising. A much higher level of monitoring and results tracking is required when deploying analytics to facilitate a machine-based decision.
Another major difference between human- and machine-based decisions is the speed and scale of the decisions being made. While the number of cross-sell offers made in a call center in a day may be impressive, it would pale in comparison to the number of decisions being made by machines optimizing output within a complex assembly line. Machine-based decisions often occur at a rate that is orders of magnitude higher than human decisions. This necessitates a more scalable, sophisticated, and performant pipeline from raw data to recommended decision.
From Many To Millions
As the practice of analytics matures within a company, opportunities that drive significant value by informing human back office or strategic decisions are typically found early. The current frontier for most companies is not assisting a few decisions that create millions in value but assisting millions of decisions that each create a small amount of value. These small individual values can add up to a massive number in total and the pursuit of this frontier is the foundation of The Analytics Revolution currently underway.
While organizations must aggressively pursue analytics that inform both human- and machine-based decisions, the differences in the requirements for each must also be understood and accounted for from the beginning. The failure to recognize the distinction has led many organizations to struggle more than needed.
Bill Franks, Chief Analytics Officer, helps drive IIA's strategy and thought leadership, as well as heading up IIA's advisory services. IIA's advisory services help clients navigate common challenges that analytics organizations face throughout each annual cycle. Bill is also the author of Taming The Big Data Tidal Wave and The Analytics Revolution. His work has spanned clients in a variety of industries for companies ranging in size from Fortune 100 companies to small non-profit organizations. You can learn more at http://www.bill-franks.com.
You can view more posts by Bill here.