Data + Technology
             =
Great UX + Accurate risk
Michal Kročil
May 2019

To challenge the status quo of everyday finance
via exceptional customer experience
by being at the forefront
in delivering cutting edge technology.
Purpose
of Twisto

Understanding the environment
Fraudulent / Irresponsible / Distressed
Implications
01
Mitigation of threats
Data sources and processing
Modelling / Technology
Non-scoring tools
02

01
Understanding
the environment

Everything moves fast.

Borders are blurred.

Reasons for customer ending in default
Distribution highly dependent on: ticket size, duration, purchased product, standard of
identification
of defaulters have the intention to repay back but do not have the means or attention to do so
10%
Distressed
of defaulters do not take the responsibilities and consequence seriously
40%
Irresponsible
of defaulters know from the very beginning, they are not going to play by the rules
50%
Fraud

Fraudsters
-actively trying to find any weak spots
-different level of sophistication
-different level of „investment“
1
2
3
4
5
6
7
The critical thing is to identify them and prevent them from completing the transaction. Once it is
completed, very thin chances of recovering any money.
Goods attractivity
Delivery methods
Identity verification
Application scoring
KO criteria
Velocity
Anomalies
Mistakes must not be repeated!
Mitigation

Irresponsible
-trying to „get away with it“
-younger and lower education
-not necessarily evil people
-pressure / opportunity / rationalization
1
2
3
4
5
They might prove to be profitable customers at the end but they need to be well navigated.
Do not try to change them! Understand them and work with how they are.
Positioning / Marketing
Psychology of communication
Intensity scaling
Application scoring
2nd chances?
Psychology!
Mitigation

Distressed
-Out of money
-Complicated life situations
-Disoriented / ashamed
-
1
2
3
4
5
6
You should be friendly and make an effort to win them back. They might become great success
stories.
Fraudsters and Irresponsible might pretend to look like Distressed. Hard to distinguish.
Prevention / early recognition
Behavioral scoring
Notifications scaling
Ease of 1st step
Customer support
Recovery options
Prevention and care!
Mitigation

02
Mitigation of
threats

UX

vs.
Risk

Complexity needs to stay.
But customer needs to be shielded from it by technology.


Data
-As few directly from customer as possible
-As many from other sources as possible
-Regularly reconsider data sources
-Pre-processing
-Check consistency
-Junk in = junk out
-Paid sources / Margin
1
2
3
4
5
6
7
8
9
Perfect data collection, understanding and making them work for you is obviously one of the most
critical aspects of any highly automated online operations.
Customer-entered data
Transaction details
Delivery method
Technical data
Shopping history
„Framework data“
Behavioral data
Interaction data
3rd party data
Think about

Modelling
-Scoring and CLTV
-History -> Prediction
-Credit risk vs. Fraud risk
-Application /  Behavioral
1
2
3
4
5
6
7
8
9
Cool and important, but plays its role efficiently only as long as it is situated in an environment
that provides support in case of inaccurate assessment (= fall back plan).
Data source consistency
Data quality and volume
Target variable homogeneity
Technological environment
IT implications
Accuracy measurement
Robustness
Monitoring vs. Recalibration
Cut-offs (DR vs. Contribution margin)
Think about

Non scoring tools
-Complementary to scoring
-Not as accurate and sexy
-More robust
-Passportable internationally
-
1
2
3
4
5
6
7
No single tool that we use is the „silver bullet“ for risk mitigation. Well cultivated environment
in which particular tools complement each other. We approach the risk from different vectors and
that does the trick.
Product / process rules
KO rules
Verification scaling
Relationships
Velocity
Anomalies
Manual review

Think about







+ Data
+ Technology
+ People
= Success

Thank you!
Michal Kročil
michal.krocil@twisto.cz
+420 608 966 643