Talk delivered at University of Tartu's Data Science Seminar, 17 February 2021. The talk explains the role of process mining as a self-service data analytics technology for business teams.
Process Mining in Action: Self-service data science for business teams
1. Process Mining in
Action
Self-service data science for
business teams
Marlon Dumas
Professor of Information Systems @ University of Tartu
Co-founder @ Apromore
2. Meet Tom
- Customer Excellence Manager @ XYZ
- Tom cares about customers, recurrent
sales revenue, efficient service delivery, …
- Every week, he has different questions:
- Why did our churn rate increased last month?
- Why did the number of customer complaints
keep rising?
- Why are our response times so slow even
though we have added more staff?
- Should I automate parts of my customer
service process? Should I buy this brand new
chatbot? Should I buy the brand new co-
browsing platform I saw last week?
3. Tom’s company has tons of data
in their information systems!
- Customers’ web site visits
- Customers’ orders
- Customers’ complaints
- The activity of salespeople
- The activity of customer service staff
- Your shipments, product returns, …
- etc.
4. How to use these data?
4
https://www.youtube.com/watch?v=z9b9ZeU5aac
• Nice option, very useful to address specific challenges
• …but will you keep calling on them every day you
need an insight?
Hire a data science consultancy that delivers
successfully on 85% of its projects
• Possible in larger companies, but how much time it
takes them to gain your domain & business
knowledge?
Hire a data scientist or data science team
• Is it possible?
Set up a self-service-system that allows you to
analyze the data yourself.
5. Digital Footprint
Every process leaves digital footprints (transactions)
Data Preparation
Data is prepared (e.g. multiple datasets are merged)
Event Collection
Transactional data is collected from enterprise systems
and other sources
Process Analysis
Process mining algorithms are used to extract process
models and other process analytics.
Step 01
Step 02
Step 03
Step 04
Process Mining:
Self-Service Data-Driven Process Analysis
12. Process map with duration overlay
Process performance
dashboards
Performance Mining
13. Process Mining
/
event log
discovered
process model
Automated Process
Discovery
Conformance
Checking
Variants Analysis
Difference
diagnostics
Performance
Mining
Business rules /
normative model
Enhanced
process model
event log’
16. Uptake by organization size
MarketsandMarkets, Process Analytics Market – Global Forecast to 2023, May 2018
17. Case 1:
Process mining @ Nordic financial company
• Context: Mid-sized European payment systems provider operating in multiple countries
• Goal: Analysis of customer onboarding and customer support processes (B2B sales)
• Questions: Why are we performing in terms of customer satisfaction and resolution
times better in some countries and for some customer segments and not for others?
• Data sources: SAP CRM and ServiceNow, centralized via a data warehouse solution
• Timeframe: 8 weeks of data extraction & analysis, continued use aftewards
18. Positive deviance
Practices prevalent in best-performing
countries. For example, we found that
performing some activities earlier in the
process lead to better customer feedback.
Negative deviance
Practices associated with poor customer
feedback. For example, certain rework
loops caused by incorrect data collection
(for a type of customer) lead to delays.
Outcomes (after ca. 6 months)
• Process changes leading to reductions in customer onboarding time
of several days in lower-performing countries
• Changes leading to reductions in rework loops, increase in NPS
• Analysts are able to perform regular review of the process in days,
instead weeks (more than 3 x speed-up)
Case 1:
Process mining @ Nordic financial company
19. Case 2:
Process mining @ Australian pension fund
Identifying Inefficiencies in the
Claims Process
Identifying Complexity of the Claims
Process
Augmenting the Speed to Analysis
Proactive Conformance/Compliance
Analysis
Variant Analysis
Showcase the Flexibility of the Tool
Model the new Claims Process
Validate the Cost to Serve
20. Case 2:
Process mining @ Australian pension fund
82% of all pension claims cases were following the
“happy path”, i.e. were compliant with the process model
(straight-through processing).
But the 82% only accounted for 2 out of the total of 474
case variants, suggesting there were various non-
compliant cases.
21. Case 2:
Process mining @ Australian pension fund
The remaining 18% of cases had various errors, rework
loops, or were withdrawn at different stages of the
pension claims process.
22. Case 2:
Process mining @ Australian pension fund
Significant expected
ROI
Estimated annual ROI
of over AUD
Cost savings of AUD
150K during
project phase
Increased Speed to
Process Improvement
2.6 times
faster compared
to traditional PI
approaches
Increased Process
Efficiency
Expected process
efficiency gains
between 5-30%
1 2 3
23. Case 3:
Process mining @ Cineca
Fast facts
• One of the major EU computational centers
• 600 IT staff
• 2000+ requests per month
Processes analyzed
• Change request process
• Help request process
• Fault handling process
• Variants per geographical region and university
served
• 10 months of data
24. Case 3:
Process mining @ Cineca
Adoption of positive
deviances
• Replicating behavior of
top performing offices
• Training of IT service operators
Changes in ticket management
system
• Identified reasons for some
types of bounce-backs between
teams
• Alerts for real-time monitoring of
certain state changes
25. HSPI, Process Mining: A Database of Applications, 2020
Process Mining is Everywhere!