Analyst Insight: Most forecasting models and AI approaches used today are considered statistical models, which require large amounts of high-quality data to build the model. Often, data is difficult to find and obtain. This limits the ability of today’s tools to quantify the operational impacts of supply chain risks for ranking and prioritization. Structural models focus on causal relations and connections (i.e., cause and effect), which provides a no-data or low-data option for building models.

As companies try to look into the future to make strategic decisions about risks and issues in their supply chains (e.g., disruptions, transportation constraints, labor strikes), the most common approach used today is the statistical modeling approach. Although this approach has served companies well in the past, the VUCA (volatility, uncertainty, complexity, ambiguity) environment in which we now find ourselves has exposed two major limitations with statistical models when it comes to long-range, strategic what-if scenario planning and analysis of risks. These limitations prevent companies from ranking and prioritizing the long list of risks that have been identified within their supply chains. Consequently, companies cannot ascertain which risks to focus on first and which risks, if any, can be ignored because they have no significant impact at all.

Statistical Modeling versus Structural Modeling

Statistical modeling and structural modeling are almost polar opposites. Statistical models treat the process or system itself as a “black box,” with no concern for the activities and decisions within that process or system. Statistical models only look at the data coming out of a process or system. Hence, data is the core requirement for building statistical models. On the other hand, structural models look at these activities, decisions, policies and other causal connections within the process or system to build the model. With structural models, data is brought in to validate the model or calibrate the model. Data is never required to create the structural model. 

Dependence on Data and Disregard for Causal Connections. 

Because data is required to build a statistical model, if no data is available, then no model can be built. As a result, statistical models cannot forecast for situations that are new, or that have not been encountered previously. Many of the risks within our supply chains are new and unexperienced. Hence, statistical tools are no longer the appropriate tools for forecasting the operational impacts of these risks, and they are no longer appropriate for ranking and prioritizing these risks for action. Moreover, with no knowledge of causal connections, statistical models do not provide effective long-range forecasts. Instead, statistical models are only valid for short-term forecasts.

With the inclusion of causal connections and the independence from data for model building, structural models tend to offer a better approach for understanding the long-range implications of risks or situations that may be new or unheard of, because the causal structure captures the decisions and activities that would occur over the long-term, regardless of data.

Outlook: Companies must adopt structural models to use alongside their current statistical models to better understand the operational impacts of various risks throughout their supply chains. This adoption will allow companies to create and analyze solutions when there is insufficient data to feed and build the statistical models. Furthermore, structural models will allow for more effective long-range, strategic what-if scenario planning.

Resource Link: www.scmblox.com