IBMInstituteforBusinessValue|ResearchInsights

GeneratingROI

withAI

Sixcapabilitiesthatdrive

world-classresults

HowIBM

canhelp

ClientscanrealizethepotentialofAI,analytics,anddatausingIBM’sdeepindustry,functional,andtechnicalexpertise;enterprise-gradetechnologysolutions;andscience-basedresearchinnovations.

FormoreinformationaboutAIservicesfromIBMConsulting,visit

/services/artificial-intelligence

FormoreinformationaboutAIsolutionsfromIBMSoftware,visit

/Watson

FormoreinformationaboutAIinnovationsfromIBMResearch?,visit

/artificial-intelligence

2

1

Key

takeaways

Maturecapabilities

differentiatetheAIprojects

thatdeliverthehighestROI.

MostAIprojectsaren’t

profitableenough—yet.

AsAImatures,ROIonenterprise-wideinitiativesaveragedonly5.9%,wellbelowthetypical10%costofcapital.

Dataiscrucial—butonly

onepieceofthepuzzle.

Buildingontrusted,high-qualitydataatscaleboostedreturnsonAIinvestmentstoasmuchas9%.

Companiescanstrike

goldwiththerightapproach.

Best-in-classcompaniesthathavebuiltsixmaturecapabilitiesreportedaverageROIof13%onAIprojects.

2

3

AlleyesareonAI

CantheROIofAImatchthehype?

Artificialintelligenceissuddenlysexy—again.

GenerativeAIhastakenthebusinessworldbystorm,withlargelanguagemodels(LLMs)suchasOpenAI’sChatGPT,Baidu’sERNIE,Google’sLaMDAandFacebook’sLLaMAsplashedacrossthenews.Andexecutivesaren’timmunetothehype.ReferencestoAIonearningscallswereup77%year-over-yearinearly2023.1

Andthemoneyisfollowing.AIisbecominganever-largercomponentofITbudgets,withworldwidespendingonAI-centricsystemsexpectedtohit$154billionthisyear—up27%over2022.2

Butwillenterprisesspendtheseresourceswisely?AsAImodelsbecomefaster,smarter,andmorereliable,organizationsareracingtocapitalize.Canthereturnoninvestment(ROI)matchexpectations?

Theanswer:Yes—butonlyiforganizationstakeadisciplinedapproach.Tocometothisconclusion,wesurveyed2,500globalexecutivesin34businessandtechnologyrolesacross16countriesfromallmajorregions.WeaskedthemtodeconstructhowtheircompaniesareinvestinginAItoday,whatreal-worldROIisbeingproduced,andwhichelementsarerequiredtoboosteffectiveness.InpartnershipwithOxfordEconomics,wethenanalyzedwhatkeybusinessandtechnologycapabilitiesareconnectedtothemostsuccessfulAIinitiatives.(See“Studyapproachandmethodology”onpage31.)

4

6%

5%

4%

3%

2%

1%

0%

Companiesthatdevelopmature

AIcapabilitieswillgenerateprofits—

notjustmediabuzz.

OurfindingsrevealyawningoutcomegapsacrossAIprojects.Fewdeliverthefinancialvalueshareholdersexpect.Infact,averageROIonenterprise-wideinitiativesisjust5.9%,wellbelowthetypical10%costofcapital.Yet,therearedistinctimprovementsasyoumovealongtheAImaturitycontinuum—withbest-in-classcompaniesreapinganenviable13%ROI(seeFigure1).

Sowhatsetstheseworld-classperformersapart?Andhowcanleadersacrosssectorslearnfromtheirsuccess?Thisreportwilloutline:

–WhyadhocAIprojectsdeliverlessvaluethanstrategicprograms

–TheimpactoftrusteddataandthevirtuouscycleofAI-datasymbiosis

–Sixkeycapabilitiesthatdefineandenabletop-tierorganizations.

BuildingamatureAIorganization—onasolidfoundationoftrust—isrequiredtounlockAI’sfullpotential.Companiesthatgetitrightaregeneratingsignificantbusinessvalue—notjustmediabuzz.

FIGURE1

Acutabovetherest

Best-in-classcompanies

deliverAIROIthatexceeds

theircostofcapital.

AIRoI

14%

13%

13%

30%

12%

more

10%

Averagecostofcapitalforcompanies

~10%

9%

8%

7%

5.9%

AverageROIofallorganizations

AverageROIfortop10%oforganizations

5

BeyondopportunisticAI

Asorganizations

figureoutwhere

andhowtodeploy

AI,boldbets

translateto

biggergains.

OrganizationshavebeenbettingonAIforthebetterpart

ofadecade—andthelearningcurvehasbeensteep.3

Somegotcaughtupinthe“wowfactor”ofthetechnology,forgettingtoalignprojectstostrategy.OtherssawAIasahammer,andeverybusinessproblemanail.Almostallstruggledtoscaletheirimplementationsbeyondexperiments,proofsofconcepts,andpilots.

Thegoodnewsfromourresearch:Manyorganizationshaveturnedacorner.AIrolloutsaremoresuccessfulthanever,withmorethantwiceasmanyexecutivessayingtheirorganizationusedAIeffectivelyin2021(54%)thanin2020(25%).TheyalsoexpectAIinvestmenttogrowto6.5%ofITspendby2024.

Overall,returnoninvestmenthasbeenrisingsteadilysince2020.Forenter-prise-wideAIinitiatives,averageROIhasgrownfromjustover1%inearly2020tonearly6%bytheendof2021.4ThiscouldbearesultofthepandemicpushingorganizationstoinvestinAI-drivensolutionsthatwouldexpediteremoteworking,enhancetheuserexperience,anddecreasecosts.

TogaugewhetherAIROIhaskeptpacewiththistrend,wesurveyedmorethan350executivesagaininAprilandMay2023.WefoundthatAIROIhascontinuedontheexpectedgrowthtrajectory,reachinganestimated8.3%in2022(seeFigure2).5

6

10%

Still,thesereturnsarelowerthanthecostofcapital,whichistypically10%acrossmostindustries.Overall,fewerthanoneinfourorganizationsinoursurveysaythey’veachievedAIROIhigherthan10%.

Inessence,AIisfollowingthe“J-curve”patterntypicalfortransformativetechnologies.6Adoptingemergingtechatscalerequiresreimaginingbusinessmodels,workflows,skills,andmanyotherfacetsofbusiness.Returnsoftenstagnatewhileteamsworkoutthekinks.Yet,ascapabilitiesmature,performancecanrisequickly.Inthisenvironment,enterprisesneedtohaveastrategicplanforscalingtheimpactofAIovertime.

FIGURE2

Returnsareontherise

ROIfromAIprojectsgrew

morethan6timesbetween

2020and2022.

AIRoI

9%

8%

7%

6%

5%

4%

3%

2%

1%

0%

8.3%

5.9%

21%

13%

Early2021

2022

(estimated)

2020

Late

2021

Source:2020:Deloitte/ESLAIsurvey;Early2021:IBVAIethics

survey;Late2021:IBVAIcapabilitysurvey;2022:IBVgenerativeAI

pulsesurvey,April-May2023.

7

OuranalysisrevealshowROIimprovesacrossacontinuumofAImaturity(seeFigure3).Theaverageenterprise,whichpursuesadhocand/oropportu-nisticAIinitiatives,isthelaggard.In2021,thosethatintentionallyembeddedAIinproducts,services,businessunits,andfunctionssawROIclimbabove7%.Asmaturityextendedastepfurther,withAIdeployedaspartofastrategicbusinesstransfor-mation,returnsimprovedagain,to8%.

AsorganizationsfigureoutwhereandhowtodeployAI,boldbetstranslateintobiggerandbiggergains.Atthetopofthecurveisourbest-in-classcategory,reportinganaverageROIof13%.Bytakingasteady,balancedapproachtoadoptingAI—whichincludesbuildingdataandanalyticsskills,developingamultidisciplinaryapproach,creatingdiverseteams,andtrainingteamsthroughAICentersofExcellence(CoEs)—they’vedevelopedcomprehensivecapabilitiesthroughouttheenterprise.

FIGURE3

AI-centricstrategiesboostROI

AligningAIwithbusinessprioritiesyields

muchhigherROIthanadhocprojects.

Product

embedded

Newbusiness

modelenabler

Organicgrowth,

innovation,and

differentiation

Transformation

andplatform

strategy

EmbeddingAIintocoreproducts

UsingAItohelpachieveplatformeconomics

AIstrategiccontinuum

Horizontallydeployed

Verticallyintegrated

Customer/operationaleffectiveness

Opportunistic

Customer/

operational

effectiveness

DeployingAIinto

functions

Adhoc

Strategic

importanceofAI

Varied

IntegratingAIwithinbusinessunit(s)

Companytypes

ROIof4.7%7.2%7.4%8.0%

organizations

Perspective

FromAlphabet

toWalmart:

Becomingan

AI-firstbusiness

Legacyenterprisesarelikeindustrial-eracities.They’reconnectedbywindingcorridorsthatgreworganicallyratherthanstraight,efficientavenues.Theyfacepersistentchallengeswithmodernization,asprogressisoftenhinderedbycentury-oldinfrastructure.

AI-firstcompaniesstartwithacleansheetofpaper.Leaderscanbemorecreative,moreflexible—andmakeemergingtechnologiescentraltothebusinessmodel.“Ihaven’tencountereddigitalnativeswhosay,‘I’mgoingtoinnovate.’Ionlyhearthosestatementsfromlegacyorganizations,becausethey’retryingtogetoutofthatrut,”saidRaviSimhambhatla,ChiefDigitalOfficeratAvisBudgetGroup,inarecentIBVreport.“Fordigitalnatives,it’sallaboutdisruptingthemselves.”7

Digitalnatives,suchasAlphabet,Netflix,Amazon,andMeta—high-growthbusinesseswithAIatthecore—haveseenoutsizedreturnsonAIinvestments.8YetsomelegacybrandshavealsothrivedwithAI.Walmart,forinstance,usesAItomatchitsinventorytoshiftingcustomerneeds.Thebrandtapscustomerandshoppingtrenddatatoanticipatewhereandwhenpeoplewillwantspecificproducts.ThisletsWalmartstockeachwarehousewiththerightitems,stream-lininglogisticsandenablingspeedydelivery,evenduringpeakshopping

seasons.9

Thiscapabilitydidn’tappearovernight.Itwasbuiltonresponsibledatacollectionandcuration,thecreationofflexiblealgorithms,andaholisticapproachtotechnology.Takentogether,theseinitiativesproduceAI-driveninsightsWalmartcantrust.

Walmart,whichhasbeenaleaderindataandanalyticsfordecades,understandsthatAIrequiresmeasurementandoptimizationtoachieveitsfullpotential.ItsabilitytotrackdesiredoutcomesanddiagnosechallengesenablesthecompanytodrivemorevaluefromAI—andbuildascalablecapabilitythatcanbeleveragedacrossmanycurrentandfuturebusinessapplications.

Walmart’ssuccessdemonstratesthat,whileatraditionalenterprisecan’tbecomeadigitalnative,itcantransformtoemulateinbornagility.Companiesthatfocustheireffortsinkeyareas—embeddingAIintocoreoperations—willseebetterresultsthanthosethatspreadthemselvestoothin.

8

DataandAI:

Feedingavirtuouscycle

Datacanclosehalfthegapbetweenaverageandworld-classAIROI.

Tobuildaworld-classAIorganization,onefactorcomesfirst:Howanorganizationchooses,collects,governs,andusesitsdata—thatabundantbutelusiveresource—eitherenablesorconstrainswhatAIcanachieve.

Datahassometimesbeencomparedtooil:avaluableresourcethat’sexpensivetoextractanddifficulttoprocess.Ifdirty,itcanpolluteanentireecosystem.Butwhentappedresponsibly,it’sworthbillions.

That’sbecausereliable,representational,consensualdataisfoundationaltotrustworthyAI.Peoplewon’tuseAIsolutionstheydon’ttrust—andorganizationsthatplacegreaterimportanceonAIethicsreportagreaterdegreeoftrustfromtheircustomersandemployees.10

ItalsohelpsclosetheROIgap.Companieswithhigh“datawealth”aren’tyetattheworld-classlevel,buttheyhavelargestoresofhigh-qualitydata,monetizedataeffectively,andsaytheirdataistrustedbyinternalandexternalstakeholders.Ouranalysisrevealsthattheseattributesdrivehigher-than-averageROIandenablemoreeffectiveAIprojects(seeFigure4).

FIGURE4

Datadifferentiators

Companieswithmoreholisticdata

practicesseebetterbusiness

Datawealthoutperformers

9.0%

outcomes.

Allothers

4.8%

ROIrealizedfromoverall

enterpriseAIcapabilities

EffectivenessofAIprojects47%77%

9

Perspective

Disruptlikeadigitalnative

Lyfthasdisruptedthetransportationindustrywithdata-drivenprocessesthatoptimizebusinessdecisionsandredefinecustomerexperiences.Byleveragingtechnologytotapunmetmarketdemand,itbroke$1billioninrevenueinitsfirstyearofoperation.Bytheendof2022,thatfigurehadtopped$4billion.11

Lyftfocusesonmeetingcustomerneedsinrealtime,usingmachinelearningmodelstomakehundredsofmillionsofdecisionseachday,includingoptimizingrideprices,matchingriderswithdrivers,andpredictingarrivaltimes.12

Makingreal-timeinferenceswithmachinelearningatscalerequiresaccesstoavastamountofdataandcomputationresources,optimizedprocesses,andatalentedteamofdatascientists,engineers,andAIexperts.AndLyfthastonsofinformation:20.3millionactiveridersinQ42022,hundredsofmillionsoftripsperyear.13Thismassivestoreofdatapowersreal-timebusinessdecisionsthatreducecosts,optimizeresources,andstreamlinethecustomerjourney.

10

Ineffect,high-quality,high-value,trusteddataunlockshalfoftheROIimprovementweseeinbest-in-classorganizations.Thatsaid,dataaloneisnotenoughtofullyrealizeAI’spotential.Whiledataquality,quantity,robustness,value,andtrustareallimportant,howbusinessesharnessdatahasabiggercumulativeimpactonROIthanwhatdatatheyhave.

Today’stop-performingChiefDataOfficers(CDOs)specializeingettingvaluefromtheirorganization’sdata.Anelitegroup—just8%ofCDOsinIBV’smostrecentsurveyofdataleaders—reapmorevaluethanpeerswhilespendingless.14What’skeyishowtheyuseAItoimprovetheirdata:ThreeoutoffoursaythatapplyingAItotheirdatahelpsthemmakefasterandbetterbusinessdecisions.

So,it’snotjustaboutusingdatatoimproveAI—AIcanalsohelpcompaniesmakebetteruseofdata.It’savirtuouscycle.AsMircoBharpalania,SeniorDirectorofCrossDomainSolutionsfortheLufthansaGroupsaid,“AIissocriticalbecauseitactuallyopensuptheworldofthedatathatwe’resittingon.”15

11

Sixkeycapabilitiesthatenableworld-classresults

Whatenablessomeorganizationstoachieveworld-classROIfromtheirAIinvestments?Howdotheyamplifyhigh-quality,trusteddatatounlockfinancialandbusinessvalue?

Toanswerthesequestions,wecarefullyanalyzedourstudyresults,lookingforpatterns,insights,andapplicablereal-worldlessons.Welearnedthatbest-in-classAIperformersbuildcapabilitiesacrosssixkeyareas,inaholistic,integratedway—withtrustatthecore(seeFigure5):

–Visionandstrategy

–AIoperatingmodel

–AIengineeringandoperations

–Dataandtechnology

–Talentandskills

–Cultureandadoption

12

Trust

Talentandskills

Deployanenterprise-wideapproachtodevelopAIethics,skills,andtalent

Cultureandadoption

AIengineeringandoperations

Developahuman-centeredapproachtoAI—withdynamic,openfeedbackloopsacrosstheecosystem

FIGURE5

Becomingbest-in-class

CompaniesthatseethehighestROI

fromAIhavematured6keycapabilities—

withtrustatthecore.

6key

capabilities

Dataandtechnology

Buildcorecapabilitiesthatsimplify,automate,control,andsecureaccesstodata

Visionandstrategy

IdentifywhereAIcanboostcompetitiveness,innovation,andperformance—andprioritizeaccordingly

forhigh

AIROI

AIoperatingmodel

EmbedanAIoperatingmodel

intothefabricandcultureofthe

organization

DeployAIsolutionsthatare

flexible,user-friendly,and

scalable

13

#1Visionandstrategy

Don’tthrowAIateverything

ApplyingAI,automation,oranyothertechnologytopoorlydesignedprocessesstilldeliverssubparoutcomes.Byassessingwherestrategicinvestmentisplannedforcoreandnon-corefunctions(forexample,customerservice,marketing,supplychain,finance,andsoforth),aswellasbusinessunits,leaderscanuncoverstrategicopportunitiestoembedAI.

Awell-thought-outAIstrategycancatalyzetransfor-mationeffortsandincreasetheROIofindividualAIprojects(see“BostonScientificspends$50,000tosave$5million,”page15).Accordingtoourresearch,organizationsthatviewAIasimportanttotheirbusinessstrategyare1.8timesmorelikelytobeeffectivewiththeirAIinitiativesandachievenearlytwicetheROI(seeFigure6).

Leadersalsobalancecompetitivedifferentiationwithcostoptimization.Someareevenleveragingpubliclyavailableandopen-sourceAIresourcestodeliverfaster,cheaper,andmorescalablesolutionstomarket(see“FoundationmodelslaythegroundworkforAI’sfuture,”page14).EthicalquestionsabouthowthesetoolshavebeentrainedwillalsoplayapartinAI’sfuture—socompaniesneedtodefinewheretheystandbeforetheypushtoofarforward.

FIGURE6

Strategycomesfirst

CompaniesthatuseAIto

advancestrategyseenearly

twicetheROI.

8%

1.7x

more

7%

AIISimportantintheareaofenterprise/businessstrategy

8.0%

6%

5%

4%

AIisNOTimportantintheareaofenterprise/businessstrategy

4.7%

3%

2%

1%

0%

ROIrealizedfromoverall

enterpriseAIcapabilities

14

Perspective

Foundationmodels

laythegroundwork

forAI’sgenerative

future

AIisgettingsmarterandfastereveryday.Butmostsolutionsarestillbespoke.They’retrainedusingaspecificdatasettocompleteapre-definedtask—aprocessthatisbothenergy-intensiveandtime-consuming.16

TomakeAIinvestmentsmorecost-effective,companiesneedflexible,reusablemodelsthatcanbeappliedinavarietyofways—includinggeneratingnewcontent.Today’sfoundationmodelsarepavingapathtowardthisfuture.

TheyofferanopportunitytoaccelerateandscaleAIadoption,asfoundationmodelscan,intheory,beappliedtomanydomains.Forexample,LLMscantransformhowinformationisgeneratedandsharedacrossanorganization.Itjustneedstobeadaptedforsemanticsearch,classification,prediction,summarization,andtranslation.

TheadoptionoffoundationmodelsisalsosupportedbyasetofemergingAIengineeringbestpracticesthathavegonemainstream.Frommodeldevelopmenttopromptengineering,thesecommonpracticesandapproachesstreamlinecollaborationacrosstheenterprise—andtheecosystem.AsetoflayeredstacksandAIarchitectureswithstrongopensource,ecosystem,andresearchcontributionsarealsogivingrisetocommon,re-usabledevelopmentanddeploymentapproaches.

Whilefoundationmodelsofferrealpromiseandpotential,theyalsocomewithnewchallenges.Forone,theyrequiresignificantcompute,storage,andnetworkresources,whichmakesthemenergyintensive.Trainingonelargenaturallanguageprocessingmodelhasroughlythesamecarbonfootprintasrunningfivecarsovertheirlifetime.17

ItisalsoimportanttoconsiderhowtheusagescaleofafoundationmodelinfluencesROI.ALLMthatistrainedtoservehundredsofmillionsofusersmaydelivermorevaluefasterthanamodelthatisusedbyonlythousands.Inasmallerscaledeployment,optimization,fine-tuning,specialization,andportabilitymaynottranslatetonear-termreturns—eventhoughtheLLMcanbeusedformanydownstreamtasks.

Otherchallengesthatcomewithlargermodelsincludetrustworthiness,explainability,andtransparency.Addressingtheseissuesrequiresadditionaleffort,investmentand,insomecases,newinventionsandsolutions.Teamsmustunderstandwhatlargemodelscando,howtheyshouldbedeployed,andwhattypeofdatacurationisrequired.Broaderdataengineeringskillswillbecritical—aswillaseriousfocusonethicalissues.

Likeanyotherdisruptivetechnology,therearetrade-offsthatcomewithadoptinggenerativeAIandfoundationmodels.Successwillonlycomefromexperimentationanditeration.Especiallyforenterprises,thisjourneywillinvolvebalancingthescalesbetweenthevaluegenerativeAIcancreateandtheinvestmentitdemands.ThefutureofAIwillbedefinedbythosewhohittherightmark.

Casestudy

BostonScientific

spends$50,000to

save$5million

BostonScientificwantedtoautomateitsstentinspectionprocesstoimproveaccuracywhensearchingfordefects,suchasbrokenlinksorsurfaceimperfec-tions.Accurateinspectionsarecriticalforsuccessfulclinicaloutcomes.18

Thecompanyhasapproximately3,000expertsdoinginspections,costingseveralmilliondollarseachyear.BostonScientificconsideredaneuralnetworkmodeltohelpcutbackonmanuallabor,butthosemodelsrequiremuchmoredatathanithadonhand.Andcollectingorgeneratingthisdatawouldbeimpracticalandcostprohibitive.

Thesolution?First,teamsscaleddowntheproblembyfocusingonsmallerandnarrowertasks.Then,theyreduceddatarequirementstoalignwiththenewfocus.Lastly,theyleveraged“off-the-shelf”open-sourceAImodelstostreamlinetheinspectionprocess.

Theresult?$5millionindirectsavings—deliveredonamodestbudgetofroughly$50,000—aswellasincreasedaccuracy.Nowthecompany’semployeescanidentifyissuesfasterandfocusontheimportantworkthatonlyhumanscando.

15

16

#2Operatingmodel

Ditchthesciencefairmentality

GroundbreakingAIisbuiltonafoundationofopeninnovation.However,leadingcompanieslearntomitigateagainstthemyththatanythinggoesininnovation.19Tokeepexperimentsandimplementa-tionsinlinewithstrategy,organizationsmusttreatAIasadiscipline.Theymustoutlineethicalprinciples,developrigorousgovernance,andemphasizepragmatismovertheory.

ThisstartswithunderstandingwhichAIoperatingmodelbestalignswiththebusinessneed(forexample,centralizedversushub-and-spokeversusdecentralizedstructures).OurresearchfindsthatorganizationswithhighdatawealththathavealsoembeddedanAIoperatingmodelintothefabricandcultureoftheorganizationareabletogenerateupto2.6timesmoreROIthantheirpeers.20

Whatdoesthislooklike?Oneexamplerevolvesaroundhowcompaniescreateminimumviableproducts(MVPs).LeadersshouldoutlineaclearprocessforapplyingAI—startingwithidentifyingthebusinessproblemthesolutionhopestosolve.Bysettingcleargoalsforevenexperimentalrollouts,companiescanchoosetoadvanceonlythemosteffectiveAIprojects.

#3Engineeringandoperations

AgileDevOps+automatedITOps+MLOps=AIOps

AIengineeringandoperations(AIOps)bringsbigideastolife,servingasaflywheelfortheoperatingmodel.Itintegratespeople,processes,andplatformstoapplyAIatspeedandscale(see“BestsellerunlocksAIvalueinfastfashion,”page17).Andorganizationsthatsuccessfullydesignprocessesthathelpteamsbuildtoscale—whilealsomonitoringtheperformanceofAIapplications—seeupto2.6timeshigherROI.

EngineeringdisciplinecanacceleratethisAIflywheelandmakeitworkeffectively.JustasmanycompaniesuseDevOpsandothersoftwareengineeringapproachestospeedupprojectswithoutsacrificingquality,AIOpshelpsshortendevelopmentcycles,improvecollaboration,increaseoperationalefficiency,anddeploysolutionsmoresuccessfully.21Standardizationandstructuredfocusareessentialtokeepupwiththepaceofinnovation—withoutsacrificingtheprinciplesofethicalAI.

Casestudy

BestsellerunlocksAIvalueinfastfashion

Inthefashionindustry,around80%ofmerchandiseissoldacrosstwoseasonseachyear.Everythingelseishighlydiscounted—orgetsdonatedordumped.Thisover-productiontranslatestosuboptimalprofitsandpresentsanenormoussustainabilityissueforclothingdesignersandretailers.22

Tohelpteamsmoreaccuratelypredictdemand,clothingandaccessorycompanyBestsellertook10,000images(oneseason’scatalog)anddevelopedanAImodelforeachofitsfourbrands.Injustthreeweeks,thecompanywasabletodevelopandtrainaconvolutionalneuralnetworktoclassifyanimageaccordingtovariousfeatures.Deep-learningdetailswerethenfedintotraditionalanalysismodels(forexample,regressionorprincipalcomponentanalysis)tohelpthecompanybetterunderstandthefactorsthatdrivesales.

IncorporatingthisinformationintoBestseller’scoreforecastingengineincreasedthecompany’ssellingefficiencyfrom78%to82%—andreducedthenumberofdesignsamplesitneededtocreateforeachbrandby15%.

#4Dataandtechnology

Supportindustrial-strengthscaling

Anyonecancreateaproofofconcept.ButforAImodelstobeeffective,useful,andtrustworthy,theymustbeproperlyintegratedintooperationalsystems.WhatacompanycandowithAIisdefined,inlargepart,byhowitselects,governs,analyzes,andappliesdataacrosstheenterprise.Becausehumansarefallible,teamsneedskillsandprocessesthathelpensuretherightdataischosentopowerAImodels.

ThisalsohasamajorimpactonAI’sROI.At

world-classorganizations,datateamsreviewgovernance,management,ethics,literacy,andotherframeworksneededforpeopletoaccess,understand—andtrust—enterprisedata.ITteamsassessinfrastructureandprocessestobalanceAI

experimentationwithindustrial-strengthscaling(see“HowIBM’sChiefAnalyticsOfficerhelpsboostAIROI,”page18).

17

EveryonehasbigideasforAI.TheofficeofIBM’sChiefAnalyticsOfficer(CAO)helpsthosevisionsbecomereality.

TheCAOofficepartnerswithbusinessunitswithinIBMtoidentifyopportunitiestoimproverevenue,savetime,andaddalayerofintelligencetoday-to-daybusinessworkflowswithmachinelearnin