3. Hiscox global presence
3
USA
Atlanta
Chicago
Dallas
Las Vegas
Los Angeles
New York City
San Francisco
White Plains
Guernsey
St Peter Port
Latin American
gateway
Miami
Bermuda
Hamilton
Europe
Amsterdam
Berlin
Bordeaux
Brussels
Cologne
Dublin
Hamburg
Lisbon
Lyon
Madrid
Munich
Paris
UK
Birmingham
Colchester
Glasgow
London
Maidenhead
Manchester
York
Asia
Bangkok
Singapore
4. Take a step change in the
way we use data and
analytics at Hiscox
5. Our solution was to run Data Labs
Highly focused projects to deliver value from data
5
Goal: Demonstrate value from using data
We didn’t want to do a large
data strategy project
We needed to show where
data can work
(and where it can’t)
We needed to do this quickly,
with controlled investment
Solution: Focus on key decisions the business makes
Small, focused, pragmatic projects delivering value from
data by improving the decision making process
Hiscox Data Labs
6. We had a clear message on how data would help
4 areas to tackle for us to make progress
6
Identify the decisions
that have the biggest
impact on our
business.....
... and the technology
infrastructure required
to support data and
analytics.
... that we have
expertise to identify
ways to improve these
and run analytics on a
regular basis...
... ensure we are
collecting and storing
good quality data that
underpin those
decisions...
Analytics Framework Technology Roadmap
Cultural Roadmap
Data (Int / Ext)
7. Focusing on decisions is at the heart of this
It takes time to identify the key decisions
7
Identify the decisions
that have the biggest
impact on our
business.....
... and the technology
infrastructure required
to support data and
analytics.
... that we have
expertise to identify
ways to improve these
and run analytics on a
regular basis...
... ensure we are
collecting and storing
good quality data that
underpin those
decisions...
Analytics Framework Technology Roadmap
Cultural Roadmap
Data (Int / Ext)
Focus of Data Labs
8. Hiscox Data Lab – 4 stages to deliver value
Decisions are the key focus throughout
8
Baseline
Map the current state and
identify all decisions
Opportunities
Prioritise decisions to decide
the key areas of focus
Ideas
Develop proof of concepts and
understand value
Deliver
Move from POC to Minimal
Viable Product – deliver value
1
2
3
4
9. Case study: US Direct business insurance
Using the Data Labs framework to drive value
9
• SME and micro business insurance
• US customers only
• Purchased via a website (Hiscox or
partners)
10. Align to business strategy and focus on decisions
Customer churn used as proxy for satisfaction
10
Deliver meaningful, tailored products that customers value
Goal:
How can we provide an improved
experience for our customers?
Who is most likely to benefit from our
products and how do we market to
them?
Key
Decisions:
Predictive customer churn model
− Who is likely to cancel?
− Why are they going to cancel?
− When will they cancel?
Idea:
11. 3 components to the churn model to understand the
full picture about customer cancellations
11
Who When Why
How likely is a
customer to complete
the policy year?
When do cancellations
take place over a policy
year?
For what reason will the
customer cancel?
• Predictive machine
learning model
• Over 20 predictive factors
used
• Simple analysis of
cancellation dates
• Machine learning
approach led to worse
results
• Predictive machine
learning model
12. “Who” model – tiered approach
Helps to understand the model and take action
12
1. Input policy
profiles
2 . Score policies in
“Who” model
Tier 5
(20% of policies)
Tier 4
(20% of policies)
Tier 3
(20% of policies)
Tier 2
(20% of policies)
Tier 1
(20% of policies)
Highest
probability of
cancelling
Lowest
probability of
cancelling
• Partner channel,
Industry, products
purchased,
switcher, etc.
• A probability of
cancelling in this
policy year
• 0 – 100%
3. Using predefined
thresholds,
segment into tiers
Within each tier we can track:
• New Binds
• # of Cancellations
• Retention rate...
All vs. an expected baseline
based on model / history
13. High probability of cancellation
Tiers used to focus actions to improve decisions
13
Tier 5
(20% of policies)
Tier 4
(20% of policies)
Tier 3
(20% of policies)
Tier 2
(20% of policies)
Tier 1
(20% of policies)
How can we provide an
improved experience for
our customers?
Decision
Targeted actions to
improve customer
satisfaction e.g.
coverage consultation
Action
14. Low probability of cancellation
Tiers used to focus acquisition spend
14
Tier 5
(20% of policies)
Tier 4
(20% of policies)
Tier 3
(20% of policies)
Tier 2
(20% of policies)
Tier 1
(20% of policies)
Who is most likely to
benefit from our products
and how do we market
to them?
Decision
Target acquisition spend
on businesses similar to
those in tiers 1 and 2
Action
15. “When” model key in testing framework
Reduces feedback time on impact of actions
15
Without model
Long time to feedback
Action
Review results vs. high level
metric (can take up to a
year)
Hard to iterate due to length
of process
With model
Short time to feedback
Action
Monitor vs.
expected
on a
monthly
basis
Evaluate
impact and
iterate
Repeatable
Cycle
16. Practicalities of making this work
Automate as much as possible
16
Customer
Data
Policy data is stored in our
underwriting systems
17. Practicalities of making this work
Automate as much as possible
17
Customer
Data
SQL
Database
Data stored into a SQL
database overnight
18. Practicalities of making this work
Automate as much as possible
18
Customer
Data
SQL
Database
Predictive
models
Predictive models run in R
from policy data in SQL
database
19. Practicalities of making this work
Automate as much as possible
19
Customer
Data
SQL
Database
Predictive
models
Model results are joined
onto policy information and
stored back in SQL
20. Practicalities of making this work
Automate as much as possible
20
Actions
Customer
Data
SQL
Database
Predictive
models
Customer tier and other
characteristics used to take
targeted actions
21. Practicalities of making this work
Automate as much as possible
21
Actions
Testing
Framework
Customer
Data
SQL
Database
Predictive
models
Actions taken stored in
testing framework and will
affect customer behaviour
22. Practicalities of making this work
Automate as much as possible
22
Actions
Testing
Framework
Customer
Data
SQL
Database
Predictive
models
Customer cancellations
recorded in underwriting
system and stored in SQL
23. Practicalities of making this work
Automate as much as possible
23
Actions
Testing
Framework
Customer
Data
SQL
Database
Predictive
models
Cancellations are
compared against
expectations to analyse
impact of actions
24. Key takeaways:
Focus on decisions
You can be more confident in the business changing
Get to an MVP quickly
Don’t aim for perfection in version 1 – show value first
Notas del editor
Bronek’s guide on how to move the needle in the right direction:
Do simple analysis well on key business decisions
Take action on the back of this analysis
Test the results, learn then repeat