Recently, machine learning algorithms surpass humans intelligence in many areas (go, chess, poker). Operational optimization (logistic, back-office) and customer behaviour predictions (marketing, sales) are some of the top priorities in companies to digitize their business.
But only a few can remember that it all started in Bletchley Park with the need to break the Enigma Code. Without business analysis techniques they probably wouldn’t have succeeded. BA approaches that were used back in the day are still valuable today.
We will present two real (banking sector) cases and their results to demonstrate analytical phases of designing, developing and using predictive analytics models that process customer data daily and recommend actions, based on predefined business rules and decision points in workflows. From Stakeholder Needs aligned with their Value we will show how to build smart predictive algorithms to determine the “next best action” and “preferred channel” in the Context of better CX.
Defined KPI’s measure VALUE daily and enable BA to monitor effectiveness and efficiency continuously, detect potential issues and take necessary corrective actions. In 5 months we increased the VALUE to 450% and needed 16 days to achieve ROI.
Key takeaways:
- Examples of different approaches we have taken to implement valuable predictive analytics solution, including what works and what doesn't
- How BAs can balance between external Customer eXperience view and internal stakeholders need to maximize the value of the project
- Large quantities of data exist, but the value is in analytics, not only the right algorithms that will work, but that they improve CX and add value.
- Approaches to ensure that algorithms and procedures used within project are "good enough"
BA and Beyond 19 Andrej Guštin - Mirror mirror on the wall Who's the wisest of them all
1.
2. Mirror, Mirror on the wall, who's the wisest of them all?
BA Perspective on Predictive Analytics and Artificial Intelligence
Andrej Guštin, IIBA Chapter Slovenia, Vice President; CREA pro, CEO
Agenda:
I. Short introduction
II. Case – Customer behavior
III. Key takeaways
5. Queen
Magic mirror on the wall, who is the fairest one of all?
Magic Mirror
Famed is thy beauty, Majesty. But hold, a lovely maid I see. Rags
cannot hide her gentle grace. Alas, she is more fair than thee.
Queen
Alas for her! Reveal her name.
Magic Mirror
Lips red as the rose. Hair black as ebony. Skin white as snow.
Queen
Snow White!
Photo from: http://disney.wikia.com/wiki/Snow_White
Mirror‘s predictive analytics algorithm
?
7. What „really“ helped - behind breaking the Enigma code
BA perspective
• Prototyping (10.36)
– The first prototype was too slow
• Solution performance goals (10.28)
– Clear KPI – 24 hours change of Enigma settings
• Data mining (10.14)
– Finding useful patterns and insights from data („weather“ „nothing to report “)
• Estimations (10.19)
– Gardening - to encourage a target to use known „plaintext in an encrypted message“
• Risk analysis and management (10.38)
– Confidentiality and non-contamination of the „sample“
7/22
9. Case background – the story
• Since economic crisis in 2008, Slovenian
banks have been deeply involved in the
collection process due to the increased
quantity and volume of overdue
outstanding receivables.
Growth of non-performing loans
Decline in the number of employees
• Operational efficiency
optimization led them to
decrease the number of
employees, so collectors were
overloaded with tasks and
documents.
10. Recovery process – From need to value
• Need: how to optimize collection process and increase the volume and amount of
collected payments.
• Stakeholder: back-office, customer service, call center, clerk, middle management
• Context: economic situation, as described
• Change: from human to machine decision making.
• Solution: predictive model (R) for probability calculations. Selectively targeting
the right debtors with the right collection strategies at the right time was
proposed by the Solution and integrated processes.
• Value: optimal allocation of resources to maximize the amount collected while
minimizing collection costs.
11. Soft recovery
Contractual
obligations…Sell products Contract
Execute daily tasks
Hard recovery StopRescheduling
Customer
status
Overdue
receivables
DW
Bill of exchange
Letter
Write-off
Call
1. DEFINE OPTIMAL STEPS
2. EXECUTE OPTIMAL STEPS
Call
Internal compensation
Letter (Reminder)
Write offs
3. DASHBOARD
Daily transaction
90 days
External law firm
Collection and recovery – typical steps in the process
Internal settlement
1 day
12. Development of predictive model
Model
Algorithms
Cursors
Rules
Historical data Machine learning Result
New data for processing The calculation of probability Result
Model
DevelopmentDailyusage
What is the probability, that this
Customer will be late with this
payment?
Probability!
13. ## Confusion Matrix and Statistics
##
## Reference
## Prediction default no-default
## default 9 1
## no-default 2 180
##
## Accuracy : 0.984
## 95% CI : (0.955, 0.997)
## No Information Rate : 0.943
## P-Value [Acc > NIR] : 0.0041
##
##
## 'Positive' Class : default
##
98,4%
Behavior prediction index
13/22
Results – statistics
What we predict?
➢Probability of default
➢Preferred channel
➢Next best „offer“ - step
➢Propensity to buy
14. How do we measure the results?
• We used survival curve to present the
results.
• We chose only one (1) KPI to measure
Solution performance (AUC)
• Observation time interval from 0 to 90 days
of overdue
• Understand what AUC90 actually means?
• Set the baseline value for AUC KPI90
• Focus on Retail segment
14/22
18. What works?
BA approaches to implement valuable predictive analytics solution
• Prototyping
2-4 months for experimenting - poor results
• Solution performance goal
Clear KPI – AUC90[Retail]
• Data mining
Useful patterns in data exists („Pay day“; „Strong Days“; „ Friends“)
• Risk analysis and management
CX: be professional, be honest, be compassionate
19. Key takeaways:
How to implement valuable predictive analytics solution?
How to evaluate what works and what doesn't?
How to balance between CX and internal project goals?
How to understand data?
How to ensure "good enough" algorithms and procedures used?
STEP BY STEP, EVOLUTIONARY
AGREE ON SINGLE KPI
KNOW YOUR CUSTOMER
FIND USEFUL PATTERNS AND INSIGHTS
FEEDBACK LOOP
20. „Computers are our mirrors:
whether we marvel or shudder
at the latest AI,
we’re merely looking at ourselves.“
Source: https://www.newscientist.com/article/mg23130803-200-how-alan-turing-found-machine-thinking-in-the-human-mind/
21.
22. Andrej Guštin is a cofounder and CEO at CREA pro, a
leading Slovenian consulting company focused
comprehensively on business process management and
innovation.
Vice president of IIBA CHAPTER SLOVENIA since 2009