loyalty

VOCLoyaltySurveyCanAimCustomer

RetentionProjects

TamiAxcellFebruary26,2010

yisameasure

lthplanorganizationhasudthe

analysisofasimplevoiceofthecustomer(VOC)surveytofindthedrivers

healthplan

hasmanyopportunitiestoimproveoperationalefficiencies,thisloyalty

analysisisaimedathelpingfocusSixSigmaDMAIC,DesignforSixSigma

(DFSS)andLeanprojectstoimprovesubscriberretention.

Theanalysismethodsarebestdescribedusingdatatoillustratethe

audhereishypothetical.

TheSurvey

Theloyaltyscoreisfiguredonthebasisofthreequestionsthat

subscriberstothehealthplananswerusingascaleof0to10,with10

stionsare:

ntendtore-enroll?

ourecommendtheplantofamilyandfriends?

eplanmeetyourexpectations?

Subscribersareconsideredloyaliftheyscore8to10onallthreeloyalty

questionsandareconsidereddisloyaliftheyscore0to4onanyone

actionivaluatedforthedriversofloyalty–quality

ofcare,accesstocare,customerrvice,claimspaymentand

rly,thedriversarecodedas

satisfiediftheyscore8to10andasdissatisfiediftheyscore0to

y,acomment

ctionasksthesubscriberhowtheplancanrvethembetter.

Table1:CustomerServiceandLoyalty

CustomerServiceLoyalty

DisloyalApatheticLoyalTotal

Dissatisfied808724191

level

correspondstothefivedriverspluscost,benefitsandamiscellaneous

tegoryissplitintosubcategoriesthataremorespecific.

Forexample,customerrvice(CS)hassubcategoriesof“MedicalOffice

CS,”“HealthPlanCS,”“DoctorCS,”rgersubcategories

arefurtherdividedtoclarifytheissue.

TheDataba

Thesurveydataarejoinedtodemographicinformationsuchasmarital

status,age,county,tenureinthehealthplanandemployergrouptype.

Medicalutilizationissummarizedbyfamilylevelcountsofprovider,

ographic,utilizationandcustomer

rvicecallcategorieswillbereferredtoasthedemographicvariables

andmanyofthearecodedintoavarietyofrollups(orcategorylevels)

mple,Tenure2hastwocategorylevels

–sixmonthstothreeyearsandmorethanthreeyearswiththeplan.

Similarly,isimportanttotrack

thedinrollmentofsurveyrespondersandthenon-responders.

TheChiSquareAnalysis

Themaindatabaissplitintoananalyticdatatdependingonthe

ststepistoperformchisquarecross-tabtablesof

loyaltybyallsurveyquestions(drivers)andthedemographicvariables.

Byconvention,theoutcome(loyaltyordinrollment)isinthecolumn

ofacross-tabandtheexplanatoryvariable(thedriverordemographic)

Count

PercentofRow

ChiSquareContrib.

42

328.93

46

3.79

13

64.29

100

Apathetic

Count

PercentofRow

ChiSquareContrib.

43

7

0

398

66

134.08

166

27

88.83

607

100

Satisfied

Count

PercentofRow

ChiSquareContrib.

14

1

55.19

234

20

85.04

898

78

102.67

1,146

100

Total

137

7

719

37

1,088

56

1,944

100

PearsonChi-Square=862.820/DF=4/p-Value=0.000

yconvention,ifthechisquarepvalueislessthan

0.05,thentheexplanatoryvariableisconsideredtohaveasignificant

nfluenceissignificant,thenthe

loyaltyanddisloyaltyrowpercentagesarecomparedtothetotalrow

percentage.

Rowpercentagedistributionandthecell’scontributiontothePearson

chisquarestatisticarebothvaluablestatisticsthatcanbeudto

e

1forexample,42percentofthowhoaredissatisfiedwithcustomer

llprovidesthegreatestcontributiontothe

tethattheoverallloyaltyis

56percent,butthosatisfiedwithcustomerrvicehavealoyaltyof

78percentandthodissatisfiedwithcustomerrvicehavealoyalty

of13percent,therefore,itisclearthatcustomerrviceisimportant.

Whereas,ifthosatisfiedwithcustomerrvicehad61percentloyalty

andthodissatisfiedwithcustomerrvicehad58percentloyalty,then

customerrvicewouldbealessimportantfactor.

gvaluesor

sohelpwiththeidentification

thatmost

ofdemographicvariableshaveanumberoflevels,suchasTenure2and

thelevelsmayhavesmallcellcounts(lessthanfive)

t

-significant

variablescanberuledoutanddonotneedtobeincludedinthelogistic

onlyonecategorizationforeachvariableinalogistic

model;forinstance,donotubothTenure2andTenure3.

TheLogisticRegressionAnalysis

Thecross-tabshelptoidentifywhatthebashouldbeforthelogistic

mple,aloyaltylogisticregressionanswersthequestion

ofwhattheoddsofbeingloyalareifthesubscriberissatisfiedwith

customerrvice,comparedtotheoddsofbeingloyalifthesubscriber

isdissatisfiedwithcustomerrvice.

Themathismorecomplicated,butthinkofitasaratioofthepercentages

prentedinthefirstcustomerrviceexampleabove:78percentdivided

retthisasthosatisfiedwithcustomer

rvicearesixtimesmorelikelytobeloyalthanthodissatisfiedwith

eordenominator,therefore,isimportantfor

atisticalprogramschoothebaaccordingto

ore,itisa

goodpracticetorenamethecategoriessotheanalystlectstheba.

Considerforexample,thecustomerrvicevariablewiththefollowing

categories“dissatisfied,”“apathetic”and“satisfied.”Ifthe

statisticalprogramchoosanascendingalpha,itwilllectapathetic

,itmakesthe

interpretationabittricky,becautheoddsratioofloyaltyfortho

satisfiedcomparedtothoapatheticisgreaterthanone(78/27=2.9)

andtheoddsratioofloyaltyforthodissatisfiedcomparedtho

apatheticwillbelessthanone(13/27=0.5).Second,theremaynot

bestatisticaldifferencebetweenapatheticandsatisfied,buttheremay

,itisnotalways

ingtotheloyaltyrow

percentagesinthecross-tabs,itiasytoidentifythecategorywith

thiscategoryasthebasothat

theoddsofloyaltyarerelativetothisandwillthereforebegreater

thanone.

Sincethedriverstendtobemorehighlycorrelatedwithloyalty,itis

usuallybesttorunthedriversanddemographicsinparatelogistic

models.

InterpretingCommentsandIdentifyingProjects

Thevolumeofaparticularcommentcategoryisnotnecessarilyagood

tsarebestillustratedby

abubblediagramthatplotsvolumeasthesizeofthebubbleagainstthe

xandyaxespercentagesofloyaltyanddisloyaltyofthomemberswho

bleswithhighdisloyaltyandlowloyaltyare

easilyidentifiedastargetsforSixSigmaprojects.

Theloyaltydriverscanbeillustratedinanimportance-performanceplot.

Usingtherawdatafrom0to10,calculatetheaverageofeachdriver

(performance-xaxis).Thencalculatethecorrelationcoefficientto

loyaltyforeachdriver(importance-yaxis).Calculatetheaverage

performanceandimportancescoreandplottheasreferencelinesto

-performance/high-importancequadrant

performance/importanceplotsandbubblediagramsareeffectivemethods

rereinforcedwiththe

tance,thosatisfiedwithcustomer

rvicearetwotimesmorelikelytobeloyalcomparedtothosatisfied

withqualityofcare,whoare15timesmorelikelytobeloyal(relative

tothodissatisfied).

Addingthethreeloyaltyquestionstoeverysurvey,suchasannualhealth

plansurveys,canhelpteaapartspecificattributesofthekeydrivers

gloyaltyanalyscan

thenbeprentedwithconfidencetoscopeandcontroltheDMAIC,DFSS

andLeaninitiativesdesignedtoimproveretention.

Figures1,2and3showwaystoprovidegraphicillustrationsofloyalty

1plots

2showscustomer

rviceinthehighimportanceandlowperformancequadrant;therefore,

3charts

example,“DoctorCS”shouldbeaddresdfirst,eventhough“Medical

OfficeCS”hasthegreatestvolume.

Figure1:RetentionVersusLoyalty

Figure2:ImportanceVersusPerformance

Figure3:CustomerServiceComments

RelationshipBetweenLoyaltyandRetention

Thefinalpointistounderstandtherelationshipbetweenretentionand

woyearsofsurveydata,retentionratescanstarttobe

ents

canbedeveloped,suchasdisloyalmembersdinrolltwiceasoftenas

herunderstandtheretentionratesandthesurvey

bias,comparetheretentionratesofsurveyresponders,non-responders

ionalso

canbeudasanoutcomevariableinthecross-tabsandlogistic.