Business process improvement based on event logs Pnina Soffer.
-
Upload
karin-kelly -
Category
Documents
-
view
218 -
download
3
Transcript of Business process improvement based on event logs Pnina Soffer.
Business process improvement based on event logs
Pnina Soffer
My research: business processes beyond control flow
• Abstracting from BP modeling languages using a state-based goal-oriented process view.
• Addressing various process-related problems in a coherent manner– Goal reachability assessment– Effect of data deficiencies on business processes– “How to think” when creating a process model– Flexibility and impact of process changes– Learning from experience– Compliance checking and diagnosis
Process mining
• Mines event logs of an information system, to:– Understand and represent the actual process– Check for compliance with required standards and
procedures– Identify performance bottlenecks– Identify improvement opportunities
Example 1: Identifying process workarounds and related problems
• Workarounds are intentional deviations from procedures
• Identifying and analyzing workarounds can guide process improvement
• Exploratory study in 5 organizations revealed 6 workaround types and problems related to them
• Mining event logs of the relevant IS enabled quantifying 4 of these types
Workaround diagnosis: example resultsOrganization Process
Number of instances
% instances with workarounds by type
Bypassing Post-hoc updates
Role violation
Post-hoc reporting
1 PR approval 3688 5.1% 1.3% 2.7% 5.9%
2PR approval 6920 53.2% 8.8% 22.3% 12.0%Purchase ordering 4211 6.8% 7.2% 24.4% 12.6%
3PR approval 21289 75.3% 25.0% 3.5% 68.1%Purchase ordering 5217 11.9% 4.8% 9.0% 4.1%
Average in all processes 30.5% 9.4% 12.4% 20.5%
Example 2: Business process learning
• Learning: the ability to induce from past experience in order to improve present and future performance.– Avoid repeating past mistakes– Know how to replicate past successful executions
• Resource used: log of past process executions• Main idea: to learn based on– What has been done (process path)– In what situations (context)– What has been achieved (outcome, performance)
• What should be considered as “context”?• How to identify and recommend the best
process path for a given context?
Two main questions:
• Aims to maximize the chances for a successful sale• Has a lot of information about each customer:– Approximate age, clothing style, height, figure, eye color,
hair color, state of mind…
• Needs to use this information for an effective classification to determine how to approach each customer
Context identification:The sales person’s problem
• An effective classification of process instances supports the selection of an appropriate path to achieve better outcomes
• Facing a large amount of contextual information, how to identify the relevant factors to establish an effective classification?
• Solution approach: feature selection and decision tree growing, identifying context variables and conditions that help predicting action-result combinations
The general problem
Identifying best process paths
• Classifying process instances by their context. • Mining each context-based group separately, to
identify decisions that have the highest effect on the outcomes achieved.
• Automated generation of decision rules accordingly, recommending options to select at given situations.
Current status
Performance % instances % improvementBefore After
Excellent 20.4 23.3 14.4Acceptable 75.8 76.4Unacceptable 4.6 1.0 76Total delivered instances 92.8 92.7Undelivered instances 6.4 6.5Rejected instances 0.8 0.8
• Current solution evaluated by simulation with promising results.
• Still ongoing research
Ongoing research directions
• Learning in medical processes: – High dimensional context– Compliance with clinical constraints– Processes without a clear completion point
• Addressing exceptions and low-volume variations• Providing real time decision support
• Main difficulty: a need for high quality logs to experiment with.