For decades, business process optimization has been largely about art and craft (and sometimes wizardry). Apart from narrowly scoped approaches to optimize resource allocation (often assuming that workers behave like robots), a lot of business process optimization relies on high-level guidelines, with A/B testing for idea validation, which is hard to scale to complex processes. As a result, managers end up settling for a "good enough" process. Can we do more? In this talk, we review recent work on the use of high-fidelity simulation models discovered from execution data. The talk also explores the possibilities (and perils) that LLMs bring to the field of business process optimization.
This talk was delivered at the Workshop on Data-Driven Business Process Optimization at the BPM'2023 conference.
4. 4
The process model is authoritative
• No deviations, no workarounds
The simulation parameters accurately reflect reality
• …in reality, they are often guesstimates
A resource only works on one task instance at a time / a task is performed by one resource
• No multi-tasking / no multi-resource tasks (teamwork)
Resources have robotic behavior (eager resources consume work items in FIFO mode)
• No batching, no prioritization
• No tiredness or stress effects, no interruptions, no distractions
Undifferentiated resources
• Every resource in a pool has the same performance as others
No time-sharing outside the simulated process
• Resources fully dedicated to one process
5. End Result
Business process simulations based
on incomplete models,
guesstimates, and simplifying
assumptions are not faithful
optimization based on such
models is at best perilous
5
9. Given
• one or more event logs recording the execution
of one or more processes
• one or more performance indicators that we
seek to maximize/minimize
• a process model, decision rules, resource
allocation rules, other process knowledge
• a set of allowed changes to the process model
and associated rules
Find
• Possible sets of changes to the process to
optimize the performance measures
13. Conversational Process Optimization
• Search-Based Process Optimization is about exploitative process redesign
• Repeatedly applies a set of predefined adaptations
• Does not put into question the existing process structure
• Cannot handle unforeseen changes
• Conversational Process Optimization
• Makes search-based optimization a step in a human-in-the-loop optimization
approach
• Brings in general knowledge together with domain knowledge to transform human
directives into search space specifications
15. Summary
• ATAMO Process Optimization
• Expert-Driven Process Optimization with Simulation-in-the-Loop
• Expert-Driven Process Optimization with Data-Driven Simulation
• Search-Based Process Optimization
• Conversational Process Optimization
16. Tactical vs Operational Process Optimization
• The approaches reviewed focus on tactical optimization
• The goal is to go from an as-is to a to-be process
• Operational process optimization is also a fertile ground for research
• Prescriptive process optimization
• Triggering predefined interventions at runtime to optimize case outcomes
• Augmented process execution
• Triggering adaptations at runtime to respond to drifts in process behavior, including previously
unobserved or unforeseen changes
17. References
Data-Driven Simulation
• Camargo et al. Automated discovery of business process simulation models from event logs. Decision Support Systems
134:113284, 2020
• Chapela-Campa et al. Can I Trust My Simulation Model? Measuring the Quality of Business Process Simulation Models. BPM
2023, pp. 20-37
• De Leoni et al. Investigating the Influence of Data-Aware Process States on Activity Probabilities in Simulation Models: Does
Accuracy Improve? BPM 2023: 129-145
Search-Based Process Optimization
• Satyal et al. Business process improvement with the AB-BPM methodology. Inf. Syst. 84: 283-298 (2019)
• López-Pintado et al. Silhouetting the Cost-Time Front: Multi-objective Resource Optimization in Business Processes. BPM
(Forum) 2021: 92-108
• Peters et al. Resource Optimization in Business Processes. EDOC 2021, pp. 104-113
Conversational Process Optimization
• Barón-Espitia et al. Coral: Conversational What-If Process Analysis. ICPM Doctoral Consortium / Demo 2022
• Berti et al. Abstractions, Scenarios, and Prompt Definitions for Process Mining with LLMs: A Case Study. BPM Workshops
2023.
• Berti & Sadat Qafari: Leveraging Large Language Models (LLMs) for Process Mining (Technical Report). Arxiv 2307.12701
(2023)
18. References
Prescriptive Process Monitoring
• Fahrenkrog-Petersen et al. Fire now, fire later: alarm-based systems for prescriptive process
monitoring. Knowledge and Information Systems 64(2): 559-587 (2022)
• Kubrak et al. Prescriptive process monitoring: Quo vadis? PeerJ Comput. Sci. 8: e1097 (2022)
• Dasht Bozorgi et al. Prescriptive process monitoring based on causal effect estimation.
Information Systems 116: 102198 (2023)
• Padella & de Leoni: Resource Allocation in Recommender Systems for Global KPI Improvement.
BPM (Forum) 2023: 249-266
• Weytjens et al. Timed Process Interventions: Causal Inference vs. Reinforcement Learning. In BPM
Workshops 2023.
Augmented Process Execution
• Dumas et al. AI-augmented Business Process Management Systems: A Research Manifesto. ACM
Transactions on Management Information Systems 14(1): 11:1-11:19 (2023)
• Kurz et al. Reinforcement Learning-Supported AB Testing of Business Process Improvements: An
Industry Perspective. BPMDS/EMMSAD@CAiSE 2023
We have developed a tool called Simod capable of generate simulation models automatically based on an event log. The tool combines an automated process discovery technique to extract a process model, with trace alignment and replay techniques to extract the simulation parameters, and a hyper-parameter optimizer to evaluate and search for the best simulation model parameters configuration. Simod has been integrated into a beta state on the Apromore platform and has been submitted to the demo track of the same BPM 2019 conference.