DIGITAL BEHAVIORAL DESIGN

Presentation on theme: "DIGITAL BEHAVIORAL DESIGN"— Presentation transcript

1 3NirEyal.Hooked:Howtobuildhabit-formingp
3NirEyal.Hooked:Howtobuildhabit-formingproducts.Penguin,2014ThePowersofHabit4,RichardThalerandCassSunsteinÕsworkon4CharlesDuhigg.Thepowerofhabit:Whywedowhatwedoinlifeandbusiness,volume34.Rand

2 omHouse,2012ChoiceArchitectureinNudge5,D
omHouse,2012ChoiceArchitectureinNudge5,DanArielyÕsworkonBehavioral5RichardHThalerandCassRSunstein.Nudge:Improvingdecisionsabouthealth,wealth,andhappiness.HeinOnline,1999EconomicsinPredictably

3 Irrational6,DanielKahnemanÕspioneering6D
Irrational6,DanielKahnemanÕspioneering6DanAriely.Predictablyirrational.HarperCollinsNewYork,2009workondecisionmakinginThinkingFastandSlow7,RobertCialdiniÕs7DanielKahneman.Thinking,fastandslow

4 .Macmillan,2011Inßuence It omputationalN
.Macmillan,2011Inßuence It omputationalNeuroscienceRecentresearchhasmadestridesinunderstandingthebrainÕsmotivationmachinery.Thestudyofbrainstructures,chemicals,andcircuitsresponsibleforhowwel

5 earnfromexperience21have21particularlyfr
earnfromexperience21have21particularlyfromreinforcementmadesigniÞcantprogressinthepastdecadefrombasicresearch,medicine,andevenartiÞcialintelligenceandroboticsresearch.Wenowhaveclearerundersta

6 ndingsofhowthebrainformshabits, muchchea
ndingsofhowthebrainformshabits, muchcheaperthanyoucouldhavein themutually-reinforcingrelationshipbetweenacademicscience,high-techcompanies,ispushingtomakesthesetoolseasiertouse.frameworksaref

7 reelyavailable,anyonecanstartusingpatter
reelyavailable,anyonecanstartusingpatterndetectors,predictionengines,andautomationpipelinesthatmighthave,10yearsago,takenseveralPhDÕsworthofinnovation,andonlybeenusablebythehandfulofexpertswh

8 obuiltthem.Bettertoo,dozensoffreevideoco
obuiltthem.Bettertoo,dozensoffreevideocoursesfromleadinguniversitieshelpanyone,globally,startbuildingsophisticatedmachinesthatlearn.2323DaltontookAndrewNgÕsduringgraduateschool.100%recommende

9 d.Today,yourappcansenseauserÕsbehavior.I
d.Today,yourappcansenseauserÕsbehavior.ItcansendthatdatatotheCloudinstantly.ThatdatacanbeprocessedbyPersuasiveAIdesignedtomodifytheirbehavior.ThatAIcaninstantlyrecommendtoyourapphowtomodifyit

10 sdigitalenvironment(itsUX)toinducethatus
sdigitalenvironment(itsUX)toinducethatusertochangehisbehavior.Allofthiscanbebuilt-andoperated-cheaplyandscalably-touchingnotonlythatuserÕsmindbutmillionsofmindsgloballyeverysecond.2424thisisl

11 iterallywhatwedoat BehavioralDesigncanbe
iterallywhatwedoat BehavioralDesigncanbeanextremelystrongdrivingforceinbuildingthatworldwhenusedwithinastrongethicalframework.Asfaraswecantell,thereareatleasttwomajorethicalquestionsunderpinn

12 ingBehavioralDesign.1.)IsanyBehavioralDe
ingBehavioralDesign.1.)IsanyBehavioralDesignethicalatall?2.)IsthisparticularuseofBehavioralDesignethical?ÒBehavioralDesignpatternsaredarkorlightdependingonhowyouusethem.DonÕtletthevillainsbet

13 heonlyoneswithsuperpowers.Ó!BehavioralDe
heonlyoneswithsuperpowers.Ó!BehavioralDesignisasetoftechniquesforpersuasion.Itisatechniqueofcoercion.Whenusedproperly,BehavioralDesigntakesadvantageofcognitivebiasestomakeparticularbehaviorsm

14 orelikely:itdoesnotforcecertainactions.I
orelikely:itdoesnotforcecertainactions.Itdoesnotuseviolenceorthreatstochangebehavior.ItdoesnotliterallyrestrictsomeoneÕsabilitytoactornot.Itmerelyprovidestestable,provablewaytoincreasethechan

15 cesthatbehaviorschange:itdoesnotforcecha
cesthatbehaviorschange:itdoesnotforcechange.ItisaTechnologyofBehavior,notaTechnologyofForce.Tothatextent,thetechniquesofBehavioralDesign,andDesignersthemselves,mustrespectpersonsÕintrinsicrig

16 htstofreedomof AlignmentwithSocialGood3.
htstofreedomof AlignmentwithSocialGood3.AlignmentwithaUserÕsDesires1.TransparencyBehavioralDesign,andourusesofit,aremostethically-aligned behaviors.AmbientCommunicationOften,youÕllneedtoprese

17 ntcomplexordenseinformationtosomeone.Amb
ntcomplexordenseinformationtosomeone.AmbientCommunicationprescribesthatusingnon-textcommunication,suchascolor,size,texture,pattern,motion,sound,vibration,ortimecanhelpyoucommunicatecomplexcon

18 tentto roupStructureIfyourAppconnectspeo
tentto roupStructureIfyourAppconnectspeopletooneanother,howdoyouknowthebestwayeachuserwouldprefertoexperiencethissocialcomponent?Differentusersexhibitdifferentpreferencesfortheirinteractionsw

19 itheachother:somefeelmostmotivatedalone,
itheachother:somefeelmostmotivatedalone,othersinsmallgroupsofpeers,andsomeinthefullgazeofpublic.OptimalGroupStructureproposesthatthereexistpredictable,optimalscopesandnaturesofinteractivitybe

20 tweenappusersthatwillbestmotivatethemtoc
tweenappusersthatwillbestmotivatethemtochangetheirbehavior.PersonalizationYoumayhaveheardofthehypothesisthatstudentshaveÒlearningstyles,Ó34andthatstudentslearnbestwhenmaterialisadaptedto34Vis

21 ual,Auditory,Tactile,Converger,etc.thats
ual,Auditory,Tactile,Converger,etc.thatstyle.ThathypothesishasbeensoundlyrepeatedlyfalsiÞed.3535ScottOLilienfeld,StevenJayLynn,JohnRuscio,andBarryLBeyerstein. )mentallyassociatedtotheperforma

22 nceaparticularaction.Youcanthinkofitlike
nceaparticularaction.YoucanthinkofitliketheÒIfthis�ÐthenthatÓpairingofacuetoanaction:ifsensecue,thendoaction.Non-BehavioralDesignersonlyfocusontheactionasthewhole understandingwhethero

23 rnotauserwillact.HisMATModel3838Dr.FoggÕ
rnotauserwillact.HisMATModel3838Dr.FoggÕsMATModel,onlinesuggeststhatusersperformahabitwhentheyhaveadequate[M]otivation,[A]bility,andinthepresenceofaSynthetic[T]rigger.AsaBehavioralDesigner,yo

24 uneedtobalancehowmotivatedauserisagainst
uneedtobalancehowmotivatedauserisagainsthowdifÞcultataskis.Ifanactionistoohard,theuserwonÕtdoit.Iftheactionistooeasy,thenyouÕvewastedtheuserÕsmotivationalpotential.Theywerereadytodosomethingh

25 ard,butyoudidnÕtaskenoughfromthem.Ideall
ard,butyoudidnÕtaskenoughfromthem.Ideally,yourproductshouldpresentasyntheticcueforanactionthatsomeoneisabletoperformwhentheyÕremotivatedtoact.SomeActionsareeasiertoturnintouserhabitsthanother

26 s.WhenweatBoundlessMindhelpourclientside
s.WhenweatBoundlessMindhelpourclientsidentifywhatactionsinsidetheirappwouldbebesttoturnintoahabit,weusethefollowingcriteria:1.SmallActionsarebetterthanlargeActionsFavorashortactionthatcanbequ

27 icklyaccomplishedoversomethingthattakesa
icklyaccomplishedoversomethingthattakesalotoftime,rigour,focus,energy,orotherscarceresources.Theuserisalmostalwaysbusy.Theyhavetimeforyourapp-andyourhabit-solongasitcouldalreadyÞtintotheirlif

28 e.2.SpecificActionsarebetterthangeneral,
e.2.SpecificActionsarebetterthangeneral,largerbehaviorsWhatuserbehaviorsinyourapplendthemselveswelltobecomingahabit?What RememberthatonefoodordrinkyoujustCANÕThaveanymorebecauseofthatonetimei

29 tmadeyoureaaaallysick?Yeah.ThatÕsthekind
tmadeyoureaaaallysick?Yeah.ThatÕsthekindofone-shotlearningthebrainÕsgoodatfor TotheBehavioralDesigner,ÒrewardÓdoesnÕtjustmeansomeprizeearned.ÒRewardÓisbroaderthanagiftorabonusorathingwecanmea

30 sureindollars.AsBehavioralscientistsandB
sureindollars.AsBehavioralscientistsandBehavioralDesigners OneofSkinnerspigeonsinanoperantchamber.Whenthelightcameon(cue),abirdwouldpeckatthebutton(action)andthebuttonwouldclick(feedback).AtÞ

31 rstbirdsdidthismerelybychance.Butasfoodp
rstbirdsdidthismerelybychance.Butasfoodpellets(reward)beganbeingreleasedaftersomepecks-unexpectedly-thebirdsquicklylearned.Somebirdsgotapelleteverypeck.Somegotpelletsveryrarely.Some,inbetween

32 .Somegotapelleteveryfewminutes,somegotap
.Somegotapelleteveryfewminutes,somegotapelletatrandomtimes.Somegotapelletafterevery3rdpeck.Somegotapelletafterarandomnumberofpecks.Skinnernoticedsomethingfascinating.Somebirdsstartedpeckingmu

33 ch,muchmorethanothers.!"#$Thepelletswere
ch,muchmorethanothers.!"#$Thepelletswereallthesame.Thebirdswerelargelysimilartooneanother.Sowhataccountsforthedifference?Hefoundthatthepatternofpeck"pelletorpeck"no-pellet47waschangingtheirpe

34 cking.47HecalledtheseÔschedulesofreinfor
cking.47HecalledtheseÔschedulesofreinforcementÕThebirdsthatreceivedpelletsafteraseeminglyrandomnumberofpecks(whathecalledaVariableRatioreinforcementschedule)learnedtoassociatethesyntheticcuel

35 ightwiththebutton-peckactionmuchfasterth
ightwiththebutton-peckactionmuchfasterthananyoftheothergroupsofbirds.Andtheypeckedmuch,muchmore.Theirfuturepeckingwasbeingactivelyprogrammedbytheconsequencesoftheiractions(overwhichheheldcont

36 rol.)NotonlydidSkinnerhaveevidencethatbe
rol.)NotonlydidSkinnerhaveevidencethatbehaviorisshapedbyreward,hefoundthatthepatternofreward-whenandhowitÕsgiven-wascriticaltoshapingbehavior.AfewyearsearlieratMcGillUniversity,Drs.OldsandMil

37 nertookamoredirectapproachtoshapingbehav
nertookamoredirectapproachtoshapingbehavior.Theyskippedthetaste-budsentirelyandinsteadplacedelectrodesdirectlyintoratsÕbrains.48ElectricalstimulationintheSeptumbrainregionwould48SpeciÞcally,t

38 heyplacedtheelectrodesintoabrainregionkn
heyplacedtheelectrodesintoabrainregionknownastheSeptum(locatedimmediatelybeneaththeBasalGangliaÕsmainpleasurecenter).causetheemotionalsensationofpleasurefortherats.Thentheygavetheratsasmallle

39 vertheycouldpush.Thatleverwouldsendaburs
vertheycouldpush.ThatleverwouldsendaburstofelectricitytotheSeptum.Andman,theratÕsLOVEDthatlever.Infact,theratswouldpressithundredsoftimesanhour.Sometimeseventhousandsof RewardsoftheHuntReward

40 softheHuntsatisfyourdesireforconquest.Th
softheHuntsatisfyourdesireforconquest.Thegainsofvictory.Thesetiecloselytotheprimaryreinforcersthatourbrainsevolvedbecausetheymostcloselymimicreality(hence,thenameHunt.)BehavioralDesignersofte

41 nuseRewardsoftheHuntingames,insettingswi
nuseRewardsoftheHuntingames,insettingswithcompetition,and(surprisingly),insituationswherethenextblastofdopaminemightbejustafewswipesaway.5252WeÕrelookingatyou,Tinder.Dotheuseractionsinyourapp

42 haveanobviouscomponentofcompetition?Dous
haveanobviouscomponentofcompetition?Dousersseekinsideyourapp?Arethecorebehaviorsyouneedtoincreasepartofagame?Partofachase?Ifso,whatsortofRewardoftheHuntcanyouimplementtosurpriseanddelightthem

43 ?R maximizegainsfromyourcurrentusers,ita
?R maximizegainsfromyourcurrentusers,itactivelyhelpsyougrow DevelopingaCultureofExperimentation,aworkenvironmentwherepeopleareinformed,empowered,andrewardedforexperimentation,iscriticalforsuc

44 cess.AsManagement,itÕsyourjobtocreatethe
cess.AsManagement,itÕsyourjobtocreatetheemotionalandoperationalspacethathelpsProduct,Design,andEngineeringTeamsexperiment.Forthemtobecreative,takerisks,andlearnastheyiterate.IfyouÕreusingaLea

45 norAgiledevelopmentmethodology,muchofthi
norAgiledevelopmentmethodology,muchofthismightfeelintuitive.TrackingexperimentationandtinkeringasaninternalperformancemetriccanshowyouifyouÕresucceedinginbuildingaCulturethatencouragespeoplet

46 oexperimentandlearn,andforgivesandunders
oexperimentandlearn,andforgivesandunderstandswhenthingsgosideways. Goal3:PushNotiÞcationsareOwnedbyProduct,notbyMarketing.Forhistoricalreasons,MarketingandMessagingTeamsoftenownresponsibility

47 foraProductÕsuserretention.Thisismostcom
foraProductÕsuserretention.ThisismostcommoninteamsthatstartedaftertheadventoftheWeb,butbeforetheriseofMobile.BeforeMobile,theseteamsusedemailtore-engagelostusers,ormarketnewproductstoretained

48 users.WhenPushNotiÞcationscameout,itseem
users.WhenPushNotiÞcationscameout,itseemedlikeanaturalÞt:PushNotiÞcationswerejustre-engagementemails,butfaster.SothesameMessagingandMarketingTeamsbecameresponsiblefordesigningandsendingPushNo

49 tiÞcations,andbyextension,userretention.
tiÞcations,andbyextension,userretention.. ! . . . Weunderstandhowthiscametobe,wejustthinkitÕsaterriblesetup.ThisturnedPushNotiÞcationsintoachannelforspam.Spamthatjusthappenedtob

50 eclosertousersÕbrains.ItcreatedthePush-o
eclosertousersÕbrains.ItcreatedthePush-overloadworldweliveintoday.ItexplainswhyPusharejustoftentreatedassmallercrappieradvertisements,insteadofthoughtofascues. Segmenting arerun.Werecommendse

51 gmentingaportionofyouruserbasetoconducta
gmentingaportionofyouruserbasetoconductarolloutexperiment.Deploythenewchangestothatsegment.MeasurehowuserbehaviorKPIsarechangingovertime,andhowthesechangesaredifferentbetweenthewholeuserbasea

52 ndtheexperimentalsegment(or,betweenacont
ndtheexperimentalsegment(or,betweenacontrolgroupandatreatmentgroupintheexperimentalsegment).IfsheseesstatisticallysigniÞcantchangesinuserbehaviorthatcorrelatewiththechange(andnoothermodiÞcati

53 ons,norchance),thatÕsasignalthattheinter
ons,norchance),thatÕsasignalthattheinterventionisworkingasdesigned!Sheshouldfeelempoweredscalingoutthedeployedchangetotheentireuserpopulation.Repeat!Often!IfthemeasuredbehavioralKPIsarenÕtcha

54 nging,orarechanginginwaysthatsuggestthey
nging,orarechanginginwaysthatsuggesttheyÕrehurtingengagementandretention,shecanrollbackthedeployment,documentwhathappened,explorerootcauses,andreturnagainwithadifferentexperiment.Noharmdone:s

55 omethingextremelyvaluablewaslearned. Int
omethingextremelyvaluablewaslearned. Inthisexample,aKPIyoumightconsidertrackingforthisteammaybehowsubtleofachangeinuserengagementcanbeaccuratelypredictedbythetooltheyÕredeveloping.Astheyexper

56 imentanditerateandexperimentanditerate,y
imentanditerateandexperimentanditerate,youcouldexpectthemtomakesomequantiÞableprogresstowardimprovingthetoolÕsprecisionandaccuracyatpredictinguserengagementchanges.If,evenwithmorework,theirto

57 olÕspredictionsdonÕtbecomemoreaccuratewi
olÕspredictionsdonÕtbecomemoreaccuratewithtime,itÕslikelythatÕsasgoodasthemeasureisgoingtoget,andtheteamshouldstartworkingonadifferentgoal.BecausetheyÕrelargelyoperatingattheedgeofbothhumanun

58 derstanding,andtechnology,theirworkwillo
derstanding,andtechnology,theirworkwilloftenresembleacademicresearchmorethananything.Because,tobefair,theyÕreliterallybuildingtoolsnooneÕsevermadetosolveproblemsnooneconsideredsolvable.CutÔem

59 someslack.Considerfosteringcollaboration
someslack.ConsiderfosteringcollaborationsbetweentheseDataScienceTeamsandotherinternalteamswitheventslikeinternalsymposiaanddatahackathons.Thesewillgivethedatascientistsvisibilityonwhatproblem

60 sotherteamsareseeing,andshowtheotherteam
sotherteamsareseeing,andshowtheotherteamswhatkindofproblemstheDataScienceTeamisabletosolve.Also,Ô20%timeÕthatallowstheDataScienceTeamfreedomtotinkercancreatethekindofinnovationthatmanagementd

61 idnÕtknowtoaskfor.6767Alotofgreatscience
idnÕtknowtoaskfor.6767Alotofgreatsciencestartswithabeeranddoodles.Trustus.WeÕre ehavioraliststoFloatOnceyourProductteamhasincorporatedBehavioralDesigndeeplyintotheirprocessesandtools,theBehav

62 orialistsandpolymathsyouÕvestaffedupwith
orialistsandpolymathsyouÕvestaffedupwithshouldbeallowedtodissociatefromtheProductTeamtofocustheirattentiononyourotherbusinessprocesses.Grantthemtheirownbudget,aplacethemorthogonaltoyourstanda

63 rdorganizationchart(perhapsreportingtoaC
rdorganizationchart(perhapsreportingtoaCTOorCOO),butnoimmediateabilitytoforcechanges.Treatthemlikeunderpowered,in-houseconsultants.TheirfreedomtounhookfromProductandobservehowotherhuman-facin

64 gpartsofthebusinessoperatewillnotjustspr
gpartsofthebusinessoperatewillnotjustspreadthesetechniquesandtheirimpacttotherestofyouroperations,itwillencouragethetypeoflateralthinkingthatwilldrivefutureinnovationbackwithintheProductTeami

65 tself.OnceotherteamsarefamiliarwiththeBe
tself.OnceotherteamsarefamiliarwiththeBehaviorTeam,andtrusthasbeenestablished,encouragetheBehaviorTeamtoproviderecommendationstotheseotherbusinessunits,andincentivizethe certainpredictableway

66 s.Inthischapter,weexploreseveralexamples
s.Inthischapter,weexploreseveralexamplesofProductsandEnvironmentsyouÕrealreadyfamiliarwith,andhighlighttheBehavioralDesigntechniquestheyÕveusedtoengineerhowpeoplebehave.ReinforcementLearninga

67 ndCues:FacebookFacebookhasmasterfullyemp
ndCues:FacebookFacebookhasmasterfullyemployedBehavioralDesigntechniquestobuildoneofthemosthabit-formingproductsonanyoneÕsphone.WhiletheyÕveusedseveraldifferenttechniques,wefocushereontwo:Rein

68 forcementLearningandCues,whichtheyÕvecom
forcementLearningandCues,whichtheyÕvecombinedtogetheranimplementationofCARModel.ÒPINGÓyouhaveanewnotiÞcation!CananyonereallyresisttheurgetoopentheappandseewhatÕsinside?!ThatnotiÞcationstartst

69 henextchainofeventstounfold:youunlockyou
henextchainofeventstounfold:youunlockyourphone,ÞndtheFacebookApp,andopenit.AsdescribedbytheCARModel,youÕrepresentedwithaSyntheticCue(acuedesignedbyFacebookthattheycontrol),andyouperformthedes

70 iredtargetAction.Then,whenyouopentheapp,
iredtargetAction.Then,whenyouopentheapp,therearenotiÞcationswaitingforyou.And,maybe,oneofthenotiÞcationsisdelightful!Here,FacebookemployedReinforcementLearningtocarefullyinduce AtBoundlessMin

71 d,weexploredthisconceptwithourapp,Space.
d,weexploredthisconceptwithourapp,Space.WewereinterestedinhelpingpeopleregaincontroloftherelationshipstheyhavewithappsbychangingtheimpactoftheCARModel.StimulusDevaluationoperatesbyincreasingt

72 hetimedelaybetweenanactionandthereinforc
hetimedelaybetweenanactionandthereinforcementthatauserreceivesforthatinteraction.AsdiscussedinChapters2 rewardingstimuli,whiledecreasingthechancethattherewardcan ,theleadingcausesofdeathwerei

73 nfectiousdiseases.Men,women,andchildrens
nfectiousdiseases.Men,women,andchildrensufferedTyphoid,Pneumonia,Fever,Tuberculosis,andInßuenza.Wediedofpathogens70inourwaterandfoodandurbanairfromwhichwe70forwhichwebarelyhadnamesyethadlittl

74 echoicebuttosuccumb.Andthensomethingbrav
echoicebuttosuccumb.Andthensomethingbravehappened.WetookseriouslytodevelopingarigorousTechnologyoftheBody.Tomodernmedicine,sanitization,andvaccination.Wechallengedourpreviouslyheldnormsofthes

75 anctityofbody,andmenandwomenworldwidein
anctityofbody,andmenandwomenworldwidein BJFogg.PersuasiveTechnology:UsingComputerstoChangeWhatWeThinkandDo.MorganKaufmann,2002.BillGates.Gatesnotes.URLhttps://www.gatesnotes.com/.JohnGeake.Ne