Let’s look at these individually.
Process digital twins
A process digital twin is created using real-world process event data derived from process mining to create a graphical model of current business processes.
It contains a Business Process Modeling Notation (BPMN) model extended with multiple statistical parameters about the process (such as the process impetus, the activities involved, resources required, processing times) but also about how each of these activities are connected, or impact each other — to create branching probabilities between them.
Historical data from event logs is used to show how key process parameters have responded to change to help calculate and simulate scenarios of how they might respond to process change in the future.
So, with the digital process twin as the starting point, organizations can test these different process change scenarios — and analyze their impact on business KPIs.
Calculating the ‘what ifs’ in simulation analysis
To be valuable, simulations need a credible level of accuracy in assigning likely outcomes, which is where probability calculations come into play.
This is a rapidly evolving space in which the use of generative AI and machine learning are starting to play a more significant role. In such instances, systems use AI algorithms that learn from process mining data to generate new process outcomes by mimicking the complex relationships between process parameters.
Another approach commonly used in simulation analysis (at least for now) is known as the Monte Carlo simulation.
This simulation models the probability of different outcomes in processes by repeatedly generating different inputs in specific process variables and calculating the corresponding outputs. By doing so multiple times, the Monte Carlo simulation creates a distribution of possible outcomes, allowing businesses to estimate probabilities and make informed decisions.
For example, let’s consider a hypothetical supply chain process where clothing orders are fulfilled from a distribution center (DC). The time taken to complete each step of the order from receipt to send-out can vary due to factors like worker efficiency, item availability and DC system downtime.
The Monte Carlo simulation enables us to understand the overall process performance by considering and building probability distributions for each activity's duration based on historical data. Rinse and repeat thousands of times and we arrive at a range of possible outcomes deriving from changes in any of the key processes — as well as identifying potential bottlenecks and opportunities for improvement.