This document discusses business process analysis, simulation, and optimization. It provides an overview of structural and statistical analysis, capacity analysis using dynamic simulation methods, visualization and numeric simulation techniques in business process simulation. Optimization techniques are discussed for selecting optimal scenarios. The benefits of simulation over real-world testing are speed and low cost. Process simulation best practices and caveats are also covered, such as ensuring the right model, parameters, and expertise for the intended goals.
2. Business Process Analysis, Simulation and
Optimization
Denis Gagné,
www.BusinessProcessIncubator.com
Chair BPSWG at WfMC
XPDL Co-Editor at WfMC
BPMN 2.0 FTF Member at OMG
BPMN 2.1 RTF Member at OMG
CMMN Submission at OMG
3. Analysis
Structural Analysis
The structural aspects (configuration) of a process model
Usually Statistical Analysis (using static methods)
Capacity Analysis
The capacity aspects of a process model
Usually Dynamic Analysis (using discreet simulation
methods)
4. Simulation
Visual Depiction (Visualization & Animation)
User is presented with a (sometime interactive) animated
depiction of the business process model
Numeric Simulation (Discreet Events)
User asked to provide values for input and decision
parameters of a simulated business model
Role Play (Serious Gaming)
User asked to take actions and make decisions within a
simulated business environment
5. Optimization
selection of a best scenario (with regard to some criteria)
from some set of available alternatives
Almost impossible without tool support
Sub optimization caveat
Optimizing the outcome for a subsystem will in general not
optimize the outcome for the system as a whole.
6. Why Simulation
Training & Learning
Although very popular in support of operations, limited use in other Business
Modeling disciplines
Persuasion & Selling
Simulating results of a proposed solution
Cause and Effect simulation
Analysis of a Business Situation
Simulation complements static analysis techniques with a dynamic perspective
Verification & Validation
Validation: Are we doing the right “thing”?
Verification: Are we doing “it” right?
7. Benefits of Simulation
Advantages of simulation over testing on the
real world include:
speed
low cost
no disturbance of current operations
8. Business Process Simulation
Working Group (BPSWG)
Goal
Define a Specification for the Parameterization and
Interchange of Process Analysis Data allowing Structural
and Capacity Analysis of a process model providing for
Pre-execution and Post-execution optimization.
We are interested in the Data («in/out») (commonly the «what»)
and its Interchange not its interpretation or its use (the «how») or
tool smart.
9. BPSim Conceptual Model
BPSim
BPMN or
XPDL
Scenario
(Input)Process
Model
Reference
Reference
instance instance
Scenario
(Output)
Reference
instance
Reference
10. Scenario
Used to provide complementary information to a BPMN or
XPDL business process model in the context of process
analysis, simulation and optimization (ASO).
A collection of element Parameters augmenting the
Business Process Model
Scenario can capture:
input parameter specification for ASO; or
results from ASO; or
historical data from past real world execution of the business
process model.
12. Business Process Simulation
Best Practices
The Right Model for the Right Goal
Align Modeling Objectives with Simulation Objectives
Abstraction
Fidelity
Validity (soundness and completeness)
The Right Answer to the Right Question
Make sure to instrument your business model with parameters that are
actual indicators (influencers) of what you wish to explore
The Right Expert for the Right Task
Although conceptually simple to grasp, successfully (meaningfully) using
numerical simulation for business modeling still requires some expertise
(Advanced Mathematical Skills)
13. Business Process Simulation
Caveats
Unrealistic User Expectations
Simple Press-Button Simulation
Deterministic behavior assumptions
A Business Model is a Simulation Model (not necessarily)
Their goals (purposes) may be misaligned
Be Mindful of Misleading Results (Garbage in Garbage Out)
A simulation model that is fidel &valid with uncharacteristic data can lead to Incorrect
conclusions or predictions, Negative Training, …
Sub-Optimization
Partial or sub-model optimization can lead you astray
16. Integrated set of visual toolkits for modeling, simulating,
analyzing and optimizing business processes.
17. Visual representations within and across the toolkits are
synchronized and provide the user with multiple
perspectives exposing the behavior of a complex process
and the root causes of that behavior.
18. Initial resource allocation uses a two phase approach. First the process is simulated with
unlimited resources in order to get the demand for each performer (the workload drivers). Then
a resource is defined for each performer with a capacity determined by the workloads, and the
process is simulated a second time. The results of the second run are then loaded into the other
toolkits, ANALYZE and OPTIMIZE.
19. To validate a model we use Gantt Charts displaying information about the path of a work item
arrival through the process. They are synchronized with the Process Preview, a compressed view
of all the pages in the model which in turn is synchronized with the MODEL toolkit. The Gantt
charts are synchronized with MODEL and with other chart groups and tables within ANALYZE.
20. For a simple process like our example there is not much to validate; for more complex, ‘real-life’
models, however, the combination of Gantt chart, process preview and time slicer turns out to
be quite powerful. And if the travel of a work item through the process comes to an untimely
end, due to inconsistent XOR branching, for example, the ANALYZE toolkit will mark this by
moving the time slicer and scrolling the preview.
21. When we focus on the workflow process as depicted in the MODEL toolkit and want to know
whether and where a particular node appears in a Gantt chart, we do a right-click on that
flownode and select Synchronize. The Gantt time slicer is moved to point where the work item
arrives the first time at that node, if at all. Figure 5 depicts the case for the subprocess process
premium and work item arrival 11.
22. We see that cycle times are mostly distributed between 30 and 135 minutes, and then at the
right edge there a few outliers. We select them for a closer look. When we browse through the
selected Gantt charts by means of the spinner we see that all contain the process premium
branch, some end approved and others rejected, and wait times in front of the tasks receive
application and KYpremiumC differ but their sums remain the same.
23. Path patterns classify the work item arrivals by counting how often the work item passes
through the different flow nodes. The third ANALYZE pane shows the path patterns. Pattern 3,
the one with the smallest size, characterizes those work items that pass through process
premium and get rejected. If we click on the Arrival List cell of a pattern, the subset is selected
in the History chart and we can scan all the histories with the same path pattern.
24. The resource idleness analysis for the whole period of time and – in a copy – the same data for
a single day. It seems that a maximum cycle time of about 2.5 hours may be unnecessarily low
for the given application.
25. From the OPTIMIZE toolkit we can start our Smart Allocation tool for any of the scenarios
available in ANALYZE or for a fresh simulation. Once started, the Smart Allocation tool allows in
fast iterations the changing of the number of copies (‘clones’) of each resource, simulation of
the given set of work item arrivals and comparing of the results in terms of resource and cycle
times.
26. We have a simple idleness reduction algorithm that removes clones from a resource as long as its
remaining average idleness does not fall below 20%. Here it comes up with a much leaner allocation,
with two immediate consequences.
The average per-arrival cost goes down from $14.45 to $9.40.
The cycle time goes up to about 5 hours, with two groups of outliers, one around 19 hours and one
around 64 hours.