Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
Customer Response to Usage Based InsuranceAn Analysis of Social Media Sentiments
Nick KuceraFebruary 25, 2014
Nick Kucera
Discussion
Social Media in General
Long term potential
Twitter Data
Analysis results
Conclusions
1
The Growth in Social Media
Social Media Platforms
I need to eat
I have fiveI ate
This is where I ate
I have five connections that recommend me because I eat so
llate
Why am I eating?
well
This is a review of where I ate Let’s all eat
t th
Watch me eat
together
3
Is the Industry Taking Advantage of Social Media?
• Insurance companies are investing significant resources in a social media presence
• Current and potential customers are voluntarily sharing intimate details of their life with the world
• Current and potential customers are interacting with• Current and potential customers are interacting with companies on a very personal level
• This information can be applied in different ways (service, marketing, competitive monitoring)
4
Twitter Data
• Twitter data ‐ over 3 million insurance total tweets from January, 2012 to present (Keyword
• Advantages of social media data
y p yof Allstate, State Farm, Geico, etc.)
• Data– Content of the tweet
– Unfiltered– Broad view of non‐
customer reactionso e o e ee– Specific tweet recipient– Sender of the tweet
Language of tweet
– Facilitates more timely analysis of trends
– Language of tweet– Where the tweet originated– Link to a picture of user
“This new world will– Latitude and longitude of the
user– Date and time of tweet
This new world will undermine the polling
industry” Fabio Rojas. How Twitter can help
d l
5
predict an election
Text Mining Data Processing
Data Processing Steps
• Remove punctuation and symbols (retain @ and #)
• Parse the tweet (35 words worked for Twitter – will need many more for other
Tweet ID
User Tweet Word1 Word2 … Word35
1 @mosley Text of tweet
W1 W2 … W35
ysources)
• Change table structures f ifrom tweets in rows to tweets in columns – keep indicator of order
Tweet ID Word Order Word1 1 Word11 2 Word2… … …1 35 Word35
• Correct spelling errors• Add word indicators
7
GEICO Top 10 Keywords
9 3%9.1%8.9%
piglike
commercials
GEICO Keywords
20.0%14.2%
11.2%10.6%10.5%
9.3%
commercialinsurance
moneyswitching
carpig
93.4%31.3%
0.0% 10.0% 20.0% 30.0% 40.0% 50.0% 60.0% 70.0% 80.0% 90.0% 100.0%
geicort
Percentage of Tweets
• Advertising (commercial, pig, commercials)• Brand recognition (“Saved money by switching car insurance”)• Sharing component significant (rt 31%)• Sharing component significant (rt 31%)
8
Analysis – Identifying Themes in the Data
Analysis – Identifying Themes in the Data
• Clustering/Segmentation– Unsupervised classification techniquep q– Groups data into set of discrete clusters or contiguous groups of cases
– Objects in each cluster tend to be similar, objects in different clusters tend to be dissimilar
10
Keywords – State Farm
Cluster #Number of
Tweets Category Keywords Description1 1,491 sponsorship around hit miss play run Home Run Derby2 3,778 sponsorship anyone college comes kids lebron College Sports Sponsorships3 1,198 advertising business music win Music Festivals, Music Download6 2,762 competition care cool funny much nationwide pretty Comparisons with Nationwide7 4,689 sponsorship another around football looks lost packers week Football sponsorships, Packers, Hish School Football8 1,260 advertising couple help jingle million nobody nothing remember sDo the jingle9 1,080 advertising another long named pretty talk wife woman Jake from State Farm, woman singing "Waterfall"10 5,142 advertising aaron came check discount double giants lost maybe p Discount Double Check, Packers12 1,217 quotes buy coverage insurance month quote start State Farm Marketing, Insurance Quotes15 471 advertising discounts drive insurance less through Discounts19 1,000 sports discounts giants hanging less more packers spent time Packers & Giants20 2,941 competition change having same Changing Insurance Companies25 2,414 marketing about care don florida insurance kids rates Great rates26 45 advertising accident another anyone anything believe commercial French model27 1,152 claims claims customers emergency help lost nationwide sandHurricane Sandy
11
Analysis – Identifying Themes in the Data
• Clustering/Segmentation– Unsupervised classification technique– Groups data into set of discrete clusters or contiguous groups of casesObjects in each cluster tend to be similar objects in– Objects in each cluster tend to be similar, objects in different clusters tend to be dissimilar
• Association Analysisy– Identification of items that occur together in the same record
– Can lead to sequence analysis as well, which considers timing and ordering of events
12
Long Term Potential of UBI
Long Term Potential of UBI
1 million
107 million
Significant Market
PotentialPotential
20%
14
“But Wait…” ‐ Glen Renwick, CEO of Progressive
“Intellectually…” “Surveys of Prospective Snapshot Users”
"Yeah
p
Yeah, why not?"30%"No way
in hell"40%
"Take
"Maybe, I need to
kSnapshot Option"100%
know more"30%
“Getting consumers to engage in a product that they were never asked to engage i [hi t i ll ] i bi b d i t ll t ll th ld h d ”
15
in [historically] is a bigger burden intellectually than we would have assumed.”
Snapshot Data
Twitter Data Subset
• Keyword searches for terms “Progressive” and “Snapshot” (3 400 tweets from September
• Advantages of social media data
(3,400 tweets from September 2012 to June 2013)
• Data
– Unfiltered– Broad view of non‐
customer reactions– Content of the tweet– Specific tweet recipient– Sender of the tweet
– Facilitates more timely analysis of trends
– Language of tweet– Where the tweet originated
Link to a picture of user– Link to a picture of user– Latitude and longitude of the
userD t d ti f t t
17
– Date and time of tweet
Analysis Process
• Analysis– Text mining– Review of individual tweets
• Categorization of tweetsStatus of user– Status of user
– Sentiment of the tweet– Positive feedback– Misconceptions– Customer complaints
P i i– Premium savings
18
Snapshot Analysis Results
Customer Status
Snapshot customer16%
Status
16%
Do not have snapshot48%
Being Monitored26%
Just received snapshot, just starting
the test
20
the test10%
Sentiment of Comment
Tone
31%90%
100%
Tone by User Status
Positive26%
30%7%
62%
31%
60%
35%
60%
70%
80%
of T
wee
ts
58%25%
30%
10%30%
40%
50%
Perc
enta
ge o
Negative55%
12%
39%29%
0%
10%
20%
Do not have
Just received
Being Monitored
Snapshot customerNeutral
19%have
Snapshotreceived
Snapshot, just starting
the test
Monitored customer
Status
21
Positive Neutral Negative
Non‐Customer Negative Responses
0.0% 5.0% 10.0% 15.0% 20.0% 25.0%Percent of Tweets
Non-Customer Negative Responses“im waiting for someone to leak documents showing that the #nsa can track people w/
th i h t thi ”
23.4%Privacy
the progressive snapshot thing”
18.2%Misunderstanding of Program
14.1%Rate Can Go Up
13.4%Concerned About Potential Results
22
Customer Positive Responses
• Price change vs. additional services
0% 20% 40% 60% 80%Percent of Tweets
Customer Positive Responses
• Customer validation• Some consumers like data76%Savings
14%Excitement
6%Tracking
“I'm liking Progressive's SnapShot device ‐not 'cause it'll lower my rates but 'cause it
tells me when I've done something I
4%Better Driving
23
tells me when I ve done something I shouldn't have.”
Customer Complaints
• Considerations– Behavior modificationComplaint
Starting Test Testing Customer Total
– Customer opinion– Feedback timing– Context
Driving Habits 34% 23% 16% 23%
Beeping 3% 21% 10% 16%Snapshot Evaluation 1% 20% 12% 15% – Context
– Negative reinforcement
Evaluation 1% 20% 12% 15%
Annoying 21% 13% 10% 14%
Premium Change 3% 10% 16% 10%
“Listen, Progressive Snapshot. You & I
Installation 28% 1% 0% 4%
will NEVER agree on what a ‘hard brake’ is & no amount of beeping is gonna
change that. #justsayin”
24
Reported Snapshot Discount
%0%5% 0%100%
ry
Reported Snapshot Discount by Tone
45 45
50
Reported Snapshot Discount
0%
2%
35%
17%%
60%
70%
80%
90%
Dis
coun
t Cat
egor
2930
35
40
45
ets
50%
83%93%
100%15%
92%
30%
40%
50%
60%
ntag
e fo
r Eac
h D
20
29 26
20
25
30
Num
ber o
f Tw
ee
0%
50%
8%0%
10%
20%
Zero Less than 10%
10% to 19%
20% to 29%
30% and G t
Tone
Per
cen
12
5
10
15
10% 19% 29% GreaterReported Discount
Positive Neutral Negative
-Zero Less than
10%10% to 19%
20% to 29%
30% and Greater
Reported Discount
“Just got an email that I'm getting a 7% discount on my auto “I've had the Progressive snapshot plugged in to my car
25
Just got an email that I m getting a 7% discount on my auto insurance after having that Snapshot thing. That's pretty
@Progressive. ;).”
I ve had the Progressive snapshot plugged in to my car to record my driving and get a discount. Out of 30% I only
got 2%. I suck.”
Conclusion
• UBI is here to stay, and will be a game‐changer• UBI has the potential for significant growth• Potential growth will not be achieved, however, without
listening closely to the voice of the customer• Ultimatel it ill help create a more positi e c stomer• Ultimately, it will help create a more positive customer
experience and improve the potential for UBI long‐term
26