analysis of internet purchasing behavior · if aig equals 1 and 11.5
TRANSCRIPT
9/28/2011
1
Analysis of Internet Purchasing Behavior
October 3, 2011CAS I F S iCAS In Focus Seminar
Experience the Pinnacle Difference!Kevin LevittSenior Vice PresidentcomScore, Inc. Roosevelt Mosley, FCAS, MAAAPrincipalPinnacle Actuarial Resources Nick KuceraConsultantPinnacle Actuarial Resources
Discussion TopicsCurrent State of Insurance MarketingCustomer response analysescomScore background and datacomScore background and dataDescription of researchCharacteristics of shoppersModel development
Analysis of quotes submittedAnalysis of policies purchasedAnalysis of policies purchased
Impact of price on conversionUnstructured dataAdditional Research
2
9/28/2011
2
Current State of Insurance Marketing
Explosion in the investment that insurers are making in marketingMany different advertising mediums are being used
Traditional, internet, social media, etc.Has led to the understanding that customer behavior is about more than price
Customer service, reputation, convenienceDifferent elements are important to different market segments
3
Customer Response Analyses
Marketing Effort Quoting A l iQuoteSale
AnalysisConversion AnalysisLapse / Cancelation AnalysisRenewal QuoteRenewal
AnalysisRetention Analysis
4
9/28/2011
3
Customer Response Analyses
Quoting analysis: analysis of the likelihood of a prospective insured obtaining an insurance quote from youyouConversion analysis: analysis of the likelihood of a prospective insured that has received a quote purchasing insurance from youLapse/Cancelation analysis: likelihood of an insured not lapsing or canceling the policy mid-termR i l i l i f h lik lih d fRetention analysis: analysis of the likelihood of a current insured renewing with you
5
Customer Response Modeling –Challenges for Insurance Companies
Model structure and parameterizationNew territory – learning curveNew territory learning curvePriorityInternal data availabilityInternal data applicabilityAvailability of price change dataAvailability of price change dataMeasuring market competitivenessApplications
6
9/28/2011
4
comScore Background & Data
7
comScore is a Global Leader in Measuring the Digital World
NASDAQ SCOR
Clients 1700+ worldwide
Employees 900+
Headquarters Reston, VA
Global Coverage 170+ countries under measurement;43 markets reported
Local Presence 32 locations in 23 countries
V0910 8
9/28/2011
5
What We Do….We provide digital marketing intelligence that helps our customers make better-informed business decisions and implement more effective digital business strategiesp g gWe measure the continuous online activity of 1 million people in the U.S. who have granted us explicit permission to confidentially measure their Internet usage patterns.Our consumer panel is a representative cross-section of the U.S. population, worldwide regions and individual countriesWe also have permission to:
Survey panelistsSurvey panelistsMatch to third-party databasesAppend offline data
9
The Trusted Source for Digital Intelligence Across Vertical Markets9 out of the top 10INVESTMENT BANKS 9 out of the top 10AUTO INSURERS
47 out of the top 50 ONLINE PROPERTIES45 out of the top 50
4 out of the top 4WIRELESS CARRIERS 9 out of the top 10PHARMACEUTICALCOMPANIES9 out of the top 10
9 out of the top 10INTERNET SERVICEPROVIDERS
45 out of the top 50ADVERTISING AGENCIES9 out of the top 10MAJOR MEDIA COMPANIES
9 p 0CONSUMER FINANCECOMPANIES9 out of the top 10CPG COMPANIESV0910
10
9/28/2011
6
Auto Insurance Quote Detail
Data is captured from what panelists see using scraping technologyscraping technology
11
Auto Insurance Quote DetailData for 5 Top Auto Insurance Company Sites– Quote
• ZIP code• Bodily injury liability limits
– Drivers• Age• GenderBodily injury liability limits
• Coverage package• Premium quoted• Final Purchased Premium• Company name• Homeownership• Whether SSN entered• Primary driver education
• Gender• Marital Status• Industry/Occupation
– Vehicles• Vehicle year/make/model/type• Vehicle use• Annual mileage• Comprehensive deductibles
• Current Insurance Information• Whether currently insured• Length of gap in coverage• Length of current carrier• Length continuously insured• Prior BI Limit
• Collision deductibles– Incidents
• Incident Type• Incident Description
12
9/28/2011
7
Auto Insurance Quote Detail Data –Source of Traffic Data
Source of traffic is broken out into the following categories:
P id hPaid searchNatural searchWebmail sitesOther referredNon-referred
For search, we know the search engine, click type (paid/natural), and the search phrase For webmail sites and other referred, we know the referring site
13
Pinnacle & comScore Research
Population Target DescriptionMarketing Effort
Auto Insurance Website VisitorQuote Initiated Of those that visit the website, people with what characteristics are more likely to start the quote process?
Quote Initiated Quote Submitted Of those that initiate the quote process, what is the likelihood that they complete it?Q t P li Bi d Of those that actually submit the quoteQuote
Quote Submitted Policy Bind Of those that actually submit the quote, how many complete the purchase?Sale
14
9/28/2011
8
Characteristics of Shoppers
15
Data Summary – Means of Entry
570 559 54660080.0%
Model - Quote SubmittedMeans of Entry
46.8%32.6%
463546
411300400500
30.0%40.0%50.0%60.0%70.0% F
requency
of
Submitted
per
Number
of
Vis
16
9.7% 5.4% 5.5%0100200
0.0%10.0%20.0%
Natural Search Non-Referred Other Referred Sponsored Search Webmail
Quotes
1,000
sitors
Means of EntryExposure Percentage Frequency per 1000
9/28/2011
9
Data Summary – Session Day/Time
120.00 30.0%
Model: Purchase Made During Session
40.00
60.00
80.00
100.00
10.0%
15.0%
20.0%
25.0%
Purc
hase
Fre
quen
cy (p
er 1
,000
)
ure
(Ses
sion
s w
ith c
ompl
eted
quo
te)
-
20.00
0.0%
5.0%
Weekday M
orning
Weekday D
ay (9-5)
Weekday Evening
Weekday Late (10PM
+)
Weekend N
ight
Saturday
Sunday
Expo
s
Session Day/Time
Data Summary – Driver1 Age
100.00 18.0%
Model: Purchase Made During Session
20 00
30.00
40.00
50.00
60.00
70.00
80.00
90.00
4.0%
6.0%
8.0%
10.0%
12.0%
14.0%
16.0%
Purc
hase
Fre
quen
cy (p
er 1
,000
)
re (S
essi
ons
with
com
plet
ed q
uote
)
-
10.00
20.00
0.0%
2.0%
4.0%
<=18
19 20 21 22 23 24 25-29
30-34
35-39
40-44
45-49
50-54
55-59
60-64
65+
P
Expo
sur
Age Driver 1
18
9/28/2011
10
Data Summary – Number Vehicles/Drivers
90.00 70.0%
Model: Purchase Made During Session
20 00
30.00
40.00
50.00
60.00
70.00
80.00
20.0%
30.0%
40.0%
50.0%
60.0%
urch
ase
Freq
uenc
y (p
er 1
,000
)
re (S
essi
ons
with
com
plet
ed q
uote
)
-
10.00
20.00
0.0%
10.0%
1 / 1
1 / 2+
2 / 1
2 / 2
2 / 3+
3+ / < 3
3+ / 3+
Pu
Expo
sur
Policy Number Vehicles / Drivers
19
Data Summary – Bodily Injury Limit
20
9/28/2011
11
Model Development
21
Modeling Techniques
Neural N kDecision TreeNetwork Regression AnalysisEnsemble
22
9/28/2011
12
Quotes Submitted Analysis
23
Submitted Likelihood Model – Decision Tree
9/28/2011
13
Decision Tree – English RulesLow Estimate
IF total_pages < 11.5AND total ssl page < 9 5AND total_ssl_page < 9.5Quote Initiated: 0.4%
High EstimateIF aig EQUALS 1 AND 11.5 <= total_ssl_page AND 4.5 <= visit_time AND means_of_entry IS ONE OF: WEBMAIL NON-REFERRED SPONSORED SEAR NATURAL SEARCHNATURAL SEARCH AND total_pages < 15.5Quote Initiated: 71.4%
Quote Submitted – Number of Prior Visits
1 0001.200
Re
Number of Prior Site Visits
1.0000.889
0.7790.668
0.5570 4000.6000.8001.000l
ative
Likeli
Submitting
Q
260.0000.2000.400
0 1 2 3 4
hood
of
uote
Number of Prior VisitsRelative Likelihood of Submitting Quote
9/28/2011
14
Policy Purchased Analysis
27
Age of Driver
131% 16%18%1.40 Model: Purchase made during session
41% 47%67% 69% 76% 72%
100%117% 111% 118%
104% 104% 103%
62%4%6%8%10%12%14%
0.400.600.801.001.20
Expo
sure
Per
cent
age
Rela
tivity
20%0%2%4%
0.000.20
Age Driver 1Sessions with Completed QuoteModeled RelativityOne Way Relativity
28
9/28/2011
15
Education
45%50%2.50 Model: Purchase made during session
70%100%
126% 131% 146% 155%196%
149%
15%20%25%30%35%40%
1.001.502.00
Expo
sure
Per
cent
age
Rela
tivity
0%5%10%0.000.50
Primary Driver EducationSessions with Completed QuoteModeled Relativity
29
Price Sensitivity – One Session
106%60%1.20 Model: Purchase made during session
100%76% 79% 71%
52%76%
10%20%30%40%50%
0 200.400.600.801.00
Expo
sure
Per
cent
age
Rela
tivity
0%10%
0.000.20
Percent Final Quoted Premium greater than Minimum Quoted Premium
Sessions with Completed QuoteModeled Relativity30
9/28/2011
16
Price Sensitivity – One Month
31
Regression Preliminary Results
560%
1 151.20 Model: Purchase made during session
115%
100% 97%10%20%30%40%50%
0.951.001.051.101.15
Expo
sure
Per
cent
age
Rela
tivity
0%10%
0.850.90
Did user come to insurer’s site from another insurer’s site?
Sessions with Completed QuoteModeled Relativity
9/28/2011
17
Unstructured Data
33
Search Phrase Analysis
CategoriesSpecific companies
ProcessRemove punctuationSpecific companies
“Quotes” – auto, homeowners, life, insuranceValue – cheap, affordable, etc.Specific agents
Remove punctuation and non-characters [|./!#?*$%()-@]Parse phrase into wordsSummarize the frequency of each wordCorrect for misspellingsp g
Circumstance - teenOther – TurboTax, hydrogen fuel cell
p gBuild cluster analysis to determine key word associations
34Insuranced insurance: insurance0 insurance3 insurances insurancer insurancej insurancel insuranceg insurrance inssurance insurannce insurancce iinsurance innsurance insuraance isnurance insuarnce inusrance insuranceco
9/28/2011
18
Cluster Text Analysis
35
Summary & Additional Research
36
9/28/2011
19
Cumulative Lift
37
Lift Chart
38
9/28/2011
20
Additional Research
Population Target DescriptionMarketing Effort
Auto Insurance Website VisitorQuote Initiated Of those that visit the website, people with what characteristics are more likely to start the quote process?
Quote Initiated Quote Submitted Of those that initiate the quote process, what is the likelihood that they complete it?Q t P li Bi d Of those that actually submit the quoteQuote
Quote Submitted Policy Bind Of those that actually submit the quote, how many complete the purchase?Sale
39
Data Considerations
Voluntary panelOnly captures internet shopping trendsOnly captures internet shopping trendsDoes not marry online shopping with offline shoppingMay not track same user on separate computersp
40
9/28/2011
21
Potential Applications1. For potential insureds with a higher likelihood of purchasing
a policy, steps can be taken within the quote and purchase t th t th t d t d tprocess to ensure that the customer does not drop out
2. For potential insureds with a higher than average likelihood of purchase, insurance companies would be able to market more aggressively to these potential insureds
f l d h h l h3. For segments of potential insureds that have a lower than average likelihood of purchasing a policy, an insurance company can identify and investigate where these lower than average likelihoods are and try to increase the likelihood of purchase
41
Questions?Questions?
Experience the Pinnacle Difference!