TDWISnapshotSeries

2024

StateofAI

Readiness

ByFernHalper,Ph.D.

TDWI'sresearchexaminesthecurrentstateofAI,howreadyorganizationsaretoimplementit,andkeyareascriticalforsuccess

Co-sponsoredby:

TableofContents

ResearchMethodology 2

TheScopeandImportanceofAI 3

TheOverallStateofAIReadiness 4

TheStateofOrganizationalReadinessforAI 6

TheStateofDataReadinessforAI 8

TheStateofSkillsandOperationalReadinessforAI.10

BuildingDataandAILiteracy 13

TheStateofGovernanceReadinessforAI 14

AIModelGovernance 15

ConsiderationsandBestPracticesforAIReadiness 16

AboutOurSponsor 18

AbouttheAuthor 20

AboutTDWIResearch 20

2024

Stateof

AIReadiness

ByFernHalper,Ph.D.

?2024byTDWI,adivisionof1105Media,Inc.Allrights

reserved.Reproductionsinwholeorpartareprohibited

exceptbywrittenpermission.Emailrequestsorfeedback

toinfo@.Productandcompanynamesmentionedhereinmaybetrademarksand/orregisteredtrademarks

oftheirrespectivecompanies.Inclusionofavendor,

product,orserviceinTDWIresearchdoesnotconstituteanendorsementbyTDWIoritsmanagement.Sponsorshipofapublicationshouldnotbeconstruedasanendorsementofthesponsororganizationorvalidationofitsclaims.Thisreportisbasedonindependentresearchandrepresents

TDWI’sfindings;readerexperiencemaydiffer.The

informationcontainedinthisreportwasobtainedfrom

sourcesbelievedtobereliableatthetimeofpublication.

Featuresandspecificationscananddochangefrequently;

readersareencouragedtovisitvendorwebsitesforupdatedinformation.TDWIshallnotbeliableforanyomissionsor

errorsintheinformationinthisreport.

1

2024StateofAIReadiness2

ResearchMethodology

ThisStateofAIReadinessReportexaminesthe

currentstateofAIandhowreadyorganizationsareto

implementit.IthighlightskeyareascriticalforAIsuccess:

organizationalreadiness,datareadiness,skillsandtools

readiness,operationalreadiness,andgovernancereadiness.Itexamineschallengesorganizationsarefacingingetting

readyforAI.Additionally,itprovidesconsiderationsandbestpracticesformovingforwardwithAI.

Forthisstudy,TDWIexaminedseveralsurveysand

assessmentsthatwerunthroughouttheyear.Data

fromthisreportcomesprimarilyfromthe2024TDWI

AIReadinessAssessment.Over100respondentsfrom

variousindustriesandcompanysizeshaveparticipatedintheassessmenttodate,and113completedresponsesarereflectedinthefiguresinthisreport.Additionally,data

fromover250respondentstothe2024TDWIDataandAnalyticsSurveyisusedinthisreport.

ThisreportwassponsoredbyMongoDB,Pecan,andSAP.

2024StateofAIReadiness3

TheScopeandImportanceofAI

Fromatechnologicalstandpoint,AIisanumbrellaterm

AI

encompassingamyriadofmethodologiesandtechniques.Itleveragesadvancesinmathematics,computerscience,

computationallinguistics,cognitivesciences,and

robotics,amongothers.PopularAItechnologiesinclude

machinelearning,naturallanguageprocessing,and

neuralnetworks,whichcollectivelydrivetheintelligent

capabilitiesofmodernAIsystems.TDWIseesorganizationsbuildingAImodelstopredictchurnandothercustomer

behavior,identifyfraud,determinewhenmaintenancewillbeneeded,recommendproducts,andpredictdisease,

ComputerScienceMathematics

CognitiveSciences

ComputationalLinguistics

MachineLearning

Neural

Networks

amongotherusecases.Fromasocietalperspective,AI—particularlygenerativeAI—hasthepotentialtotransformindustries,enhancedecision-making,andsolvecomplexproblems,makingitanimportantforceintheongoing

digitalrevolution.

Practicallyspeaking,AIcanprovidetangiblebenefits

suchasdeeperinsights,increasedproductivity,improved

Natural

LanguageProcessing

Robotics

customerservice,andgreateroperationalefficienciesthatdrivecostsavingsandstrongertop-linegrowththatdeliverslargerprofits.InTDWIresearch,forinstance,weseethat

organizationsimplementingmoresophisticatedanalyticssuchasAIaremorelikelytoderivetop-orbottom-linebenefitsfromtheiranalyticseffortsthanothers.1Inotherwords,thereisreal,tangiblevaluefromAI.

1See2023TDWIBestPracticesReport:AchievingSuccesswithModernAnalytics,onlineat

/bpreports

.

2024StateofAIReadiness4

ToachievethebenefitsfromAI,organizationsneedto

understandtheproblemstheywanttosolveandhaveasoliddatafoundationthatsupportshighvolumesofdiversedata.TheywillneedtohaveorganizationalsupportandaculturethatchampionsAI.Thisincludestheskillsandtrainingto

makeAIareality.EnterpriseswillneedoperationalmodelsandteamsinplacetodeployAIintoproductionandensurethatthosemodelsstaycurrent.TheywillneedtogovernAItoensureitmeetscompliance,quality,andethicalconcerns.

WiththeadventofgenerativeAI,moreorganizationsare

feelingthepressuretoputanAIprogramtoworkintheir

company.Withoutthenecessaryfoundations,successwill

bedifficult.Executivebackingforthemoveisnecessarybutit'snotenough;organizationsmustbereadyforAIwiththeconsiderationsalreadymentioned.Forthesereasons,TDWIdevelopedtheAIReadinessAssessmentsoorganizations

couldseejusthowpreparedtheyaretostartanAIjourney.

TheOverallStateofAIReadiness

Therearenumerousinterrelatedfactorsthatformthe

currentstateofAIreadiness.Readinessisnotsimplya

matteroftryingoutAIsolutionsorbuildingapredictive

analyticsmodel.AIreadinessinvolvespeople,processes,

andtechnologies.Inthe2024TDWIAIReadiness

AssessmentwemeasuredfivekeydimensionsofenterprisefitnessandreadinesstouseAIoperationally:organizationalreadiness,datareadiness,skillsreadiness,operational

readiness,andgovernancereadiness.

Theoverallmedianscoreforallenterprisesrespondingtothe2024assessmentwas62outof100,whichputsrespondentsattheStandardizingstageofreadiness.

(Figure1).Likewise,forallrespondents,eachdimensionscoredanaveragethatputitintheStandardizingstage.

stage1stage2stage3stage4stage5

Organizationalreadiness

12.6/20

Data

readiness

12.2/20

Skills

readiness

10.8/20

Operationalreadiness

11.9/20

Governancereadiness

11.6/20

Figure1.ThestagesofAIreadinessandparticipants’averagescoresforeachdimension.

2024StateofAIReadiness5

DuringtheStandardizingstage,thecompanyisintheprocessofputtingastrategyinplaceforAI.Thismayincludesomepreliminaryusecasesanorganizationislookingtoputinplace.Theymayalreadybebuildingproofsofconcept.Leadershipistypicallyonboard

withAIandunderstandstheimpactitcanhaveonthe

business,althoughitmaynotunderstandwhatittakestoimplementAI.ThatalsomeansthatleadershipmaynotyetbecommittedtoinvestingthenecessaryresourcesforAIdevelopment,includingtechnology,talent,andtraining.

Inthisstage,anorganizationistypicallycollectingmore

thanjuststructureddata,althoughitmaynotbeanalyzingityet.Thismayincludeemails,callinteractionnotes,

freeformtextinsurveyanswers,socialmediadata(blogs,tweets),machine-generateddata,geospatialdata,real-timeeventdata,audio,video,weblogs,clickstreams,

scientificdata,anddemographicdata.Theyareworkingtomakethisdataeasilyaccessible.Theytypicallyutilizearangeofdataplatformsincludingadatawarehouseordatalake,bothonpremisesandinthecloud,tomanagethisdiversedata.Theseplatformscansupportbuildingpredictiveanalyticsandmachinelearningmodelsagainststructureddata.Enterprisesaretryingtoimplementanarchitecturetosupportdatagrowth.

Anorganizationatthisstageistypicallystartingtobuildpredictivemodelsandmayhaveafewdatascientists

inplace.Theyrealizethatthereareotherskillsthatareneededtoo,suchasdataengineersandoperationstobuildpipelinesandputmodelsintoproduction.They

Inthisstage,anorganizationistypically

collectingmorethanjuststructureddata,

whichcaninclude:

Emails

Geospatialdata

Demographicdata

VideoAudio

Tweetsand

blogs

Surveyanswers

mayormaynotbebeginningtohirethiskindoftalent.

Typically,anorganizationisonlystartingtothinkaboutAIgovernanceoroperationalizingAI.

TDWIhasseensimilarresultsinothersurveys.For

instance,inouryearlydataandanalyticssurveywesee

thatorganizationsaremovingtowardspredictiveanalyticsandmachinelearning.Theirtopprioritiesaretoputadatainfrastructureinplacetosupportthesemodernanalyticsandtogovernthisenvironmenteffectively.2

ThefollowingsectionsdrillintoeachoftheAIReadinessdimensionsinmoredetail.

2Unpublished2024TDWIsurvey.

2024StateofAIReadiness6

TheStateofOrganizationalReadinessforAI

OrganizationalcommitmentiscriticalformovingforwardwithAI.Organizational

readinessscoredthehighestmarksintheassessment.Thisincludesleadership

commitment.AtTDWIweseethatalthoughexecutivesandothersmayappreciatethevalueofAI,thatdoesn’tnecessarilymeantheywillprovidethehelpneededtobuildadatastrategy,workonaculturetosupportit,fundit,orcommunicatetheeffort.Forinstance,intheAIReadinessAssessment,althoughmanyorganizationsareworking

onanAIstrategy,onlyabout30%haveoneinplaceandareexecutingonitorare

currentlyintheprocessofbuildingitoutnow(Figure2).Over35%areintheprocessofthinkingaboutthestrategy.OtherTDWIresearchfoundsimilarresults.

9%

OurAIstrategyisalreadyinplaceandweareexecutingonit

MycompanyhasadefinedstrategyforleveragingAItoenhanceourcompetitiveadvantage

21%

Yes,weareinthe

processofbuildingout

anAIstrategynow

37%

20%

No,butweplanto

dosointhenext

year

Yes,weareintheprocess

ofthinkingthroughhow

todothatnow

Figure2.StatusofAIstrategy.

No,andwe

13%havenoplans

todoso

2024StateofAIReadiness7

Likewise,manyorganizationsareonthefencewhenitcomestowhethertheirorganization’sleadershipiscommitted

toinvestingthenecessaryresourcesforAIdevelopment

(includingtechnology,talent,andtraining)andrecognizingthelong-termvalueoftheseinvestments(Figure3).Less

thanhalfoftheassessmentrespondentssaytheirleadershiphascommittedfunding.

Respondentsarealsostillworkingonotherorganizational

readinessfactorsforAIsuchasaculturalshifttowards

innovation,continuouslearning,andadaptabilityrequired

forsuccessfulAIimplementationaswellasacultureoftrusttowardAI.Forinstance,inthereadinessassessment,only

28%ofrespondentsstatedtherewasacultureoftrustforAIintheirorganization(notshown).AboutathirdsaidtheyhadacollaborativeculturetosupportAI(notshown).

Myorganization'sleadershipiscommitted

toinvestingthenecessaryresourcesfor

AIdevelopment

13%

32%

37%

6%

11%

Completelydisagree

Completely

agree

Disagree

Agree

Neutral (Neitheragreenordisagree)

Figure3.LeadershipcommitmenttoAI,includinginvestments

intechnology,talent,andtraining.

2024StateofAIReadiness8

TheStateofDataReadinessforAI

Organizationalcommitmentcanmakeorbreakan

enterprise’sreadinessforAI,butsuccesswithAIrequiresmore.OrganizationsmustalsohavethedataplatformsinplacetosupportAI.AIoftendealswithlargevolumesofdiversedata.AImodelscanbecomputationallyexpensivetobuildandoperate,andthisisespeciallythecasewithgenerativeAI.Itiscriticalforanorganizationtohavea

soliddatafoundationtosupportitsAIefforts.

Theoveralldatareadinessscorewas12.2,secondonlyto

organizationalreadiness.Thismiddle-of-the-roadscore

isinlinewithwhatweseeinotherTDWIresearch.Many

organizationshavemadethemovetoclouddatawarehousesandclouddatalakestosupportdiversedatatypes.Inmanycases,theseorganizationshavehybridenvironmentsthat

areoftenmultiplatform.AIReadinessAssessmentresponsesindicatethatmanycompanieshavenotyetachievedthe

maturityintheirplatformstobecompletelyreadyforAI.

Forinstance,animportantpartofbeingreadyforAIisthe

abilitytointegratedatafrommultiplesourcestobuildan

enrichedandrobustdatasetfortrainingamodel.Althoughorganizationsaremakingprogressonthisfront—slightly

morethan40%ofrespondentscandothistoday(Figure4)—therestareonthefenceordon’thavethesystemsinplacetomakedataeasilyaccessibleforAI.

Myorganizationhassystemsinplaceto

ensurethatdataiseasilyaccessibleandcan

beintegratedfromdiversesources

40%

32%

35%

30%

30%

25%

20%

20%

15%

10%

7%

10%

5%

0%

Completelydisagree

Disagree

NeutralAgreeCompletely

(Neitheragree

agreenordisagree)

Figure4.Theaccessibilityofdatafromdiversesources,including

internalandexternaldatasets.

2024StateofAIReadiness9

InotherTDWIresearch,we’veseenthatmakingdata

accessibleformodelsisatopanalyticspriorityfor

organizations.3Moreenterprisesareestablishingthatdata

foundation.However,themajorityofassessmentrespondents(59%)eitherdisagreeorareuncertainabouthavingatrusteddatafoundationinplaceforanalytics(notshown).

Partofthechallengecanbeattributedtorespondentsstilldealingwithsiloeddatainfrastructures.Forinstance,intheassessmentweasked,“Whichofthefollowingstatements

bestdescribesthedataplatformtechnologiesyourcompanyutilizestoday?”(Figure5).Twenty-fourpercentareusinga

datawarehouseoradatamart.Another22%areutilizingadatawarehouseandadatalakebutsaytheyaresiloed.

Althoughthemajorityhavesometypeofplatform,thereareareasforimprovement.Forinstance,wheredatafromdifferentplatformsisintegrated,itistypicallystructureddata.Morethanhalfoftherespondentsdonotbelieve

theirorganizationhastheadvancedanalyticscapabilitiesandcomputationalresourcesnecessarytodevelop,train,anddeployAImodelsefficiently(notshown).

Toolsonthemarketcanautomaticallybuildmachinelearningmodels,buttheyneedthedatatobeatleastBIready(in

somesortofdatawarehouse).ThesetoolscanbeagoodstartingpointforAIprograms,buttheystillrequiremore(suchasoperationalreadinessandgovernancereadiness,discussedinthenextsections)tobecometrulyready.

Whichofthefollowingstatementsbest

describesthedataplatformtechnologies

yourcompanyutilizestoday?

Weuseflatfilesorspreadsheets

24%

Wehaveadatawarehouseordatamart

22%

Weuseourdatawarehousetogetherwithadatalakeorotherplatform,buttheyaresiloed

Weusearangeoftechnologiesincludingourdatawarehouse,datalake, cloud,orotherandwearearchitectingthemtogetherasanecosystem

Weusearangeofapproachesthatformawell-architectedenvironmentfordataaccess

0%10%20%30%40%50%

Figure5.Currentdataplatformtechnologiesinuse.

11%

38%

6%

3Unpublished2024TDWIdataandanalyticssurvey.

2024StateofAIReadiness10

Inotherwords,organizationsaremakingprogresswith

theirdatainfrastructurestosupportAI,butsomestill

haveworktodo.Thisexplainswhylessthan30%of

respondentsagreedwiththestatement,“MyorganizationhasacompanywidedataarchitectureinplaceforAIthatcanhandleusergrowth”(notshown).

Organizationswillneedtofocusontheirdatainfrastructure,especiallyiftheyplantobuildgenerativeAIapplications

utilizingcompanydata.Thisappearstobeapriorityfor

organizations;generativeAIrankedinthetopfourprioritiesforanalyticsin2024,aheadofmachinelearning.4Yet,AIorgenerativeAIsuccesswillrequireasoliddatafoundation,

capableofscalingandsupportinghighperformance.

4Unpublished2024TDWIdataandanalyticssurvey.

TheStateofSkillsandOperationalReadinessforAI

TomakeAIsuccessful,anorganizationneedstheright

skills—morethanjustdatascienceskills,thoughtheyareclearlyimportant.Enterpriseswillneeddataengineeringskills(forbuildinganddeployingdatapipelines),

operationalskills(forversioningmodels,puttingthem

intoproduction,andmonitoringthemfordriftoncein

production),anddevelopmentskills(forbuildingapps

utilizingAIandgenerativeAI).Knowledgeofthebusinesswillalsobeimportant.

Asmentioned,manyorganizationsarehiringdatascientiststohelpbuildtheskillsetneededforAI.IntheAIReadinessAssessment,aboutone-thirdofrespondentsdon’thave

anydatascientistsontheirteam,anotherone-thirdhavestartedtohirethem,andtherestalreadyhavetheminplace(notshown).

Thesedatascientistswillhelpanorganizationget

startedwithbuildingAImodels,butothers,suchasdataengineers,arealsocriticaltotheprocess.Infact,when

weaskorganizationswhatskillstheyneed,theyoftencitedataengineers.

IntheAIReadinessAssessment,weaskedwhether

respondentsemploydataengineerstobuilddata

pipelinesforAI.Theseengineersareresponsiblefor

managingandpreparinglargedatasetsforAI,ensuringdataqualityandaccessibility.

2024StateofAIReadiness11

Figure6showsthatabouthalfoftherespondentshavethisrole;therestdonot.Itwillbeimportantfororganizationstoplanforthisrolebecausebuildingandorchestrating

pipelinesisoftenchallenging.

ArecenttrendinpredictiveanalyticsandAIisto

democratizeit,inotherwordstoopenupAItoawider

audience.Insomecases,thisaudiencecanincludebusinessanalysts(thosewhobuilddashboardsandreports).TheseorganizationstypicallyhaveadatawarehouseordatalakeandanalystsarerunningBIreportsfromthisinfrastructure.

ManyoftheseanalystsarereadyforAI;theyareinterestedingrowingtheirskillset.Theymaybeboredwithsimply

producingdashboards.TheyalsounderstandthemovetoAIandwanttogotothenextstep.Theyunderstandtheirdataandthebusiness.

OneoftheissueswhenmovingintoanAIprojectishow

youwillbuildyourorganizationalculturesoitcanadopt

AImodels.Oftendatascientistsaren’tconnectedwith

thebusinessandthatcanbeanissue.Businessanalysts

dounderstandthebusiness.Intheassessment,21%of

respondentsfeltthatbusinessanalystshavetheskillstheyneedforAItoday;another43%feltthatbusinessanalysts

couldperformdatasciencewithhelpfromothers(not

shown).Thiscanrequiretherighttoolsthatareeasyenoughtouse.Forinstance,someofthetoolsonthemarket

providenaturallanguageinterfaces(oftenviagenerativeAI)tohelpemployeeswithoutdatascienceskillsbuildmodels.

Myorganizationemploysdataengineers

tobuilddatapipelinesforAI;theyare

responsibleformanagingandpreparing

largedatasetsforAI,ensuringdataquality

andaccessibility.

17%

No,andI'mnotsurewearethinkingaboutdataengineers

36%

No,howeverwerealizethisisimportantandmaybetryingtoworkonitadhocwithexistingstafforoutsourceit

20%

Wearedevelopingadedicatedgrouporstaffmembersresponsibleforthis,orweareworkingwithanoutsidegroup

15%

Wehavededicatedteammemberswithaspecificmandateandresourcesforthis

12%

Wehaveacompleteteamwithenough dataengineers,DevOps,andsimilarpersonneltosupportourfullcapacityofneeds

0%10%20%30%40%50%

Figure6.Statusofdataengineeringteamsforbuildingdatapipelines.

2024StateofAIReadiness12

Itwillalsobeimportantnottosimplyfocusonthefront

endofmodelbuildingbuttoconsiderwhoisgoingtoputmodelsintoproductionandthenmanagethem.Thisis

wheretheOpsteamcomesintoplay.Ifyoucan’tputthemodelintoproduction,yourorganizationwon’tgetthefullvaluefromit.

Yet,onlyaboutone-thirdofrespondentsagreedwiththestatement,“Myorganizationhasorplanstosoonhavethetoolsandskillsforeffectivedeployment,monitoring,andmanagementofAImodelsinproduction,ensuringthey

performasexpectedovertimeandcanbeupdatedor

retrainedasnecessary”(notshown).Usingdatascientiststofilltheserolessimplywon’tscale.

Additionally,AIoftenrequiresdeveloperstobuild

applicationsthatutilizeAI;thisisespeciallythecasefor

generativeAIapplicationssuchaschatbotsorthosethatusecompanydata.ThismaybeassimpleasusinggenerativeAItosummarizecallcenternotesoritmaybemorecomplex,suchasusinggenerativeAItogeneratepersonalized

marketingmessagesbyusingtraditionalmachinelearningmodelsinconjunctionwithgenerativeAIoutput.

Often,thisinvolvesusingnewertechnologysuchas

convertingeachwordinasentenceintoavectorusingapre-trainedwordembeddingmodelandstoringtheminavectordatabaseforusebythegenerativeAImodel.

Thismayalsoincluderetrieval-augmentedgeneration,wherethevectorembeddingmightbecombinedwithapromptandsenttothegenerativeAIsystemthatusesittogenerateanin-contextresponse.Buildingthese

applicationsoftencallsfordevelopmentexpertise.

However,lessthan30%ofrespondentsbelievethattheyhavetheskillsinthesoftwareengineeringtechniquesthatwouldberequiredforputtingmodelsintoproduction

inapplicationsorbusinessprocesses(notshown).This

illustratesthatmanyorganizationsarestillatthebeginningoftheirAIjourney.

2024StateofAIReadiness13

BuildingDataandAILiteracy

AnotheraspectofreadinessskillsforAIincludesdataliteracy.Asmentioned,insomecases,organizationsarelookingtodemocratizetheiranalyticsandmakeuseofautomatedandaugmentedtoolsonthemarketthat

canhelpbusinessanalystsandothersbuildAImodels.

Althoughthesetoolscanbeeasytouseandbusiness

analystshavethecriticalthinkingskillsandknowledgeofthebusinesstoexecute,theywillstillultimatelyneedtobetrainedinAItechniques.

Inadditiontobusinessanalysts,otherinthebusiness

mayneedtounderstandthefoundationalconceptsofAIiftheyareusingtheoutputsofmodelsfortheirjoborif

theyareutilizingtoolssuchasgenerativeAIforcreating

marketingcontent(orotherusecases).IntheAIReadinessAssessment,abouthalfoftherespondentsstated

theyregularlyscheduletrainingandencourage—ifnot

mandate—employeestoattendtoensureemployeesarewellequippedwiththelatestinAIandanalyticsknowledgeandskills,ortheyfundinternalandexternaltrainingfor

employeeswhoneedtobuildskillsorgrowprofessionally(notshown).Thisisdefinitelyastepintherightdirection.

2024StateofAIReadiness14

TheStateofGovernanceReadinessforAI

InTDWIsurveys,weroutinelyseethatdatagovernance

isatoppriorityfororganizations,especiallyastheytry

tomodernizetheirenvironments.TDWIdefinesdata

governanceasthepracticeofensuringthatbusinessdataremainsfitforuse.Itfocusesonthepeople,processes,

policies,rules,andregulationsforachievingspecificgoalsforamanageddataresourceandbuildingtrustinthedata.

AsorganizationsmovetoAI,theymustensurethattheir

dataremainsfitforuseandthatitisaccurate,complete,

andtimely.Theadage“garbageingarbageout”definitely

applieshere.Thisdataqualitywillneedtobeinplacefor

structuredaswellasunstructureddatatypes.PreviousTDWIresearchindicatesthatorganizationsareonlystartingto

thinkabouthowtoensuredataqualityfornewkindsofdata.

Additional