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

RMosley@pinnacleactuaries.com

chris.carver@speedgauge.net

Chris Carver

Yiem.Sunbhanich@tnedicca.com

Yiem Sunbhanich