Profit from the Learning Curve - Harvard Business Review

Both define a declining trend. The learning curve is the line that fits the points; and the boundaries of the band are the upper and lower limits within which ... Youhave1freearticlesleftthismonth. Youarereadingyourlastfreearticleforthismonth. Subscribeforunlimitedaccess. Createanaccounttoread2more. Operationsandsupplychainmanagement ProfitfromtheLearningCurve by WinfredB.Hirschmann by WinfredB.Hirschmann FromtheMagazine(January1964) Tweet Post Share Save Print Tweet Post Share Save Print Leerenespañol Leremportuguês Practicemakesperfect.Athingcanalwaysbedonebetternotonlythesecondtimebuteachsucceedingtimebytrying.Thiseverybodyknows.Buthowmanyknowthatthepatternofimprovementcanbesufficientlyregulartobepredictive?Howmanyrealizethatsuchpatternscancharacterize,notonlyindividualperformance,butalsothecompositeperformanceofmanyindividualsorganizedtoaccomplishacommontask? Theindustriallearningcurvequantifiessuchperformance.Ithasevolvedfromexperienceinairframemanufacture,whichfoundthatthenumberofman-hoursspentinbuildingaplanedeclinedataregularrateoverawiderangeofproduction.Suchcontinuingimprovementwassocommonintheaircraftindustrythatitbecamethenormalexpectationinthewartimemassproductionofaircraft;thus,productionandothertypesofperformancewerecustomarilyscheduledonsomebasisofprogressivebetterment. Butthisisnotthepracticeinindustrygenerally.Althoughlearningcurveshavebeenrecognizedinindustriesotherthanaircraft,theyhavenotbeenaswidelyaccepted.Instead,predictionsareusuallybasedonassumptionsoflevelperformanceandconstantcosts. InthisarticleIwillarguethatthispracticeisnotonlyincorrectbutcostly.Peopledolearn,andtheylearnaccordingtoagenerallypredictablepattern.Thelearningcurve,Ibelieve,isanunderlyingnaturalcharacteristicoforganizedactivity,justasthebell-shapedcurveisanaccuratedepictionofnormal,randomdistributionofanything,fromhumanI.Q.’stothesizeoftomatoes.Whereverpeoplestrivetodobetter,improvementsresult;otherwise,howwouldprogresstakeplace? Byfailingtocapitalizeonthisnaturalphenomenon,managerswillnotencouragecontinuedeffortsoncetheybecomeconvincedthat“furtherimprovementsarenotpossible.”Furtherimprovementsarealwayspossibleovertime,solongaspeopleareencouraged,orevenordered,toseekthem.Thus,anunderstandingofthelearningcurvebecomesofcrucialimportancetothebusinessmanager. CharacteristicCurve BeforeIdemonstrateitsapplication,however,itisnecessarytoclarifyjustwhatImeanbythelearningcurve. Learningpatternswerereportedlyfirstobservedformanufacturingoperationsin1925bythecommanderoftheWright-PattersonAirForceBaseinOhio.1Duringsubsequentyears,definitivestudiesofaircraftassemblyshowedthefollowingpattern:thefourthplanerequiredonly80%asmuchdirectlaborasthesecond;theeighthplane,only80%asmuchasthefourth;theonehundredth,only80%asmuchasthefiftieth;andsoon.Thus,therateoflearningtoassembleaircraftwasconcludedtobe80%betweendoubledquantities.Onanarithmeticchart,withlinearcoordinates,therelationshipisacurve,showingarapidinitialdeclinethatlatertrailsoff(seeExhibitI-A).Onadoublelogarithmicchart,however,itisastraightdecliningline,whichreflectsaconstantrateofreduction(seeExhibitI-B).Suchastraight-linerelationshipiseasiertodrawanduseforpredictionpurposes. ExhibitI.The80%LearningCurve Whilethelearningcurveisauniversalphenomenon,ithasmanyvariationsinform;forexample,therearegreatvariationsinthelevelatwhichacurvestarts(i.e.,thecostofthefirstunit).Thisissimplybecauseofthedifferentrangesofcomplexityofitems.Nevertheless,theslopeofthecurveiscommontoawidevarietyofexperience.Infact,itwastheregularfindingofthecommonslopeofabout80%forfighter,bomber,andtransportplanesthatstartedspeculationaboutageneraltheoryoflearningcurves.2Furtherinvestigationshowedthat,althoughoperationshavingessentiallythesameproportionsoflaborcontenthavefairlycommonslopes,otheroperationsdifferincharacteristicsandcorrespondingslopesofcurve. Thus,operationspacedbypeoplehavesteeperslopesthanthosepacedbymachines.Forexample: Inairframemanufacturethree-fourthsofthedirectlaborinputisassembly;thebalanceisrepresentedbymenengagedinmachinework.Insuchalargelyman-pacedoperation,an80%curveiscommonlyfound. Butwhentheproportionofassemblyworkislower,thedownwardslopeofthecurveisnotsosteep.Iftheratioofassemblytomachineworkis50/50,theslopeisabout85%.Iftheratioisone-fourthassemblyandthree-fourthsmachinework,theoperationislargelymachine-paced,andtheslopeisaround90%. Theseresultsmightbeexpectedsincelearningisrelatedtopeople—thefewerthepeople,thelessthecapacityforlearning. Thesevariouspercentagesprovideapointofdepartureonwhichtobasepredictions.Suchpredictions,ingeneral,haveprovedmorevalidthanthosebasedonanassumptionoflevelperformance,i.e.,aconditionofnolearning. AsExhibitII,depictingtheexperienceofseveralaircraftcompanies,shows,itwouldbepossibletodrawalinetofitthesuccessivepointsofaplot;theywouldnotfalldirectlyonit,butmostofthemcouldbecoveredbyabandcenteredontheline.Bothdefineadecliningtrend.Thelearningcurveisthelinethatfitsthepoints;andtheboundariesofthebandaretheupperandlowerlimitswithinwhichrandomvariationscausethepointstofall. ExhibitII.DataIllustratingNormalLearningCurveandSomeDeviationsSource:MiguelA.Reguero,“AnEconomicStudyoftheMilitaryAirframeIndustry,”Wright-PattersonAirForceBase,Ohio,DepartmentoftheAirForce,October1957,pp.231–235. Wheneverpointsfalloutsidetheselimits,thereisusuallyanassignablecause,justasthereisforanyotheroperationwhich“getsoutofcontrol.”Forexample,ExhibitIIindicates: A“leveling-off,”orevena“toe-up,”canoccur,suchasattheendofacontractwhenworkersaretransferredtootherproductionlinesandclosing-outoperationsbecomeinefficient. Ariseinthecurvecanoccurinthemiddleofacontracttoo,owingtoasubstantialinterruption(suchasthatcausedbyintroducingchangesinamodel,bymovingoperationstoanewbuilding,orbyhaltingoperationsforawhilesothatforgettingoccurs).Shortlyafteroperationsrecommenceandskillinhandlingchangesisacquired,thecurvedeclinesrapidlytoapproachtheoldslope.Suchabreakinthecurveoccursfrequentlyenoughtohaveacquiredthedescriptiveterm“scallop.”Infact,if,insteadofmerelyachangebeingmade,anewmodelisintroduced,oranewtypeofitemisputintoproduction,thescallopoccursinitiallyandthecurveessentiallystartsagain.Thus,thedirectlaborinputrevertsbacktowhatithadbeenwhenthefirstitemoftheprecedingtypewasputintoproduction(assumingthatthetwoitemswereofsimilartypeandconfiguration). Suchlearningcurveperformancewassocommoninaircraftassemblythatitbecameexpectedasamatterofcourse.Anditisstillusedconsistentlytodayasabasisforprojectingcosts,forecastingmanpowerrequirements,ratingpersonnel,schedulingproduction,andnegotiatingmultimilliondollarcontracts. BarrierstoAcceptance Whilelearningcurveshavebeenreportedforoperationsinotherindustries(particularlybypersonnelwhohavemigratedfromaircraftcompanies),theiroccurrenceseemstobeconsideredtheexceptionratherthantherule.Whyhasatechniqueusedsolongandvalidatedinoneindustrynotbeenadoptedincommercegenerally?Herearesomepossiblereasons: Thereisalackofawarenessthatimprovementpatternscanbereasonablywellquantified.Althougheveryonehashadtheexperienceofimprovinghisskillbyrepetition,thisexperienceisgenerallyassessedqualitatively.Improvementscanseemirregularandfluctuatingatthetimetheyoccur.Consequently,itisnotobviousthat,overthelongterm,thesewobblingimprovementscanbetracingadefinabletrend.Studiesofestimatesbasedonempiricaltrendsintheaircraftindustryindicatethatimprovementcurvesactuallyapproachareliabilitycomparabletothatexperiencedwithengineeringconstructionestimates. Skepticismthatimprovementcancontinuemaybeanotherfactorthathaslimitedmoregeneralacceptanceofthistechnique.Afterarduoushoursormonthsofefforttoachieveanincrementofadvance,itisnaturaltofeelthatthelastounceofbettermenthasbeenwrungoutofanoperation.Asaresulttheexpectationofstillfurtherimprovementseemsquiteunrealistic. Manycompaniesbelieve,“Ourbusinessisdifferent,”andconsequentlythatsuchcurvesdonotapplytotheiroperations.Credenceisgiventhisconclusionbytheobservationsofsomeexperiencedpractitionerswhoexpectthelearningcurvetobeinapplicableorhavelittlevalueinsuchindustriesasbasicchemicals,plastics,andpetroleumrefining.3Ofcourse,eachbusinessisdifferentfromothers.Nevertheless,sinceitisuniversalexperiencethatoverthelongrunamethodcanalwaysbeimprovedinsucceedingtimesbytrying,thereisabasisforbelievingthatprogressivelyimprovingperformancecanbeuniversallybroughtaboutbyeffectivetrying. Learningcurveswhicharealreadyoccurringmaynotberecognizedbymanypeople.Withthe80%learningcurvecharacterizingthemixofthree-fourthsassemblyandone-fourthmachineworkcommonlyfoundinairframemanufacture,improvementwassorapidthatthedirectlaborhoursforthetenthplanemanufacturedwerelessthanhalfthoserequiredforthefirstone.Suchamarkeddeclinevirtuallyforcedattentiontoit.Nevertheless,11yearspassedbetween1925,whenlearningcurveperformancewasfirstnotedinaircraft,and1936,whenitwaspubliclyreportedbythemanageroftheBuffaloplantofCurtiss-WrightCorporation.4Consequently,otherindustrieswithmuchsmallerratiosofassemblytomachineworkcanbeexpectedtobelesssusceptibletoimprovement,andthereforetohavecurveswithlessslope.Thesmalleramountofimprovementcouldbeobscuredbylargerforces,orbeoffsetbysuchopposingeffectsasinflation.Improvementmightalsoescapenoticebyoccurringoveralongertime,orbeattributedtoothercauses. Finally,theremayalsobealackofawarenessthatthelearningcurvecandescribegroupaswellasindividualperformance,andthatanygroupcancomprisenotonlydirectlaborbutalsobehind-the-scenespeoplewhodeliberatelyseekimprovementsintheprocess.Inairframemanufacture,forinstance,suchgroupsmayincludetoolengineerswhocontrivenewjigsandfixtures.Inothertypesofmanufacture,theremaybestaffgroupswhichworkondifferentcombinationsofoperatingconditionstoimproveyieldsordeviseinstrumentstoimprovecontrol. Thelastpointisparticularlyimportant.Yetmostmanagementshavefailedtorecognizethattechnologicalprogressisakindoflearning.Assigningspecialiststoseektechnicalimprovementsandtoincorporatetheminoperationsobviouslyhelpsbringaboutimprovement.Theindustriallearningcurvethusembracesmorethantheincreasingskillofanindividualbyrepetitionofasimpleoperation.Instead,itdescribesamorecomplexorganism—thecollectiveeffortsofmanypeople,someinlineandothersinstaffpositions,butallaimingtoaccomplishacommontaskprogressivelymoreefficiently. Thisbroaderconceptmaybethereasonwhythephenomenonhasmanynames:“manufacturingprogressfunction,”“cost-quantityrelationship,”“costcurve,”“experiencecurve,”“efficiencycurve,”“productionaccelerationcurve,”“improvementcurve,”and“performancecurve.”Suchtermsaremorecommonlyusedinwritingthaninconversation,wheretheterm“learningcurve”predominates.Itseemspreferabletoretainthegeneralname,buttorememberthatitcanbebrokendownintomeaningfultermswhicharemorespecific. TestingApplicability Iflearningcurveperformanceisanaturalcharacteristic,thensuchperformanceshouldbefoundnotonlyformoretypesofactivitiesalreadyrecognizedasresponsive,butalsoforunlikelyoperationssuchasthosenotpreviouslyreportedorbelievedsusceptible.Petroleumrefiningoffersagoodexampleofthetypeofindustrytowhichthelearningcurvemightbethoughttobeinapplicable.Itischaracterizedbylargeinvestmentsinheavyequipment,andissohighlyautomatedthatlearningisthoughttobeeithernonexistentortoosmalltobeofvalue.Letusseehowtruethisbeliefis. Inprocessingunits Petroleumrefiningcomprisessuchprocessoperationsasdistillation,cracking,andreforming.Fromtimetotime,scatteredobservationshaveappearedintheliteraturetotheeffectthatthecapacityofsomeoftheunitsperformingtheseoperationsislargerthantheirdesigncapacity.In1951,theworldwide,installeddesigncapacityoffluidcatalyticcrackingunitswas1,200,000barrelsperstreamday;however,theactualthroughputwasaboutone-thirdgreater—asmuchas1,600,000barrelsperstreamday!Thisaggregateiscomposedofthethroughputsoftheindividualcrackingunitsatapointintime.Itdoesnotshowhowtheychangedovertime. Butbyplottingtheperformanceofindividualunitsatapointintimeagainsttheirageatthattime,weobtainacluetothispattern;seeExhibitIII.Thepointsaretheratiosoftheachievedcapacitytothedesigncapacityasdeterminedfrompublishedtabulations.TheselectedpointsofExhibitIIIwerecalculatedfromthesedata.Forexample,thefirstunit(A)atthetimeofthisresearch—1958—wasoneandone-halfyearsoldandbythattimehadachievedapproximately116%ofdesigncapacity.Thesecondunit(B)atthetimeofthisresearchwasfouryearsoldandhadachievedabout125%ofdesigncapacity.Ingeneral,astheolderunitsshow(C,D,E,F,G,andH),performancerapidlyimprovedinthefirstfewyears,andcontinuedataslowerrateinlateryears. ExhibitIII.CapacityAchievedbyDifferentFluidCatalyticCrackingUnitsofVariousAges(in1958) ExhibitIVshowsthatsuccessiveannualpointsforanindividualcrackingunitindicatethatgrowthoccursinastep-wisefashion,sothatthepointsarescatteredinabandinsteadoflyingonasmoothcurve.Thepatternofimprovementindicatedbythecoloredline,whichisthesamecurveastheoneinExhibitIII,resemblestheinverseofalearningcurveonanarithmeticchart.Iftheparametersarechangedsothatthenumberofdaystoprocess100,000barrelsisplottedagainstcumulatedthroughputonalogarithmicchart,adecliningstraightlinecanbedrawnthroughthepoints(seeExhibitV).Thelinehasa“slope”ofabout90%,asmightbeexpectedfromamachine-pacedoperationwhichinvolvescomparativelylittledirectlabor. ExhibitIV.PerformanceofanIndividualFluidCatalyticCrackingUnit ExhibitV.LearningCurve—PlotofPerformanceCorrespondingtoExhibitIV Whataccountsforthisimprovementinperformance?Safetymarginsforcriticalequipmentareincludedduringdesignstagesofaprojecttoensuregettingrequired(design)performance.Thus,actualperformancecanandshouldbehigherthanthedesigntarget.Operatorswillsoonlearntotakeadvantageofbuilt-insafetymargins.However,equipmentnotconsideredcriticalindesignandwithoutextrasafetymarginsmaylimitinitialperformancetothedesigntarget.Obviously,removalofsuchabottleneckcanresultinmarkedimprovement.Butastimepasses,fewerandfewerbottlenecksremaintobeuncovered,soprogressslows. Thesecircumstancesexplainarelativelyrapidearlygrowthandasubsequentgradualslowingdownasmoreandmorecapitalandingenuityarerequiredtoeliminatefurtherbottlenecks.Inthisindustrywithitsheavydependenceonmachinery,theimprovementcurveappearstoreflecttechnologicalresourcefulness.Soitseemsreasonabletobelievethatthistechnicalskillwillcontinuetoresultinsuchenhancementpatternssolongasincreaseddemandorotherincentivesoccurtoprodthesearchforimprovement,andsolongasitisbackedupbythepresentlevelofresearchandengineeringefforts. Instart-upoperations Butiflearningisgreatestwherethemostpeopleareinvolved,thelearningcurveshouldberevealedmostclearlyinrefineryoperationswhichhaveahighlaborcontent.Onesuchcircumstancewouldbethestart-upofunitsaftershutdownperiodsforrepairs.Specialistsareonhandtoassureproperfunctioningofinstruments.Extracraftpeopleareassignedtohandleemergencies.Moresupervisorsarepresenttogiveguidance.Theregularcrewisparticularlybusyinroutingflows,openingandclosingvalves,liningoutsystemcomponents,andoperatingmanualcontrolsbeforetheautomaticonesarecutin.Shiftworkersmayattimes“doubleover.” Thisrelativelyhighhumanactivitycontentofstart-upssuggeststhatthepeopleinvolvedshouldbesusceptibletoasignificantdegreeoflearning.Andactualexperienceconfirmsthis.OveraperiodoftenyearsthetimenecessarytoputaWhitingrefineryfluidcrackingunitoftheAmericanOilCompanyonstreamdroppedtolessthanhalfthetimeinitiallyrequired. Inmaintenance Inviewoftherepetitivenatureofmuchmaintenanceworkandthecontinuingeffortsmadetoorganizeforbetterefficiency,theperformanceofamaintenancedepartmentasawholemightbeexpectedtoshowprogressivelearning.And,asExhibitVIshows,thisexpectationissound.Thepointsintheexhibitshowadecliningtrendforproductivelaborman-hoursonmaintenanceandshutdownsduringtheyears1949–1956.Attheendofthatperiodtheplotseemstobelevelingoff.Ifthemanagementhadspeculatedthenaboutthiscurve,itmighthavefeltthatithadreachedaplateau,thatmaintenancehadlearnedhowbesttodothejobsrequiredsothatafurtherdeclinewouldnotoccur,atleastforawhile.Butactually,asthe1958point(plottedasacross)shows,thetrendcontinued,endingupaboutwhereitshouldhavebeenexpected. ExhibitVI.ProductiveRefineryLaborMan-HoursforMaintenanceandShutdowns Thisisnotdifficulttoexplain.Departmentperformanceisacompositeofmanyindividualtasks.Andifitisareflectionoflearning,thenindividualmaintenanceoperationsshouldfollowestablishedlearningcurves.Thisexpectationisconfirmednotonlyinrefiningbutalsoinotherlargemanufacturingplants.ExhibitVIIdepictsarecordoftheperformanceofarepetitivemaintenanceoperationbyGeneralElectricCompanyatitsRichmond-WashingtonAECplant.Itshowsthatthetimerequiredtoreplaceagroupofpartsduringshutdownsdeclinedalonga76%curve. ExhibitVII.MaintenanceLearningCurveataGeneralElectricPlantSource:CarlA.Bennett,ApplicationofaLearningCurvetoaMaintenanceProblem,SecondAnnualQualityControlSymposiumoftheDallas-FortWorthSection,ASQC,March16,1957. Inconstruction Buildingnewitemsofheavyequipmentalsoappearstobecharacterizedbylearning.The245,000barrelrefineryatFawley,England,whichwentonstreamintheearly1950’s,couldhavebeenduplicated5yearslaterat70%oftheoriginalcost.Thisdecreaseincostrepresentsarateofdeclineofabout7%peryear.Theper-barrelinvestmentcostsofunitsforsomeindividualprocesses(thermalcracking,polymerization,catalyticcracking,andcatalyticreforming)alsodeclineprogressively.Inthecaseofthermalcracking,thedeclinecontinuedfor33years. Alearningcurvemeasureoftherateofdeclineforfluidcatalyticcrackingunitsissuggestedbytheobservationthatthesteelrequiredandtheinvestmentcostin1955wereestimatedtobeone-thirdofthoserequiredtoduplicatethecapacityoftheoriginaldownflowfluidplant,whichwasbuiltin1942.Duringtheinterveningyears,about3millionbarrelsoffluidcrackingcapacitywerebuilt.When,asinExhibitVIII,thesetwocostpointsareplottedonalogarithmicscaleagainstthecumulativeinstalleddesigncapacityoffluidcatalyticcrackingplantsandconnectedbyasolidstraightline,theslopeofthatdecliningline(inblack)turnsouttobeabout80%(inconstantdollars). ExhibitVIII.LearninginConstructionofFluidCatalyticCrackingUnits Notethatpricerisescandistortthepicture.Ifthe“Nelson’sRefineryConstructionCostIndex”forpetroleumrefiningequipmentisapplicable,theactualdollarcostin1955ofbuildinga1942unitwouldhavebeen2.2timesasmuchasin1942.5Aftertakingintoaccountthetechnologywhichenabledittobeduplicatedin1955forone-thirdthiscost,wearriveataunitcostin1955dollarsof2.2×33.3%or73%ofthe1942dollarcost.Onthisbasis,thecostreductionovera13-yearperiodcorrespondstoaslopeof94%,andisrepresentedbytheblacklineonExhibitVIII.Thefactremainsthatthetruetechnologicalbenefitismeasuredbythe80%curve(thecoloredline.) Itmightseemsurprisingtofindthattheconstructionofmultimilliondollarprocessingunits,whicharecomparativelyfewandbuilttoorder,ischaracterizedbythesame80%slopecommonlyfoundinmanufacturingmanydifferentitemsinlargenumbersonproductionlines.However,sincesuchconstructionislargelyassemblywork,thisfindingisconsistentwiththegenerallearningcurveexperience:thatis,operationswithsimilarratiosofassemblytomachineworkhavesimilarlearningcurveslopes. Thisdeclinereflectslearningbyconstructionpeopleinhowtobuildsoastoreduceunitcosts,andbyresearch,engineering,andoperatingpeopleinwhattobuild.Itprovidesacluetotherelativecontributionsofcapitalandoftechnologytolearningcurveperformance.Thefactthataplantbuilttoduplicatetheperformanceofanoriginalonecouldbebuiltwithone-thirdthesteelindicatesthatthewhat-to-buildcontributionsoftheprogressivelyimprovedtechnologyembodiedinsuccessiveplantsgreatlyoutweighsthehow-to-buildcontributionsofcapitalforthebettertoolsandequipmentemployedinbuildingthem. Thegreatextenttowhichtechnologycanbedominantoverinvestmenthasbeenshownbythreerecentstudies.Thesecoversomewhatdifferenttimeperiods,buteachindependentlyconcludesthatabout90%oftheU.S.growthinoutputperman-hourhasbeenduetotechnologicalchange,andonlyabout10%toincreasedinvestmentincapitalequipment.6 Technologyalsocontributestobuildinglargerplantswithattendanteconomies,becauseconstructioncostsdonotincreaseproportionatelywithsize.Doublingthecapacityofaplantdoesnotnecessarilydoublethecost.Rather,itincreasesitbysomelesseramount,whichcanberepresentedbyanexponentofthesize.Aplanttwiceaslargemaycostabout20.7or1.6timesasmuchasthebasesize.Ifthis0.7costexponentappliestofluidcrackingplants,thentheper-barrelinvestmentcostsofvariousmultiplesofsizedecline,asshownbythecolumnofcrossesontherightsideofExhibitIX.Sinceplantshavebeenbuilttwoormoretimeslargerthanoriginallyplanned,theeconomiesofsizefromprogressinresearchandengineeringalsocontributedtooffsettingtheeffectsofinflation. ExhibitIX.LearninginConstructionofVariousSizeFluidCatalyticCrackingUnits Inthisinstance,technologicalprogressdiddecreasecostsmorethaninflationincreasedthem.Thiscircumstancemightsuggestthatdepreciationallowanceshavebeenmorethanadequatetoprovidethecapitalnecessarytoreplaceunitswhentheyareretired.Actually,rapidobsolescenceisconcomitantwithrapidtechnologicalprogress.Acompetingfirmwithanew,low-costunithastheadvantageofasmallercapitalchargeforitsproductsandcantherebyprofitablysellthematlowerprices.Tosurviveinthefaceofsuchcompetition,companiesmaybeforcedtoreplaceexistingunitsbeforetheyarefullyamortizedandbeforedepreciationreserveshavebecomelargeenoughtopayforthereplacements.Andthisisparticularlytrueforunitsdepreciatedbythestraight-lineandothermethodsrequiredforequipmentinstalledbefore1954. Industryperformance Anindustryisanaggregateofcomponents.Wecanreasonthatiflearningincomponentsiswidespread,itshouldbereflectedinaggregateperformance.Alogarithmicplot(inExhibitX)ofman-hoursperbarrelversuscumulatedbarrelsofcrudeoilrefinedintheUnitedStatessince1860resultsinthefairlyregulartypeofdeclinesuchreasoningwouldleadustoexpect. ExhibitX.Man-hoursperBarrelRefinedinthePetroleumIndustry Otherindustriesshowsimilardeclines,asillustratedbyExhibitXIfortheU.S.electricpowerindustryandbyExhibitXIIfortheU.S.basicsteelindustry. ExhibitXI.U.S.ElectricPowerIndustry—SuggestsLearningCurveDeclineSource:JohnE.Ullman,“EconomicsofNuclearPower,”Science,April4,1958,p.140. ExhibitXII.Man-HoursperUnitofOutputinU.S.BasicSteelIndustry—SuggestLearningCurveDecline1935–1955Source:Bulletin1200,Washington,U.S.DepartmentofLaborStatistics,September1956,andothersources. Suchincreasesinaggregateproductivityreflectthejointeffectofmanyinterrelatedinfluences.Amongthemaretechnologicaladvance,increasedcapitalinvestment,bettermethodsofmanagement,increasedhealthandeducationofworkers,andimprovedcommunications.Theirover-allaim,however,isprogress,andsuchdeclinescanberegardedastheresultoflearninghowtodothingsbetter. Sinceconstruction,maintenance,processing,andstart-upoperationsarecommoninindustrygenerally,andsincelearningcurvescancharacterizesuchactivitiesinthepetroleumindustry(wheretheywereneitherpreviouslyreportednorexpectedtobesignificant),itseemsreasonabletobelievethatsimilarcomponentactivitiescanfollowlearningcurvepatternsinotherareasalso,andmayinfactalreadybecontributingtotheprogressoccurring.Thecredibilityofthisgeneralizationisreinforcedbyunderstandingtheelementsoflearningandthepracticeswhichpromoteit. ImportantImplications Whatdoesallthisseemtoaddupto?Essentiallythis.Therearetwomainfactorswhichaffectlearning:(1)theinherentsusceptibilityofanoperationtoimprovement,and(2)thedegreetowhichthatsusceptibilityisexploited.Indetail: 1.Inherentsusceptibilityisrelatedtothehumancontentofanoperation.Itisreflectedintheratioofassemblytomachinework.Thegreaterthehumancontent,thegreaterthesusceptibilityforimprovement. 2.Thedegreetowhichsusceptibilityisexploitedisrelatedtothedynamiccontentoftheenvironment—thedriveandresourcefulnessofmanagementanditsskillinstimulatingsupervisorsandtechnicalpeopletobecreativeandworkerstobeproductive.Thegreaterthedynamiccontentoftheenvironment,thegreatertheexploitationofinherentsusceptibility. Effectoffaith Oneofthefactorsaffectingthedynamiccontentoftheenvironmentisfaith.Ifprogressisbelievedpossible,itwilllikelybesought;andifitislookedfor,thereissomepossibilityoffindingit.Conversely,ifimprovementsareconsideredunlikely,therewillbelittleurgetoseekthem.Adefeatistphilosophycanbeengenderedwhichsodebilitatesaneffortthatithelpstoproducetheveryconditionitassumes.Industrialengineershavelongknownthatonceaquantitativeobjectiveisimposedonanorganization,therearestrongforcescreatedtofittheobjective.Thereareanumberofstudiesshowingthatprogressdidceasewhentheactualunittimereachedtheoriginallyestimatedtimeforthejob. Consider,forinstance,theexperienceofacompanywitha“cost-plus”subcontractfromanaircraftcompany: Thiscompanyexpectedthatcostswouldbeginatahighlevel,butwoulddropduringaninitialperiodofacquiringfamiliarity,andthenleveloff.Thisiswhatactuallyoccurred,andthesubcontractorwassatisfied. Theaircraftcompany,however,pressedforcontinuedreductioninunitcosts.Toitspeople,learningcurveperformancewassuchacommonphenomenonthattheydidnotinitiallyexplainittothesubcontractorasthebasisoftheirexpectation.Whenunitcostsceasedtodecline,however,lowerechelonsoftheaircraftcompanytalkedwithlowerechelonsofthesubcontractor,whoprovedtobeunfamiliarwithlearningcurveexperienceanddidnotbelieveit.Whenlackofprogresscontinued,theuppermanagementoftheaircraftcompanytalkedtoitscounterpartinthesubcontractingfirm,andfoundthatlevelequallyunfamiliarandunabletounderstandwhythereducedcostsalreadyachievedwerenotsatisfactory.Finally,theaircraftcompanyinsistedthatunitcostscouldandmustdecline,andthatifthesubcontractorcompanydidnotknowhowtolowerthem,hewouldbeshown.Ateamofaircraftpeopleexplainedthelearningcurveandshowedhowtogettheresultswhichproducecontinuedprogress.Thereafter,thedeclineoccurredalongthelearningcurveoriginallyexpectedbytheprimecontractor. Open-endedexpectations Thelimitingeffectof“ceilingpsychology”alsoshowsupintheexpectedperformanceofnewequipment.Ifitisdesignedtooperateatagivenrateandfailstoperformasexpected,greateffortswillbemadetobringituptotarget.Itistakenforgrantedthatratedorspecifiedperformancecanbeachieved.Butoncethatlevelisreached,expectationsareprovedjustified,andattentionisdirectedelsewhereonthepresumptionthatratedoutputisthelimitofcapacity. Suchwerethecircumstancesinthefollowingexample: Anewmachinehadagivenratedcapacity,butinitialproductionwaslessthanhalfthatcapacity.Someofficialsinthecompanyhadopposedtheselectionofthisequipmentonthegroundthattheydidnotbelievethequalityandquantitywouldbeacceptable.Itsinitialperformancesupportedtheirbelief,andtheyrecommendedthatthemachinebereplacedwiththeonetheyhadfavored. However,thosewhohadselectedtheoriginalonewerestillconvinceditcouldperformadequately,andaskedformoretime.Thefirm’smanagementengineeringconsultantsupportedthem;hehadnotedtheoccurrenceofsomeimprovementalongalinethatsuggested“learning.”Themachinewasretainedforfurthertrial,andoutputgraduallyrosetotheratedcapacityinaboutfourweeks. With“normal”outputachieved,nofurtherimprovementwasexpected,andconcertedeffortswereabouttocease.Buttheconsultantnotedthattheimprovementpatternhadcontinuedtotracealearningcurve.Hereasonedthatprogressmightcontinue,andthereforeengineerswerekeptassignedtothemachine.Theresultwasafurtherincreasetotheleveloftripletheratedcapacityinfourmonths. Thus,notputtingaceilingonexpectationsmaypermitimprovementstocontinue.Itmayalsoresultinafasterrateofimprovement.Somecompaniesreportgettingmorerapidprogresswhentheoperatingforceisnotinformedofthetargetrate. Conversely,ifbettermentisnotbelievedpossible,thentheincentivetoseekimprovementisreduced,andanatmosphereofmaintainingthestatusquoisencouraged.Assemblyworkersgettofrownonincreasesasrate-breaking,andengineerswhohaveideasforpossibleimprovementmayhesitatetopushthemiftheywillbeinterpretedas“rockingtheboat,”oriffailureriskscensure. Ifthereisalimittolearning,itcouldbeexpectedtobereachedafterquiteaperiodoftime,duringwhichskillandtechnologywouldhaveimprovedperformancetotheultimatelevel.Thisiswhatseemstooccurfrequentlyinanestablishedplant.InaCornellUniversitystudy,however,twoseparateitemswhichwerethoughttohavereachedanachievementplateauwerefoundtoshowcontinuedlearningwhentransferredtoanotherfirm.7 Thereisevidencethatimprovementcanpersistovermanyyears,andonmanymillionsofitems.Inaman-pacedoperationinvolvingtheassemblyofcandyboxes,forinstance,thelearningcurvewasfoundtohavepersistedforthepreceding16yearsduringwhich16millionboxeswereassembledbyoneperson.8Inamachine-pacedoperation,progressalsohascontinuedovertheproductionoftensofmillionsofunits. Althoughthepercentageofimprovementinsuchcasesbecomessmallafteralongperiod,thebenefitcanstillbesubstantialifthenumberofunitsinvolvedislarge. Progressbyserendipity Theassumptionthatcontinualprogresscanoccurmayalsocreateanatmospherewhichencouragesuncoveringideasorrecognizingthemwhenstumbledupon.Hereisanexampleofthispossibility: Inanautomaticoperationforforgingwrenches,apieceofhotmetalisforgedintothepropershapeinadie.Withrepeateduse,thedie“wears”orbecomeslarger,sothateventuallythedimensionsofthewrenchexceedspecificationsandthediehastobereplaced.Experienceindicatestheminimumnumberofforgingsthatcanbemadebeforereplacementisrequired,andlikewisethemaximumnumber. Ontheeveningshiftatonecompany,dimensionsoftheforgingsbegantovaryerraticallyandthusindicatedtheneedfordiereplacementmuchsoonerthanusual.Investigationdisclosedthatanewmanwasonthejob,andthathewastinkeringwiththefurnacetemperature.Whenthetinkeringwasstopped,thediecontinuedtoperformsatisfactorily.Themattercouldhaveendedtherewiththecustomaryreprimandtotheemployeetofollowinstructions. However,theinvestigatornotedthattherehadbeensomegoodaswellasbadresults.Instudyingthematter,hefoundthatthesteelusedcouldbeforgedattemperaturesmuchhigherthanthosespecifiedwithoutburningordestroyingthepropertiesofthemetal.Newdiesweresubsequentlymadesmaller,andthemetalfirstforgedattheminimumpracticaltemperature.Asthediewore(gotlarger),theforgingtemperatureswereincreased.Thehotterforgingsshrankmoreoncooling,andinthiswaycontinuedtofallwithintolerance.Thispracticepermittedadietoproducetriplethenormaloutput. Thus,evenwithautomaticoperations,theremaybetransientperiodswhenperformanceisbetterthannormal.Ifthesesituationscanberecognizedandstudiedwiththesamezealandconvictionasisappliedtobelow-normalperformance,itmaybepossibletomaketheimprovementpersist.Anoptimisticresearchclimatemayinthiswayencourageprogressbyserendipity—makinghappyandunexpecteddiscoveriesbyaccident. Wishingnotenough Thegeneralknowledgeofthepervasivenessoflearningandtheactualexperienceofhavinghaditoccur,sometimesinthemostunexpectedplaces,shouldleadtotheconvictionthatitcan,will,andmustoccur.Yetevencompaniesthathavehadmuchexperiencewithlearningreportoccasionswhenlearninghasfailedtocomeaboutaspredicted.Why?Theexplanationisthatmerelyexpectingprogressisnotenoughtobringitabout. Oneaircraftfirmstatescategoricallythatifaforemandoesnotgettheexpectedamountofimprovement,heisfired.Buttheforemandoesnotcontrolallhisenvironment.Ifheisprovidedwithimpropertools,aninsufficientbudget,ornotenoughstaffassistance—inotherwords,ifthesystemisnotright—itisthemanagement,somesay,thatshouldbereplaced. Perhapsinrecognitionofthiscircumstance,thegeneralmanagerofanotherairframecompanytookdifferentactionwhencostsofagivenmodelwerenotdecliningasexpected: Aconferencewithhisstaffdevelopedtheexplanationthatthemanychangesrequiredbythegovernmentwereresponsibleforthelackofsatisfactoryprogress.Furtherinvestigationindicated,however,thatalongwitheachchangerequiredbythegovernment,theplant’sownengineeringstaffwasmakingadditionalchangestoimprovetheoperationonthegroundthat,aslongastheinterruptionwouldoccuranyway,theextrachangescouldbeincorporatedatsubstantiallynoextracostordisadvantage. Thegeneralmanager’sactionwastoorder15%ofthedesignandtoolengineerstobetransferredtoanothercontract.Eventhoughtherewerestrongproteststhatsuchactionwouldmakeitimpossibletodothenecessarywork,hestuckbyhisedict.Asaresult,theworkpiledupinfrontoftheremainingpeoplesothattheydidnothaveenoughtimeforembellishment.After30days,thenumberofextrachangesgeneratedbytheengineeringdepartmentdroppedtofourforeachchangeinitiatedbythegovernment,andthecostcurvebegantodecline.Thenextmonth,thetransferofanadditional15%oftheengineersbroughtaboutafurtherdeclineinunitcosts.Threesuchtransfersweremade,andtheextrachangesgenerateddroppedtoanaverageofabouttwoandone-halfforeachchangerequiredbythegovernment.Concurrently,theunitcostcurveestablishedadeclinealongtheexpectedpath. Justwhythenumberoftransfersstoppedatthreewasnotexplained.Perhapsthepreviousexperienceofthegeneralmanagerwhichconvincedhimthatprogresscouldbemadealsotaughthimthat,withtheoperationofhisplantandindustry,anyfurthergainswouldintroduceoffsettingpenaltiessothattherewouldbenofurthernetbenefit. Resultofneed Thiscircumstanceillustratesthatmerelyexpectingprogressdoesnotbringitabout.Itisnotordainedbyfatetoarriveonschedule,butmustbecontinuously,vigorously,andresourcefullysought.Suchdriveisusuallytheresultofneed.Inthecaseofanimals,theneedisusuallyhunger—thedesiretosurvive.Ithasbeenfoundtoleadtolearningcurvesforawholerangeoforganismcomplexity:amoebas,ants,snails,rats,andmonkeys. Thissameneedunderliesnationalandindustrialprogress.Ithasbeensaidthatcivilizationridesontheguncarriage—thatthethreattosurvivalinawarisbothsolargeandsoapparentthatmassesofpeoplestrivewitheffortstheywouldotherwisenotmaketogetprogresswhichwouldotherwisenotresult. Survivalisalsothedriveinindustry.Theproprietorofanewbusinesswillworklonghourstogethiscompanyonitsfeet,andthreatstoprofitwillspureffortsinanestablishedfirm.Ifacompanycoasts,itlosesground;andifitcontinuestoslip,itfacesbankruptcy.Thethreatofeconomicextinctionislikeanyotherthreatofdeath.Itprodsthefirmtostriveforimprovementsandfortheprogressnecessarytocontinue.Whenimprovementsmustoccurforsurvival,thequestionofwhethertheycanoccurbecomestrivial.Theproblembecomesoneofuncoveringwhathastobethere. Conceptsnotnew Theseconceptsarewellknown.Theyarereflectedinthemotto,“ProgressIsOurMostImportantProduct,”andtheslogan,“PetroleumPromotesProgress.”Theyunderliethefamiliarphrases,“necessityisthemotherofinvention”and,“practicemakesperfect.”Itistheuniversalexperienceepitomizedbysuchproverbsthatmakesitreasonabletoexpectthatanyactivityinvolvingorganiclifeissusceptibletolearningcurveimprovement. Thecontributionoflearningcurvestudieshasbeentovalidatethisexpectationbyfindinglearningcurvesforbothsimpleandcomplexorganisms,andforvariouslevelsoforganizedactivity.Thesestudieshavebroughttheconceptintosharperfocusandputitintoamoremeaningfulandusefulframeworkbyintroducingameasureofquantificationwherepreviouslytherewasonlyqualitativeexpectation. Theseresultscanbesummarizedaslearningcurvedoctrine: 1.Wherethereislife,therecanbelearning. 2.Themorecomplexthelife,thegreatertherateoflearning.Man-pacedoperationsaremoresusceptibletolearningorcangivegreaterratesofprogressthanmachine-pacedoperations. 3.Therateoflearningcanbesufficientlyregulartobepredictive.Operationscandeveloptrendswhicharecharacteristicofthemselves.Projectingsuchestablishedtrendsismorevalidthanassuminglevelperformanceornolearning. 4.Learningisrelatedtothedynamiccontextoftheenvironment.Faithandincentivestimulateprogress,andprovidethedrivetoexerttheenergy,resourcefulness,skill,andpersistenceneededtobringitabout.Conversely,“ceilingpsychology”andthetendencytomaintainthestatusquo,ornottorocktheboat,inhibitlearning. PracticalApplications Enunciatingthesefundamentalstoamanagerwhofindshehasbeenexperiencinglearninginsomeofhisoperationsmightseemakintoexpectinghimtobeastonishedatbeinginformedthathehasbeenwritingproseallhislife.Butthereismoreherethanamerelabelingofwhatishappening.Supposeweitemizesomeoftheseapplications. 1.Iflearninghasbeenoccurring,itisreasonabletobelievethatitcancontinue,ifthedynamiccontentoftheenvironmentremainsthesame.Itisprudenttoincludethispotentialofexistingoperationsforcontinuedimprovementinexpansionplansandotherlongertermforecasts.Reflectingitincapitalexpenditureplanscouldinfluencethesizeofthenextincrementofcapacity,ordefertheincrement,andtherebyenablemoreofevolvingtechnologytobeeventuallyincorporatedaswellasprovideotherearningsonthecapitalexpenditureinthemeantime. 2.Anotherapplicationliesinchoosingbetweenmodernizingexistingplantsandreplacingthem.Itseemsreasonabletoexpectmorepotentialforimprovementtobeinnewunitsofequipmentthantoremaininexistingoneswhichhavehadmostoftheirbottlenecksremoved.Thisdifferenceinpotentialisanadvantagetothereplacementalternativewhichhasgenerallybeenconsideredintangible.Itcannowbeevaluated. 3.Thereisalsoanincentiveforacompanytorefineitspracticessoastocapitalizemorefullyonthepotentialinherentinitsdailyoperations.Ifhistoricaldataareonhandforplotting,theycanprovideanexperiencebaseforpredictingfurtherimprovement.Evenifavailable—notintermsofunitlaborrequiredversuscumulativeoutput,butascostsversustimeorotherparameters—historicaldatamaystillbeusefuliftheydefineatrendbecause,onceestablished,atrendisinclinedtopersist. Sincepastrecordsreflectpastpractices,theyshouldonlybeusedasapointofdeparture.Increasedmanagementattention,moreeffort,greaterincentives,theapplicationofahighercaliberofresourcefulnessandcreativeness,orotherchangesinthedynamiccontentoftheenvironmentmayleadtoafasterrateofprogress.Iftheaimofbetterperformanceissetinaframeworkofgoalswhichcontainarealisticallowanceforbetterment,waysofachievingimprovementmaybesoughtwhichwouldotherwisenotbesought.Suchgoalscanconstitutearealchallengetoagroup’scapacitytomeetthem.Andinthiswaytheexpectationmayhelpbringabouttheimprovement.Evidencethatthisapproachcanbefruitfulisprovidedbytwoexamples.Theywerenotoriginallyreportedaslearning,buttheyreflecttheapplicationoflearningcurvedoctrine.Thus: InDuPont’s1,000-acre,1,800-manSabineRiverpetrochemicalworksinOrange,Texas,productionattheendof1955hadreachedthehighestlevelintheplant’shistory,withtheaveragefortheyearbeing25%abovethatof1954.Plantsizehadincreasedby13%overtheprecedingtwoyears.Nevertheless,inthefaceoftheincreasedmaintenanceburden,“goalssetwithdueallowanceforincreasedproductivity”helpedreduceover-allmaintenancecostsdespitehigherwages,highermaterialprices,andhigheremployeebenefits. Theratioofmaintenancecoststoplantinvestmentdroppedfrom5.9in1953,to5.0in1954,andto4.3in1955.Inthesameperiod,thenumberofpeoplerequiredforthemaintenanceoperationwasreducedbymorethan12%(126menoutofabout975);capitaltiedupinsparepartsandextramachinerywasreducedbyone-third;nosacrificewasmadeinupkeepormaintenancestandards;and,ofconsiderablesignificance,noinjuryofseriousconsequencesoccurred.9 HerefurtherevidencethatsuchcontinuedpacingsupportedbycontinuedstrivingcanleadtocontinuedimprovementisprovidedbyHenryFord’sresultswithhisModelT.Fordwrote: “Sincewehavethefirmpolicyofsteadypricereduction,thereisalwayspressure…wehaveneverconsideredanycostsasfixed.Therefore,wefirstreducethepricetoapointwherewebelievemoresaleswillresult.Thenwegoaheadandtrytomaketheprice.Wedonotbotheraboutthecosts.Thenewpriceforcesthecostsdown…Althoughonemaycalculatewhatacostis,…nooneknowswhatacostoughttobe.Oneofthewaysofdiscoveringwhatacostoughttobeistonameapricesolowastoforceeverybodyintheplacetothehighestpointofefficiency.Thelowpricemakeseveryonedigforprofits.Wemakemorediscoveriesconcerningmanufacturingandsellingbythisforcedmethodthanbyanymethodofleisurelyinvestigation…Wehavealwaysmadeaprofitatthepriceswehavefixed.”10 TheresultsofthispolicywerereflectedinthepriceoftheModelT.In1910when12,292carsweremade,thepricewas$950.By1926when15,000,000carshadbeenmanufactured,thepricewasonly$270.Inthemeantime,thewholesalepriceindexhadrisenfrom70in1910to100in1926,sothatthepriceinconstantdollarscouldbeestimatedatabout$200.Alineconnectingthesetwopointsonadoublelogarithmiccharthasaslopeofabout86%,whichisconsistentwithlearningcurveexpectationsforanoperationwhichwaslargelyassembly,especiallysincepriceisnotthecustomaryparameteronalearningcurveplot. ThefactthatproductionoftheModelTbeganin1909,morethanagenerationbeforethefirstlearningcurvearticlewaspublishedin1936,underscoresanobservationoncemadebythephilosopher,GeorgeSantayana:“Thosewhoforgetthepast…arecondemnedtorepeatit.” Conclusion Learningisapropertyofalllivingorganisms.Theycantraceimprovementpatternscharacteristicofthemselves.Sinceorganizedgroupscanbelookeduponaslivingentities,theycanbeexpectedtoexhibitlearningandtotracesuchpatterns.Intheaircraftindustry,forexample,theycommonlydo. Suchperformancedoesnotjusthappen.Itistheresultofcontinuedseekingandresourcefulstriving.Studyofanumberofoperationswhichareimportantcomponentsofmajorindustriesrevealsthattheyhavetracedimprovementpatternswithlearningcurvecharacteristics. Inaway,suchfindingsshouldnotbesurprising.Unremittingcompetitionhasprovidedacontinuingincentiveforcompaniestolookfornewandbetterwaysofdoingthings,andtheresultingprogressiveimprovementsaremerelyconsistentwiththecommonexperiencethatathingcanalwaysbedonemoreefficientlyeachsucceedingtimebytrying. Nevertheless,discoveringsuchperformanceforoperationspreviouslyconsideredunresponsivedoesprovideadditionaltangibleevidencethatlearningcanbeanunderlying,naturalcharacteristicoforganizedactivity.Itdoesnotmerelyextendthecatalogoflearningcurves.Instead,itcanhelptobreedtheconvictionthatsuchperformanceshouldbefoundelsewhere,andtherebyleadnotonlytoscrutinizingalloperationstoseewhichadditionalonesaresusceptible,buttoassumingthatalloperationshavelearningcurvepotentialandtodevisingwaysofmakingthispotentialareality.Thus,itisprudenttoreflectlearningpotentialinplansandforecasts. Themostimportantingredientsinlearningcurveperformancearevisionandleadership.Continuedimprovementisachainofinfluenceswhichstartswiththeconvictionthatprogressispossible,continueswiththecreationofanenvironmentandsupportofworkwhichpromoteit,andresultsinaflexibilityandwillingnesstochangeestablishedpracticesformoreefficientonesastheycontinuallyevolve.Furtheringthischainispartofthepracticeofmanagement.Consequently,thelearningcurvecanberegardedasaprimarytoolofmanagement. 1.MiguelA.Reguero,AnEconomicStudyoftheMilitaryAirframeIndustry,Wright-PattersonAirForceBase,Ohio,DepartmentoftheAirForce,October1957,p.213. 2.SeeFrankJ.Andress,“TheLearningCurveasaProductionTool,”HBRJanuary–February1954,p.96. 3.Ibid. 4.T.P.Wright,“FactorsAffectingtheCostofAirplanes,”JournalofAeronauticalScience,February1936,pp.122–128. 5.OilandGasJournal,publishedmonthly. 6.Foradiscussionofthesestudies;see“TheGrowthForceThatCan’tBeOverlooked,”BusinessWeek,August6,1960,p.68. 7.RichardW.ConwayandAndrewSchultz,Jr.,“TheManufacturingProgressFunction,”JournalofIndustrialEngineering,January–February1959,p.48. 8.GlenE.Ghormley,“TheLearningCurve,”WesternIndustry,September1952,p.34. 9.PhilipS.Skaff,“TheMaintenanceChallengeinaPetrochemicalsPlant,”ASMEPetroleumMechanicalEngineeringConference,Dallas,Texas,September1956,Paper56-PET-2,p.9. 10.MyLifeandWork(NewYork,Doubleday,PageandCompany,1932),pp.146–147. AversionofthisarticleappearedintheJanuary1964issueofHarvardBusinessReview. ReadmoreonOperationsandsupplychainmanagement WH Tweet Post Share Save Print ReadmoreonOperationsandsupplychainmanagement PartnerCenter Diversity Latest Magazine Ascend Topics Podcasts Video Store TheBigIdea Data&Visuals CaseSelections HBRLearning