Discovering Digital Process Twins for What-if Analysis: a Process Mining Approach

This webinar discusses the limitations of traditional approaches for business process simulation based on had-crafted model with restrictive assumptions. It shows how process mining techniques can be assembled together to discover high-fidelity digital twins of end-to-end processes from event data.

1. Discovering Digital Process Twins for What-if Analysis: a Process Mining Approach Marlon Dumas Professor @ University of Tartu Chief Product Officer @ Apromore AUTO-TWIN Webinar – June 19 , 2024

2. How to determine if a given process change would improve a business process, and by how much?

3. The Traditional Answer: Business Process Simulation 3

4. Starting Point: Business Process Model 4

5. 1. Specify Processing Times 5 Exp (20m) Normal(20m, 4m) Normal(10m, 2m) Normal(10m, 2m) Normal(10m, 2m) 0m

6. 2. Specify arrival process & branching probabilities 6 Arrival rate = 2 applications per hour Inter-arrival time = 0.5 hour Negative exponential distribution From Monday-Friday, 9am-5pm 0.3 0.7 0.3

7. 3. Specify resource pools & task-to-pool assignment Clerk Officer System Clerk Officer Officer Clerk Credit Officer € 25 per hour € 35 per hour Mon-Fri, 9am-5pm Mon-Fri, 9am-4pm

8. 4. Analyze the results D

9. Business Process Simulation: Assumptions 9

10. End Result Business process simulations based on incomplete models, guesstimates, and simplifying assumptions are not faithful  adoption of business process simulation is disappointing 10

11. Data to the Rescue! Enterprise System (CRM, ERP, …) Event Log

12. Data-Driven Business Process Simulation Model Learning Process Constraints or Process Model Enterprise System Process Change Specification Simulation Engine Predicted Performance Profile & Reliability Estimate Digital Process Twin

13. Problem Statement

14. Non-Functional Requirements 14

15. Simod v1: Discovering Digital Process Twins for What-If Analysis Stochastic Process Model Discovery Event log Simod Control-Flow (BPMN Model) Filtering threshold Parallelism threshold Branching probabilities Trace alignment + replay to determine how many times each sequence flow is traversed Resources & Performance Resource pools Pool-task assignment Resource timetables Multi-tasking behavior Interarrival distribution Activities durations distribution Simulation Simulator Optimizer Congestion Model Discovery SplitMiner Digital Process Twin ( DPT )

16. Evaluating the quality of DPTs event log (testing set) K simulated event logs 1. Generate K simulated event logs. 2. Compare individually and report the average and confidence interval. D. Chapela- Campa , M. Dumas, A. Senderovich , et al. Can I Trust My Simulation Model? Measuring the Quality of Business Process Simulation Models . BPM 2023.

17. Simod v1 – Results Temporal performance: mean percentage error of 15-40% Huge under-estimation of activity waiting times Erratic performance with large number of resources! Erratic performance when multi-tasking present Poor ability to capture the distribution of activity sequences. Two directions: Black-box – bring in deep learning! White-box – Address limitations one by one, Kaizen style M. Camargo et al. Automated discovery of business process simulation models from event logs. Decis . Support Syst. 134: 113284 (2020) .

18. Small Improvement – Differentiated Resources 4 Clerks Same Availability Same Performance vs Differentiated Availability vs Differentiated Performance Senior Clerk Junior Clerk Pooled Allocation vs Unpooled Allocation 5-10% improvement O. Lopez-Pintado & M. Dumas Discovery, simulation, and optimization of business processes with differentiated resources. Inf. Syst. 120: 102289 (2024)

19. Small Improvement Branching Probabilities * Frequency Analysis Branching Conditions * Deterministic Expressions 0.3 0.7 0.5 Activity Decision Make Credit offer denied Notify Rejection granted Activity Decision Make Credit offer granted Notify Rejection denied 5-10% improvement

20. Less Small Improvement – Modeling External Delays Extraneous activity delays : waiting times not explained by available data. 1. Analyze the waiting time previous to each activity (since its enablement). 2. Were the resources available? 10-20% improvement D. Chapela -Campa & M. Dumas. Enhancing business process simulation models with extraneous activity delays. Inf. Syst. 122 : 102346 (2024)

21. Nice Improvement– Probabilistic Availability 21 08:00 12:00 13:00 17:00 Monday 08:00 12:00 13:00 17:00 Friday Crisp Calendar 30% 50% 90% 100% 100% 100% 100% 100% 08:00 12:00 13:00 17:00 Monday 90% 90% 90% 70% 50% 50% 5% 5% 08:00 12:00 13:00 17:00 Friday Probabilistic Calendar 20-40% improvement

22. Diminishing Effects Plateau Multitasking ?? Batching?? Prioritization?? 5-20% improvement

23. Simod v5: From Event Data to Digital Process Twin https://tinyurl.com/autoDPT