Simulation modeling provides dynamic platform for scenario planning
A three-part blog series on how simulation models aid different industries with predicting outcomes based on real-world context
The real world of business is an ever-evolving, dynamic environment. Digital transformation is about gaining and sustaining a competitive advantage and having the ability to be responsive to new business moments. This requires continuous intelligence and fast decision making. The prediction models generated by artificial intelligence (AI) and big data are point solutions based on historical data. Does it make sense to make decisions solely based on historic data, or should plausible future scenarios also be included? Leadership often wants to examine alternatives before choosing a new path. So, what can be done to capture mental models of leadership that also helps them visually examine consequences of decisions under different scenarios (context)? Business will always be dynamic, so does it make sense to make decisions based on static point solutions?
Modeling and simulation provide the methodology and technology to examine various scenarios and answer questions like these. Unlike point solutions, which assumes that structural characteristics inherent in historic data would remain the same in the future, simulation models allow us to create a model of the real word that can account for the “what if” scenarios that often occur in a dynamic environment and examine “if only” types of questions. John Sterman (MIT) refers to simulation models as “low cost learning laboratories.” No one would disagree that outcomes of a decision could vary depending upon the context or scenario. Simulation-based decisions are better than point solutions from an analytics perspective since they add context to predictive models, thereby providing a platform for examining prescriptive actions.
There are various approaches to simulation modeling. System dynamics is one of them.
In the case of operational decisions, data analytics provides models, which are evaluated for accuracy. For strategic and tactical decisions, where the implications will play out over time and space, simulation models are required. Unlike using models for accuracy, system dynamic simulation models are used to design and test policy. This methodology allows to account for time delays, feedback loops and potential unintended consequences. In analytics, models are evaluated for accuracy, whereas in system dynamics, models are used to rigorously map and design and road test policy. To illustrate this point, let us take a simple business problem of predicting employee turnover.
Inherent data helps up to a point
Human resources is trying to identify employees that are at risk of leaving. Among other attributes, HR managers have data sets that are comprised of employee performance, employee potential and summary of exit interviews. Analytic models provide managers with probability of an employee leaving, providing management an opportunity to take potential actions about what could be done to retain an employee. This is a point solution for operations that provides likelihood (prediction) patterns inherent in data.
There are many factors that affect employee churn, such as pay satisfaction, historical compensation growth, job satisfaction, work pressures, skill utilization and more. But these factors interact and do not occur at the same time in the employee’s life. In fact, the decision to leave is a result of cumulative effects on an employee. Analytics can help infer overall employee satisfaction, pay satisfaction, etc. at an aggregate level but cannot infer at an individual level. Also, metrics are not often in place to monitor employee workloads or to assess work-life balance. So, static point solutions aid the HR manager up to a certain point.
Answering “what if” scenarios with system dynamics
In today’s world of automating processes using insights from artificial intelligence and big data analytics, simulation models are the contextual equivalent when trying to predict future outcomes based on historical data. As an example, novel AI techniques to assess employee loyalty based on variables such as voice quality, enthusiasm and sentiments can provide additional information about an employee’s overall demeanor towards their job and company, thereby providing a better estimate of the odds about an employee leaving. Each person has an individual threshold; what is acceptable for one employee may be not applicable to another. But these individual quantitative, as well as soft qualitative factors (predictors), are not stand-alone; they interact and change over time. While predictive analytics provides correlations, it does not help identify leverage points for management to devise policies that would reduce overall turnover. This is where simulation using the system dynamics approach becomes useful.
Instead of focusing on individual turnover, management is really interested in improving overall financial performance by reducing hiring costs associated with the replacement of workers leaving the company. In the system dynamic methodology, a simulation model integrates predictive models (point solutions) derived from analytics (e.g., pay satisfaction, work pressure, etc.) and depict non-linear interactions among various factors (including feedback) that vary in place and time. This simulation model provides opportunity to identify high leverage points. In this way, combining insights from analytics into a simulation model helps management examine various “what-if” scenarios that can devise better policies, providing desired business outcomes.
In the next part of this blog series, we will discuss data analytics and discrete-event simulation methodology using a specific example.