lima primary insurance pricing · henry receives a proposal of 330 € he will renew he will not...
TRANSCRIPT
Massimo Cavadini – 14 October 2020
PRIMARY INSURANCE PRICING
where prediction meets the business
“The unprecedented availability of structured
and unstructured data, cloud computing and
machine learning algorithms are bringing the
pricing sophistication to the next level.”
Massimo Cavadini
Head of Actuarial Consulting and Data Analytics
Munich Re
Why pricing is so important now?
Technical Sophistication
Technical Sophistication is the Differentiating Factor
LIMA Programme 2020
De-intermediation via
Online & Digital Channels
Breakthroughs in Modelling
Techniques (ML & AI)
Disruptive Players’ Entering
the Market (e.g., Lemonade)
Insurers’ Focus on Technical Income
due to Interest Rates Decrease
Increased Agility due
to IT Flexibility and new platforms
Increased Customer Awareness
and Price Sensitivity
3
H1 2020 Pricing Survey
4iStock-663981890
We use GLMs for the majority
(50–80%) of covers
We combine traditional and machine
learning (ML) approaches for nearly
all risk and demand (renewals/new
business) models
We use Generalized Linear Models
(GLMs) for less than half of the covers
63%19%
19%
1Describe the statusof your technical sophistication
H1 – 2020 pricing survey
LIMA Programme 2020 5
We use GLMs for the majority
(50–80%) of covers
We ran some pilots but it’s hard to
bring them to production level
We combine traditional and machine
learning (ML) approaches for nearly
all risk and demand (renewals/new
business) models
Use ML in other projects (discounting,
customer lifetime value, etc.) other
than the others mentioned
We use Generalized Linear Models
(GLMs) for less than half of the covers
Use some ML in demand models
Use some ML in risk models
63%19%
19%
16%
11%
27%
46%
1Describe the statusof your technical sophistication
2Describe your use of machine learning (ML)in pricing today
H1 – 2020 pricing survey
LIMA Programme 2020 6
We use GLMs for the majority
(50–80%) of covers
We ran some pilots but it’s hard to
bring them to production level
We manually import rating tables
through our IT system, and we would
like to move to API
We combine traditional and machine
learning (ML) approaches for nearly
all risk and demand (renewals/new
business) models
Use ML in other projects (discounting,
customer lifetime value, etc.) other
than the others mentioned
We have an API connection between
our pricing tool and the rating
system/policy administration system
We use Generalized Linear Models
(GLMs) for less than half of the covers
Use some ML in demand models We manually import rating tables
through our IT system, and we do
not see reason to change it
Use some ML in risk models
63%19%
19%
16%
11%
27%
46%49%
38%
14%
1Describe the statusof your technical sophistication
2Describe your use of machine learning (ML)in pricing today
3Describe your level of automation of pricing ina rating engine system
H1 – 2020 pricing survey
LIMA Programme 2020 7
We use GLMs for the majority
(50–80%) of covers
We ran some pilots but it’s hard to
bring them to production level
We manually import rating tables
through our IT system, and we would
like to move to API
It’s better to use professional
pricing software from a known
brand in combination with coding
abilities where needed
We combine traditional and machine
learning (ML) approaches for nearly
all risk and demand (renewals/new
business) models
Use ML in other projects (discounting,
customer lifetime value, etc.) other
than the others mentioned
We have an API connection between
our pricing tool and the rating
system/policy administration system
It’s better to hard-code all your
models in SAS, R, Python, etc.
We use Generalized Linear Models
(GLMs) for less than half of the covers
Use some ML in demand models We manually import rating tables
through our IT system, and we do
not see reason to change it
We are considering switching to a
different pricing software provider
Use some ML in risk models
63%19%
19%
16%
11%
27%
46%49%
38%
14%
67%18%
15%
1Describe the statusof your technical sophistication
2Describe your use of machine learning (ML)in pricing today
3Describe your level of automation of pricing ina rating engine system
4Describe your use of proprietary software
H1 – 2020 pricing survey
LIMA Programme 2020 8
We use GLMs for the majority
(50–80%) of covers
We ran some pilots but it’s hard to
bring them to production level
We manually import rating tables
through our IT system, and we would
like to move to API
It’s better to use professional
pricing software from a known
brand in combination with coding
abilities where needed
We feel our technical sophistication
is in line with our peers
We combine traditional and machine
learning (ML) approaches for nearly
all risk and demand (renewals/new
business) models
Use ML in other projects (discounting,
customer lifetime value, etc.) other
than the others mentioned
We have an API connection between
our pricing tool and the rating
system/policy administration system
It’s better to hard-code all your
models in SAS, R, Python, etc.
We should ramp up our sophistication
level because we’re lagging behind
We use Generalized Linear Models
(GLMs) for less than half of the covers
Use some ML in demand models We manually import rating tables
through our IT system, and we do
not see reason to change it
We are considering switching to a
different pricing software provider
We are way ahead, and we can see
the benefit of our early move in
this space
Use some ML in risk models
63%19%
19%
16%
11%
27%
46%49%
38%
14%
67%18%
15%22%
27%
51%
1Describe the statusof your technical sophistication
2Describe your use of machine learning (ML)in pricing today
3Describe your level of automation of pricing ina rating engine system
4Describe your use of proprietary software
5What do you observe in your market/within your competitors
H1 – 2020 pricing survey
LIMA Programme 2020 9
Prediction Decision making
Pillars
Pricing excellence framework
Data management Domain knowledge Deployment
Architrave Technical price
Using the past to predict the expected claim costs
Behavioral modelling
Customer buying behaviour and portfolio elasticity
Competitive market analysis
Taking into account market positioning to improve pricing adequacy
Rate calibration
Transforming the technical prediction in a commercial rate
Scenario testing
Assessing the impact of any rate change and the effect on the
expected volume and profitability
Portfolio steering & pruning
Identifying the loss/profit making segment and the strategy to
improve the profitability
▪ Data availability
▪ Storage infrastructure
▪ Business impact of analytical actions
▪ Regulatory environment
▪ Fast implementation and execution
▪ Synergy across software solutions
LIMA Programme 2020 10
iStock-840166882
DATA MANAGEMENT 1
Leveraging structured internal information
LIMA Programme 2020
Customer
▪ Age
▪ Gender
▪ License’s age
▪ Marital Status
▪ Occupation
▪ Previous Company
▪ Tenure
▪ Private / Commercial
▪ Bonus Malus
▪ …
Vehicle
▪ Make
▪ Model
▪ Vehicle’s age
▪ Horse Power
▪ KW
▪ Cubic Capacity
▪ Fuel type
▪ Sum insured
▪ …
Claim’s History
▪ Number of Claims
▪ Paid Amount
▪ Cover Affected
▪ Reserve
▪ ALAE
▪ …
Territorial / Economical
▪ Zip-Code / Postal Code
▪ Region
▪ Payment Method
▪ Instalment
▪ …
Policy data Claims data
12
How to use external information to improve the pricing process
Data Enrichment
Vehicle Territorial / Economical Risk Contextualization Benchmark▪ Road riskiness
▪ Accidents
▪ Traffic
▪ Road works
▪ Market prices
▪ Benchmark variables▪ Ranking
▪ Distance (Eur/
Percentage)
▪ …
▪ Financial - demographic Score
▪ Revenue information
▪ Geo - demographical info▪ Population Density
▪ Life Expectancy
▪ Crime Rates
▪ Cars per Inhabitant
▪ Injured per Accident
▪ …
▪ Technical characteristics ▪ Make
▪ Model
▪ Horse Powers
▪ KW
▪ Fuel type
▪ Value
▪ Max speed
▪ ADAS
▪ …
Application
Competitive Market Analysis
Demand modelling
Rate making
Risk Network Analytics
Risk Modelling
Telematics
Fraud Identification
Data cleaning
Risk Modelling
Demand Modelling
Risk Modelling
Demand Modelling
Microzoning
Risk Network Analytics
LIMA Programme 2020 13
Data is key for successful modelling
Internal
DataExternal
data
Policy data
Claims data
External Data – Structurede.g., Population Density
External Data – Unstructurede.g., claim description
Modelling Dataset
Prediction
LIMA Programme 2020 14
iStock-918855102
Prediction 2
The race of algorithms
Neural
Networks
1950
1972
1990
1995
1996
1999
GLMs
GAMs
Random
Forests
Penalized
regression
GBM
Time
LIMA Programme 2020 16
Technical pricing is your “compass”
Profit
Loss?
or Commercial ViewRisk View
100
Technical
Price
Risk Loadings Expenses
90
120
Commercial PremiumActual Premium
LIMA Programme 2020 17
LIMA Programme 2020
Choosing the right granularityS
oph
istication
Freq Sev Propensity Excess Freq Sev Freq Sev SevFreq
Product
Risk
Premium
Bodily Injury Material Damage Collision Fire & Theft
TPL Own Damage
Comprehensive
ExcessAttritional
Cover
Peril
18
Behavioural modelling: beyond the risk estimationPaul & Henry are twins …
HenryPaul
≠
=
=INSURANCE
ABOUT
PROFESSION
AGE 38
REGION Munich
VEHICLE BMW 120i F20
Successful Popstar Unemployed
AGE 38
REGION Munich
VEHICLE BMW 120i F20
300 €CURRENT PREMIUM
330 €RENEWAL OFFER
300 €CURRENT PREMIUM
330 €RENEWAL OFFER
Image: gmast3r / Getty Images LIMA Programme 2020 19
Renewal Scenarios
Scenario A
Scenario B
Insurance
+10%
+10%
Paulreceives a proposal of
330 €
Henryreceives a proposal of
330 €
He will renew
He will not renew
+15%
+5%
Paulreceives a proposal of
345 €
Henryreceives a proposal of
315 €
He will renew
He will renew
330 €EARNED PREMIUM
70%LOSS RATIO
99 €MARGIN
660 €EARNED PREMIUM
70%LOSS RATIO
198 €MARGIN
Paul and
Henry both
receive the
SAME offer
Paul and
Henry both
receive an
ADEQUATE
offer
Image: gmast3r / Getty Images LIMA Programme 2020 20
LIMA Programme 2020
Measuring the client and portfolio elasticity
… What is the probability
of converting/renewing?
How does the probability
of converting/renewing
change?
Demand Elasticity
If we offer a premium of
$200 …
Elasticity allows to measure and simulate the impact of a price change on the volume sold
e =∆% 𝑑𝑒𝑚𝑎𝑛𝑑
∆% 𝑝𝑟𝑖𝑐𝑒
21
Outcomes
LIMA Programme 2020
Competitive market analysis
▪ Competitors’ pricing
variables identified
▪ Marginal impact of each
pricing variables
▪ Implementable formula(s)
replicating competitors’
commercial premium
Profile ID Vehicle Brand Age Driver Region Price Competitor 1 Price Competitor 2
P0 BMW 35 Region 1 200 220
P1 BMW 20 Region 1 500 230
P2 BMW 35 Region 2 210 200
P3 … … … … …
Vehicle Brand Age Driver Region
Modality Coefficient Modality Coefficient Modality Coefficient
Audi 1 18–25 2.5 Region 1 1
BMW 1 26-50 1 Region 2 1.3
Nissan 0.8 >50 0.9 Region 3 1.1
Commercial Price Competitor 1 =
Base Premium × Coeff Vehicle Brand ×
Coeff Age ×Fixed Cost Region
Advanced Predictive Modelling Techniques
22
Combining all the predictions…
LIMA Programme 2020
Risk Models
Behavioral Model +
CMA
Possible applications
▪ Price optimization at single
customer level
▪ Accurate definition of the renewal strategy
through an advanced system of caps and floors
▪ Identification of elastic customer to be retained
with specific marketing campaign
Impacts
▪ Reduction of the Loss Ratio
▪ Top-line growth (volume)
▪ Expense reduction and
CoR improvement
23
….to prepare the ground for the right decisions
LIMA Programme 2020
Reverse
engineering
Competitive
market analysis
so
ph
istica
tion
Rate
requirement
Risk models &
Technical price
Price elasticity
Segment
definition
Demand
modellingPrice
optimization
Impact Analysis
Scenario testing
Multi – year
projection
24
iStock-479446229
Decision Making 3
Understand the current status to set the right strategy
Exp
ecte
d L
oss R
atio
Expected retention
Bottom line driven strategy
Scenario 2
Top line driven strategy
Scenario 1
Current status
LIMA Programme 2020 26
0
50
100
150
200
15 30 45 60 75 100
Expe
cte
dpro
fit(%
)
Percentage of clients (%)
LIMA Programme 2020
Pruning the portfolio with a forward-looking approach
The following 45% of clients contributes
with a further 70% of profits…
…which is destroyed by the remaining
40% of clients
The best 15% of clients generates
100% of profitability
15 60 100
27
Domain knowledge wins over algorithms
2 1
Algorithms
Domain
knowledge
Domain
expertise
Algorithms
LIMA Programme 2020 28
iStock-528612314
Deployment 4
LIMA Programme 2020
Shifting the power from IT to the pricing department
Rating Tables
Scenario Testing tool
ML Algorithm
ClientPolicy Admin System
Versioned pricing
IT departmentPricing department
30
49% of the interviewed clients is manually
import rating tables through their IT system,
and we would like to move to API
TechneGCU’s service bundle to boost your technical excellence
Technical Price
Roadmap
New Business
Pricing Strategy
Augmented Renewal
Pricing Strategy
▪ We identify the main risk drivers, blending traditional methods with the most recent ML
techniques to enhance the segmentation of your portfolio.
▪ Thanks to our RE_smoother tool we improve the predictiveness of geographical
or vehicle variables, enriching your portfolio with external information
▪ TimeMapper: a tool which enables you to leverage advanced spatial analysis to improve
the risk modelling prediction for your property book, pivoting on Risk Network Analytics
▪ We help you define your commercial rate for new business, identifying the profitable
target clients
▪ Through the reverse engineering of the market premiums, we support you in defining your
strategic positioning compared to the competitors and set up your go-to rates for any market
▪ Modeling the conversion probability i.e. the elasticity of a customer at the point of sale, we
help you optimize your competitive positioning on aggregators and price comparison websites
▪ We work with you to develop an advanced renewal strategy to optimize
performance KPIs such as the expected profit or the expected volume
▪ Thanks to sophisticated behavioral models, we identify the client elasticity and
factor the propensity to renew within the final premium, if the market allows it
▪ IDIAL is a MR tool which helps you defining and setting cap and floor system based on
advanced analytics, forecasting the evolution of the portfolio over a multi-year horizon
LIMA Programme 2020 31
TechneGCU’s service bundle to boost your technical excellence
Portfolio pruning
and steering
Pricing Sophistication
Assessment Tool
Pricing strategy &
Underwriting advisory
▪ We use predictive modelling to measure the expected profitability of the book, considering the
value of the single policy over a multi-year time horizon.
▪ We use leakage and cross-subsidy analysis to measure the “wellness” of the portfolio to
identify loss making and profit-making clients along the future history of the portfolio.
▪ We measure the economic benefit of pruning, selecting only risks which will not be profitable
over the time horizon, maximizing the efficiency of the impact.
▪ Thanks to the Pricing Sophistication Assessment Tool we benchmark your pricing processes
and methodology against your peers and the market
▪ Based on the outcome, we identify the areas of improvement and we help clients improving
the pricing governance and pricing flow
▪ We set up pricing and underwriting committee to help insurtech and digital players
in defining the best pricing strategy and portfolio steering
▪ Thanks to the MR platforms we provide our clients with the most advanced cloud-based
monitoring solution to follow the growth of the portfolio in real time and act quickly
▪ We helps insurers to build their pricing team, identifying the right profiles based
on the client needs
LIMA Programme 2020 32
Thank you
Massimo CavadiniHead of Actuarial Consulting and Data Analytics
Professional experience
▪ Senior consultant advising clients on improving the portfolio
performance through the technical excellence
▪ Senior pricing analyst in the Global P&C of Generali Group
▪ Pricing analyst in the Global P&C of Allianz SE
Non project related experience
▪ Implemented an international network of pricing experts
▪ Building pricing capabilities and best practices
Clients/projects
• Leading the Actuarial and Data Analytics practice within the Global
Consulting Unit
• Pricing committee member – direct insurance companies
• Designing pricing strategies for traditional companies and start-ups
• Extensive experience in risk modelling, demand modelling and price
optimization in Europe, LATAM and MENA region.
• Coordinating ML and AI projects to improve and enhance the technical
performances.
34
Tel
Mobile
+49 (89) 3891 8805
+49 160 9642 6759
https://de.linkedin.com/in/massimo-
cavadini-2376b837
Team
Location
Start date
Languages
Actuarial Consulting & Data Analytics
Munich
July 2017
English, Italian
LIMA Programme 2020
Disclaimer/Important Notice
LIMA Programme 2020
© 2020 Münchener Rückversicherungs-Gesellschaft Aktiengesellschaft in Munich ("Munich Re"). All rights reserved.
This presentation is for information purposes only. It is not a binding offer. Any obligations between the parties can only be imposed by a binding written
agreement. Munich Re is not obliged to enter into such agreement.
This document is strictly confidential and may not be copied, distributed or reproduced in whole or in part or disclosed or made available by the recipient to any other
person. It is intended to be read by only a limited number of persons which fall within certain exemptions. Please satisfy yourself that you are one of the intended
recipients.
Munich Re has used its discretion, best judgement and every reasonable effort in compiling the information and components contained in this presentation. It may not be
held liable, however, for the completeness, correctness, topicality and technical accuracy of any information contained herein. Munich Re assumes no liability with
regard to updating the information or other content provided in this presentation or to adapting this to conform with future events or developments.
Image: used under licence from shutterstock.com
35