using telematics data effectively · • data was collected from class 1-8 trucks in seven...
TRANSCRIPT
November 27, 2017
Roosevelt C. Mosley, FCAS, MAAA, CSPAChris Carver
Yiem Sunbhanich
Using Telematics Data EffectivelyThe Nature Of Commercial Fleets
1
About the Presenters
• Roosevelt Mosley, FCAS, MAAA, CSPA• Pinnacle Actuarial Resources, Inc.• Principal and Consulting Actuary
• Chris Carver• SpeedGauge
• Yiem Sunbhanich• TNEDICCA®
2
We’re PAST the tipping point. Telematics, mobile phones, tablets, ELDs and/or cameras are present in 58% of fleet vehicles! Those who are not leveraging this data will soon realize they are being adversely selected.
Who?This is not just a session about big trucks! A fleet is any business with 5 or more vehicles.
3
• PL telematics provide a selection benefit for very homogeneous risks.
• Commercial Lines include a very heterogeneous mix of vehicles and drivers within a single business class.
• Fleet data has exposed the weakness of zone rating heavy vehicles and territory rating light vehicles.
• Driver turnover is up 270%.• 14% of vehicles are becoming
significantly safer every year, but repair costs on replacement vehicles are up 42%.
• We’re not mandating essential ADAS technologies, even though they yield a 61% reduction in frequency.
The Problem
Heterogeneous vs Homogeneous Mix
4
• The current hard market fuels specialty underwriting, but is fully underwriting everything our new reality?
• We should be looking for higher pass rates based on deep insights and fewer errors!
• Innovate with the self-equipped fleets: >58% have data and 73% are willing to share the data.
• Move away from unit rating/experience rating toward exposure rating.
• Rate the actual vehicle and the driver separately.• Require an updated driver list quarterly.
What’s the Solution?
5
Comparing different weight vehicles is difficult. Context adds more lift than traditional UBI.
Lift From Context
Relativity
Low Frequency HighExample data for 160K vehicles
6
Why Two Scores?
Driver ScoreDriver score describes the real choices a driver makes every day.
Vehicle ScoreFAIR Score®, a proprietary exposure index, tells us about the context of the driven risk by VIN.
For Commercial Lines, do not accept a single variable as descriptive of the driven exposure. People and vehicles change!
7
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
Exposure by Primary Rating Variable
ISO Rated Actual
• Vehicle Scores, combined with detailed driving data, produce a selection benefit 8.5 times more predictive than territory rating.
• Errors in exposure distribution create large rating errors for fleet vehicles, thus creating residual risk.
Misclassification Creates Residual Risk
8
Measuring speeding and braking alone only provides a self-selection benefit!
0.690.84
0.97
1.181.32
0.49
0.92 0.9
1.25
1.44
0
0.6
1.2
1.8
20 40 60 80 100
Liability Claims
BI PD
0.69
0.95 0.94
1.151.27
0.810.93 0.96
1.12 1.18
0
0.6
1.2
1.8
20 40 60 80 100
Physical Damage Claims
Collision OTC
Residual Risk Captured Through Context
9
• Data was collected from Class 1-8 trucks in seven different programs (included loss sensitive and guaranteed cost clients).
• Geographic area included the lower 48 states.
• All vehicles used telematics.
• Data was normalized by platform and vehicle weight.
• Exposure is based on time on each road segment, not miles.
Building a Contextual Risk Score
0
0.5
1
1.5
2
2.5
3
3.5
0
5
10
15
20
25
30
35
40
45
Safe 2 3 4 5 6 7 8 9 Risky
Th
ou
sa
nd
s
Actual Events versus Premium Relativity
Vehicle Count Collisions Pure Premium Relativity
10
External Context: Location Risk
Source: Analysis from TNEDICCA®
Before During
Construction of the turn lane addition
After
1.4
0.7 0
1
0
1
2
3
4
5
8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12
2011 2012 2013 2014 2015 2016
Cra
sh
Fre
qu
en
cy
Road Modification Effect Example 1: Adding a Turn Lane
After adding the turn lane, did the drivers who usually frequented this location become “safer drivers”?
11
After adding the turn lane, did the drivers who usually frequented this location become “worse drivers”?
External Context: Location Risk
Source: Analysis from TNEDICCA®
Before During
Roundabout Construction
After
2
9.4
0
1
0
5
10
15
20
25
9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12
2012 2013 2014 2015 2016 2017
Cra
sh
Fre
qu
en
cy
Road Modification Effect Example 2: New Roundabout
12
Source: Analysis from TNEDICCA®
Crash Locations Matter
Process design drives outcomes more than an individual’s behavior. Most traffic crashes consistently occur within a limited set of locations.
90%
34%
10%
66%
Proportion ofLocations
Contribution of TotalCrashes
13
From Location Risk to Contextual Risk Score
Demo:https://www.youtube.com/watch?v=2iJmRqB7pco&feature=youtu.be
14
• Telematics data
• No associated loss experience
None
• Telematics data
• Historical loss experience
Historical Experience
• Telematics data
• Loss cost experience from the same period
Concurrent Experience
UBI Score Analysis – Steps to Contextual Analysis
• Mileage
Trip Summary Information
• Hard braking
• Harsh acceleration
• Speeding
Indicators
• Internal context
• External context
Contextual Information
Analysis Data
15
Putting Telematics Data into Context
Should these trips be evaluated differently?
16
• Heading
• Change in heading (prior 4 readings)
• Speed
• Change in speed (prior 4 readings)
• Feet per second
• Change in feet per second
• Speed limit
• Speed – speed limit
• Speeding indicators (0, 5 and 10 mile buffers)
• Road class
• Hour
Data Used for UBI Scoring Analysis
• 59,000 unique trips (new trip starts when a vehicle is at rest for 60 or more minutes)
• 2.7 million miles
• Trip length average: 2 hours and 15 minutes
• Average distance traveled per trip: 45 miles
• Average mileage per day: 123
Data Elements Summary Statistics
17
0.0%
1.0%
2.0%
3.0%
4.0%
5.0%
6.0%
0 4 8 12 16 20 24 28 32 36 40 44 48 52 56 60 64 68 72 76 80 84 88 92 96
Dis
trib
uti
on
Speed
Speed
Speed
• In general, higher speed translates into a worse driving evaluation.
• Many plans use a single speed cut-off.
• In this example, a cut-off of 70 miles or hour results in 3.6% of the readings having a negative evaluation.
18
077%
123%
Speed Above Limit
0 1
Speed Above Limit
• Indicator = 1 for readings where speed is greater than speed limit
• Begins to add some context to the raw speed measure
• Still does not tell the entire story
19
0.0%
0.5%
1.0%
1.5%
2.0%
2.5%
3.0%
3.5%
4.0%
4.5%
-75
-70
-65
-60
-55
-50
-45
-40
-35
-30
-25
-20
-15
-10 -5 0 5
10
15
20
25
30
35
40
45
50
55
60
65
70
Dis
trib
uti
on
Speed Minus Speed Limit
Speed Minus Speed Limit
Speed Minus Speed Limit
• Calculation of speed minus speed limit
• Provides additional context and risk segmentation –takes a single indicator and provides additional segmentation of the risk
20
0.0%
0.2%
0.4%
0.6%
0.8%
1.0%
1.2%
1.4%
-25
-24.1
-23.2
-22.3
-21.4
-20.5
-19.6
-18.7
-17.8
-16.9
-16
-15.1
-14.2
-13.3
-12.4
-11.5
-10.6
-9.7
-8.8
-7.9 -7
-6.1
-5.2
-4.3
-3.4
-2.5
-1.6
-0.7
0.2
1.1 2
2.9
3.8
4.7
5.6
6.5
7.4
8.3
9.2
10.1 11
11.9
12.8
13.7
14.6
15.5
16.4
17.3
18.2
19.1 20
20.9
21.8
22.7
23.6
24.5
Dis
trib
ution
Change in Feet per Second
Change in Feet Per Second
Adding Historical Context – Change in Feet per Second
21
• Traditional Analysis
– Count of braking, speeding, acceleration
– Application of research studies to traditional data
• Clustering/Segmentation
– Unsupervised classification technique
– Groups data into set of discrete clusters or contiguous groups of cases
– Performs disjoint cluster analysis on the basis of Euclidean distances computed from one or more quantitative input variables and cluster seeds
– Data points are grouped based on the distances from the seed values
– Objects in each cluster tend to be similar, objects in different clusters tend to be dissimilar
• Benefit – telematics readings classified based on entire record, not just value of one element
Clustering/Segmentation
22
Cluster Distances
23
• Average change in feet per second (t – 1 to t – 2) = 156
• Average change in feet per second (t – 2 to t – 3) = -55
Cluster 2 Description
Element Cluster 17 Overall Average
Highway Road Class 45.7% 15.1%
Time of Day: 12 – 5 am 0.8% 0.4%
Time of Day: 7 – 8 am 1.0% 0.6%
Time of Day: 5 – 9 pm 10.5% 9.0%
Time of Day: 9pm – 12 am 2.2% 1.4%
24
Distance from Cluster Mean
0
0.2
0.4
0.6
0.8
1
1.2
0.0%
0.5%
1.0%
1.5%
2.0%
2.5%
3.0%
3.5%0
.7
1.1
1.5
1.9
2.3
2.7
3.1
3.5
3.9
4.3
4.7
5.1
5.5
5.9
6.3
6.7
7.1
7.5
7.9
8.3
8.7
9.1
9.5
9.9
10.3
10.7
11.1
11.5
11.9
12.3
12.7
Sp
eed
ing
Dis
trib
uti
on
Distance from Cluster Mean
Distance from Cluster Mean
Series1 Speeding
25
Number of Different Clusters Assigned to Each Trip
0.0%
5.0%
10.0%
15.0%
20.0%
25.0%
30.0%
35.0%
40.0%
1 2 3 4 5 6 7 8 9 1011121314151617181920212223242526272829
Perc
en
tag
e o
f T
rip
s
Number of Behavior Clusters Assigned to Trip
Number of Behavior Clusters Assigned to Trip
26
• Telematics data isn’t the only way to achieve pricing precision, but it helps!
• Ignoring telematics data limits pricing innovation.
• Much of the premium leakage comes from the what, by whom and how much a vehicle drives.
• Rate each vehicle, and you’ll discover at least 6% of rate.
• Knowing what’s happening on the road ahead is going to prepare you for the future.
Summary
11/27/20
18
The road ahead is clear, despite the picture in the rearview mirror:
27
Questions
28
Join Us for the Next APEX Webinar
29
• We’d like your feedback and suggestions
• Please complete our survey
• For copies of this APEX presentation
• Visit the Resource Knowledge Center at Pinnacleactuaries.com
Final Notes
30Commitment Beyond Numbers
Thank You for Your Time and Attention
Roosevelt Mosley
Chris Carver
Yiem Sunbhanich