Business Process Simulation with Differentiated
Resources: Does it Make a Difference?
Orlenys López Pintado
Marlon Dumas
September 15, 2022 - BPM’22
PIX Project
Undifferentiated Or Differentiated Resources?
4 Clerks
Same Availability
Same Performance
Differentiated Availability
Differentiated Performance
Senior Clerk
Junior Clerk
Pooled Allocation
Unpooled Allocation
2
Inter-arrival
Times
Branching
Probabilities
Business Process Simulation
Simulation Model
Event Log
Metrics
Waiting Time
Processing Time
Cycle Time
Idle Time
Resource
Utilization
BIMP
3
Business Process Simulation
What would be the cycle time of a process
if the number of daily new cases increases by 20%?
What if a resource become unavailable for a long period?
What if a resource is replaced by another with lower performance?
What if a resource changes availability from full-time to part-time?
4
Research Question and Contributions
Do pooled allocation, undifferentiated performance and availability affect the accuracy
of a business process simulation model, and if so, to what extent?
1. A business process simulation approach with differentiated resources –
Unpooled allocation, Differentiated Performance, and Differentiated Availability
PROSIMOS
2. An automated method to discover a simulation model with differentiated resources from
event logs
Proposal
https://prosimos.cloud.ut.ee (Online)
https://shorturl.at/FNPS9 (Screencast)
5
Discovering Resource Availability
00:00 01:00 02:00 ….. 08:00 09:00 ….
23:00
Monday
00:00 01:00 02:00 ….. 08:00 09:00 ….
23:00
Sunday
.
.
.
.
Junior Clerk 06/06/2022 – 08:17:03  Monday – 08:00 to 09:00
What Granule Candidates to keep?
Confidence (task-conditioned)
Support
Resource Participation (task-relative)
How many Mondays from 8:00 to 9:00?
How many timestamps covered by any calendar granule?
How many resources to keep?
6
Discovering Resource Availability – Task Allocation
What if unallocated task?
Group By Max Disjoint Intervals
No Yes
7
Discovering Resource Performance
Task Instance: Check Credit History
Star
t
End
C1 C2 C3
Processing Time = End - Start
Idle
Processing Time = C1 + C2 + C3
What if not enough Points
(i.e., task instances processing
times)?
Best Fitted Distribution
Aggregated Distribution
8
Inter-Arrival and Branching Probabilities
Inter-Arrival
1. Filter and Sort first (start) event of each trace
2. Discover Inter-Arrival Calendar
3. Discover Best Fitted Distribution (difference between consecutive cases)
Branching Probabilities
1. Reply event log over process model
2. Compute probability on (Split) Inclusive and Exclusive gateways
9
Experimental Evaluation
10 Event Logs
10
Synthetic Real Process
Unpooled Allocation Differentiated
Performance
Differentiated
Availability
SP-NP-NA
MP-NP-NA
MP-DP-NA
MP-NP-DA
SP-DP-DA
5 baselines
Earth Movers’ Distance metric (EMD), which measures the minimum cost of turning one discretized
distribution (i.e., histograms) into another, penalizing each transformation by their distance.
EMD – Cycle Times:
EMD – Work Rhythm:
11
00:00 01:00 02:00 ….. 08:00 09:00 10:00 …. 22:00
23:00
15.09.2022 1 0 35 50 58 2 0
1 0 60 50 40 2 0
0
20
40
60
80
0
20
40
60
80
Cycle times grouped into 100 equidistant bins
0
100
0
100
Real Log Simulated-Log
Experimental Evaluation
Experimental Results
EQ1.What impact does unpooled resource allocation have compared to pooled allocation?
Unpooled models offer the best results under complete differentiation of resource performances
and availability. Still, they perform more poorly in the presence of undifferentiated resources.
EQ2 What impact does differentiated resource performance have compared to
undifferentiated performance?
EQ3 What impact does differentiated resource availability have compared to
undifferentiated availability?
Differentiated performance – Best (on average) Estimating Work Rhythm
Differentiated availability – Best (on average) Estimating Cycle Times
Differentiated performance and availability – Best Results Overall
12
Summary
2. Replace curve-fitting on Inter-Arrival Discovery
1. Discover calendar with non-weekly periodicity
3. Discover Optimal parametrization
13
Future Work
Proposal:
A business process simulation approach with differentiated resources profiles
Unpooled allocation,
Differentiated Performance
Differentiated Availability
PROSIMOS: Discovering and Simulating Business Processes With Differentiated
Resources
Business Process Simulation with Differentiated
Resources: Does it Make a Difference?
Orlenys López Pintado
Marlon Dumas
September 15, 2022 - BPM’22
PIX Project

Business Process Simulation with Differentiated Resources: Does it Make a Difference?

  • 1.
    Business Process Simulationwith Differentiated Resources: Does it Make a Difference? Orlenys López Pintado Marlon Dumas September 15, 2022 - BPM’22 PIX Project
  • 2.
    Undifferentiated Or DifferentiatedResources? 4 Clerks Same Availability Same Performance Differentiated Availability Differentiated Performance Senior Clerk Junior Clerk Pooled Allocation Unpooled Allocation 2 Inter-arrival Times Branching Probabilities
  • 3.
    Business Process Simulation SimulationModel Event Log Metrics Waiting Time Processing Time Cycle Time Idle Time Resource Utilization BIMP 3
  • 4.
    Business Process Simulation Whatwould be the cycle time of a process if the number of daily new cases increases by 20%? What if a resource become unavailable for a long period? What if a resource is replaced by another with lower performance? What if a resource changes availability from full-time to part-time? 4
  • 5.
    Research Question andContributions Do pooled allocation, undifferentiated performance and availability affect the accuracy of a business process simulation model, and if so, to what extent? 1. A business process simulation approach with differentiated resources – Unpooled allocation, Differentiated Performance, and Differentiated Availability PROSIMOS 2. An automated method to discover a simulation model with differentiated resources from event logs Proposal https://prosimos.cloud.ut.ee (Online) https://shorturl.at/FNPS9 (Screencast) 5
  • 6.
    Discovering Resource Availability 00:0001:00 02:00 ….. 08:00 09:00 …. 23:00 Monday 00:00 01:00 02:00 ….. 08:00 09:00 …. 23:00 Sunday . . . . Junior Clerk 06/06/2022 – 08:17:03  Monday – 08:00 to 09:00 What Granule Candidates to keep? Confidence (task-conditioned) Support Resource Participation (task-relative) How many Mondays from 8:00 to 9:00? How many timestamps covered by any calendar granule? How many resources to keep? 6
  • 7.
    Discovering Resource Availability– Task Allocation What if unallocated task? Group By Max Disjoint Intervals No Yes 7
  • 8.
    Discovering Resource Performance TaskInstance: Check Credit History Star t End C1 C2 C3 Processing Time = End - Start Idle Processing Time = C1 + C2 + C3 What if not enough Points (i.e., task instances processing times)? Best Fitted Distribution Aggregated Distribution 8
  • 9.
    Inter-Arrival and BranchingProbabilities Inter-Arrival 1. Filter and Sort first (start) event of each trace 2. Discover Inter-Arrival Calendar 3. Discover Best Fitted Distribution (difference between consecutive cases) Branching Probabilities 1. Reply event log over process model 2. Compute probability on (Split) Inclusive and Exclusive gateways 9
  • 10.
    Experimental Evaluation 10 EventLogs 10 Synthetic Real Process Unpooled Allocation Differentiated Performance Differentiated Availability SP-NP-NA MP-NP-NA MP-DP-NA MP-NP-DA SP-DP-DA 5 baselines
  • 11.
    Earth Movers’ Distancemetric (EMD), which measures the minimum cost of turning one discretized distribution (i.e., histograms) into another, penalizing each transformation by their distance. EMD – Cycle Times: EMD – Work Rhythm: 11 00:00 01:00 02:00 ….. 08:00 09:00 10:00 …. 22:00 23:00 15.09.2022 1 0 35 50 58 2 0 1 0 60 50 40 2 0 0 20 40 60 80 0 20 40 60 80 Cycle times grouped into 100 equidistant bins 0 100 0 100 Real Log Simulated-Log Experimental Evaluation
  • 12.
    Experimental Results EQ1.What impactdoes unpooled resource allocation have compared to pooled allocation? Unpooled models offer the best results under complete differentiation of resource performances and availability. Still, they perform more poorly in the presence of undifferentiated resources. EQ2 What impact does differentiated resource performance have compared to undifferentiated performance? EQ3 What impact does differentiated resource availability have compared to undifferentiated availability? Differentiated performance – Best (on average) Estimating Work Rhythm Differentiated availability – Best (on average) Estimating Cycle Times Differentiated performance and availability – Best Results Overall 12
  • 13.
    Summary 2. Replace curve-fittingon Inter-Arrival Discovery 1. Discover calendar with non-weekly periodicity 3. Discover Optimal parametrization 13 Future Work Proposal: A business process simulation approach with differentiated resources profiles Unpooled allocation, Differentiated Performance Differentiated Availability PROSIMOS: Discovering and Simulating Business Processes With Differentiated Resources
  • 14.
    Business Process Simulationwith Differentiated Resources: Does it Make a Difference? Orlenys López Pintado Marlon Dumas September 15, 2022 - BPM’22 PIX Project