Using Telematics Data Effectively The Nature Of Commercial Fleets Roosevelt C. Mosley, FCAS, MAAA, CSPA Chris Carver Yiem Sunbhanich November 27, 2017
About the Presenters Roosevelt Mosley, FCAS, MAAA, CSPA Pinnacle Actuarial Resources, Inc. Principal and Consulting Actuary Chris Carver SpeedGauge Yiem Sunbhanich TNEDICCA 1
Who? This is not just a session about big trucks! A fleet is any business with 5 or more vehicles. 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. 2
The Problem Heterogeneous vs Homogeneous Mix 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. 3
What s the Solution? 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. 4
Lift From Context Comparing different weight vehicles is difficult. Context adds more lift than traditional UBI. Relativity Example data for 160K vehicles Low Frequency High 5
Why Two Scores? For Commercial Lines, do not accept a single variable as descriptive of the driven exposure. People and vehicles change! Driver Score Driver score describes the real choices a driver makes every day. Vehicle Score FAIR Score, a proprietary exposure index, tells us about the context of the driven risk by VIN. 6
Misclassification Creates Residual Risk 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. 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% Exposure by Primary Rating Variable ISO Rated Actual 7
Residual Risk Captured Through Context 1.8 1.2 0.6 0.69 0.49 Liability Claims 1.25 1.18 0.92 0.97 0.84 0.9 1.44 1.32 1.8 1.2 0.6 0.81 0.69 Physical Damage Claims 1.151.12 0.950.93 0.940.96 1.27 1.18 0 20 40 60 80 100 0 20 40 60 80 100 BI PD Collision OTC Measuring speeding and braking alone only provides a self-selection benefit! 8
Thousands Building a Contextual Risk Score 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. 45 40 35 30 25 20 15 10 5 0 Actual Events versus Premium Relativity Safe 2 3 4 5 6 7 8 9 Risky Vehicle Count Collisions Pure Premium Relativity 3.5 3 2.5 2 1.5 1 0.5 0 9
External Context: Location Risk 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? Before During After Construction of the turn lane addition 5 1 Crash Frequency 4 3 2 1 0 1.4 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 0.7 0 2011 2012 2013 2014 2015 2016 Source: Analysis from TNEDICCA 10
External Context: Location Risk Road Modification Effect Example 2: New Roundabout After adding the turn lane, did the drivers who usually frequented this location become worse drivers? Before During Roundabout Construction After 25 1 Crash Frequency 20 15 10 5 0 9.4 2 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 0 2012 2013 2014 2015 2016 2017 Source: Analysis from TNEDICCA 11
Crash Locations Matter Process design drives outcomes more than an individual s behavior. Most traffic crashes consistently occur within a limited set of locations. 10% 66% 90% 34% Proportion of Locations Contribution of Total Crashes Source: Analysis from TNEDICCA 12
From Location Risk to Contextual Risk Score Demo: https://www.youtube.com/watch?v=2ijmrqb7pco&feature=youtu.be 13
UBI Score Analysis Steps to Contextual Analysis Analysis None Telematics data No associated loss experience Historical Experience Data Trip Summary Information Mileage Indicators Telematics data Historical loss experience Hard braking Harsh acceleration Speeding Concurrent Experience Contextual Information Telematics data Loss cost experience from the same period Internal context External context 14
Putting Telematics Data into Context Should these trips be evaluated differently? 15
Data Used for UBI Scoring Analysis Data Elements 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 Summary Statistics 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 16
Distribution Speed 6.0% 5.0% 4.0% 3.0% 2.0% 1.0% 0.0% Speed 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 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. 17
Speed Above Limit 1 23% 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 0 77% 0 1 18
-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 Distribution Speed Minus Speed Limit Speed Minus Speed Limit 4.5% 4.0% 3.5% 3.0% 2.5% 2.0% 1.5% 1.0% 0.5% 0.0% 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 19
-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 Distribution Adding Historical Context Change in Feet per Second 1.4% Change in Feet Per Second 1.2% 1.0% 0.8% 0.6% 0.4% 0.2% 0.0% Change in Feet per Second 20
Clustering/Segmentation 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 21
Cluster Distances 22
Cluster 2 Description Average change in feet per second (t 1 to t 2) = 156 Average change in feet per second (t 2 to t 3) = -55 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% 23
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 Distribution Speeding Distance from Cluster Mean 3.5% Distance from Cluster Mean 1.2 3.0% 2.5% 2.0% 1.5% 1.0% 0.5% 1 0.8 0.6 0.4 0.2 0.0% 0 Distance from Cluster Mean Series1 Speeding 24
Percentage of Trips Number of Different Clusters Assigned to Each Trip 40.0% Number of Behavior Clusters Assigned to Trip 35.0% 30.0% 25.0% 20.0% 15.0% 10.0% 5.0% 0.0% 1 2 3 4 5 6 7 8 9 1011121314151617181920212223242526272829 Number of Behavior Clusters Assigned to Trip 25
Summary The road ahead is clear, despite the picture in the rearview mirror: 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. 11/27/20 18 26
Questions 27
Join Us for the Next APEX Webinar 28
Final Notes 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 29
Thank You for Your Time and Attention Roosevelt Mosley RMosley@pinnacleactuaries.com Chris Carver chris.carver@speedgauge.net Yiem Sunbhanich Yiem.Sunbhanich@tnedicca.com Commitment Beyond Numbers 30