procurementsupercharged:

whathappenswhenagenticaiorchestratessource-to-pay

Sponsoredby:

EP·

procureconInsIGHTS

procurementsupercharged:whathappens

whenagenticaiorchestratessource-to-pay

executivesummary

Procurementfunctionstodayfacemountingpressureto

accelerateprocesseswhilereducingmanualwork.AgenticAIrepresentstheneweststageinarti?cialintelligence

evolution,movingbeyondsimpleautomationto

autonomousagentsthatcanreason,makedecisions,andactonbehalfofprocurementprofessionals.

ThiseBookexploreshowtheseintelligentagentsare

reshapingsource-to-pay.Here,readerswilllearnhow

thesesystemsarefundamentallydifferentfromtraditionalroboticprocessautomation(RPA)models,astheytransformprocessesthroughorchestratedwork?owsthatrequire

minimalhumanintervention.

CONTRIBUTORS

ChrisGovers

GEP

AndrewTumath

GEP

Sponsoredby:9GEP2

procureconInsIGHTS

procurementsupercharged:whathappens

whenagenticaiorchestratessource-to-pay

theevolutionofagenticaiinprocurement

Al,GenAl,LLMsandAlAgents

ficiantellienceA

acineearnin

eearninviaNNs

Emulatehumanintelligencewithcomputersystems

Beyondinferencerules,fore.g.,classification(s);clustering(u-s)

Neuralnetworks(likeourbrains)withalgorithmicfeedback

Generatepredictedcontent(“typeaheadonsteroids”;images/videos)

GenAlmodels(text)pre-trainedonmassdatasetsusing“transformer”architecture.“Language”canbemultilingualprose,metadata,computercodeoreven“l(fā)egalese”

AIAgents

CommercialLLMs:GPT4/o1/o3,Llama,Gemini,Claude,Qwen,DeepSeekV3,Grok.Manyarefinetunedforreasoningcapabilities(andcanalsobeagentic)

Agentic

Frameworks

Intelligentchatbots,copilotsandautonomousagents:off-the-shelf(ChatGPT,MS-Copilot,Gemini,Perplexity,Claude,Grok,DeepSeekR1)orcustom-developed/-tuned

Source:SpendMatters

Thejourneyfrombasicarti?cialintelligencetoagentic

systemsrepresentsafundamentalshiftinhowprocurementtechnologyoperates.TraditionalAIstartedwithmachine

learningandevolvedthroughdeeplearningtogenerativeAIandlargelanguagemodels.

ThesefoundationsnowsupportagenticAI,whichcombinesallpreviousinnovationsintoautonomoussystemscapableofindependentaction.

howagenticairedeHnesautomation

Thekeydifferentiatorliesinwhatthecontributorscall

“ReAct,”whichreferstothecombinationofreasoningandaction.Whilelargelanguagemodelscananalyzedataandgenerateresponses,theycannotreasonthroughcomplexproblemsorexecutedecisionsintherealworld.

Sponsoredby:9GEP3

procureconInsIGHTS

procurementsupercharged:whathappens

whenagenticaiorchestratessource-to-pay

HOWAGENTICAIISDIFFERENTFROMRPAORLLM

AnalyzeData

04

01

TakeAction

AgenticAi

Plan

02

●

05Collaborate

Strategically

ReasonandMake

06

03

AchieveGoals

Decisions

AgenticAIbridgesthisgapbyincorporatingsixcriticalcapabilities:

?Dataanalysis

?Strategicplanning

?Reasoninganddecision-making

?Actionexecution

?Collaborationacrosstechstacks

?Goalachievement

GEP’sexperiencewithover8millionsuppliersintheir

ecosystemprovidesthedatafoundationnecessaryfor

effectiveagenticAIimplementation.Thisvastdataset

enablesagentstomakeinformeddecisionsbasedonrealmarketconditionsratherthantheoreticalmodels.

“There’sthisphrasegoingaroundthatrelates

toagenticAI:‘ReAct,’thatis,reasonandaction.LLMsallowyoutoanalyzedatatogenerate

responses,whereasAIagentsmakeadifferencebyreasoninglikeahumanbeing,?guringoutasolutiontoaproblem,andthentakingsteps.”

AndrewTumath

GEP

Sponsoredby:9GEP4

procureconInsIGHTS

procurementsupercharged:whathappens

whenagenticaiorchestratessource-to-pay

Anticipatesanddeliversonuserneedsproactively

Dynamicallyorchestratesdecisionsandwork?ows

Self-optimizesandself-regulateswithouthumanintervention

autonomousserviceecosystems

self-organizingsystems

agenticaievolution

networkofagents

Outcome-drivenwith

autonomous,end-to-endexecution

Context-awareagents

autonomouslyretrievereal-timedataacrosssystems

Delivershyper-personalizedexperiencesthrough

continuousadaptation

traditionalai

saassolutions

Requireshumaninputthroughdashboardsandinterfaces

Siloedappswithdataintegrationissues

Needsfrequentupdatesandmanualtuning

generative&agenticai

evolutiontowardautonomousserviceecosystems

Theevolutioncontinuesasorganizationsprogressfrom

manualprocurementsolutionstowardnetwork-basedagentsystemsthatorchestrateentirework?owsautonomously.Inthisfuturestate,agentswillanticipateuserneedsanddeliversolutionsproactively.

Currentimplementationsalreadydemonstratesigni?cantvalue,butthetechnologycontinuesevolvingrapidly.

OrganizationsthatbeginimplementingagenticAI

todaypositionthemselvestobene?tfromincreasinglysophisticatedcapabilitiesasthetechnologymatures.

keysuggestions

?UnderstandtheReActframeworkbefore

implementingagenticAIsolutions.This

combinationofreasoningandactioncapabilitiesdistinguishesagenticAIfromprevious

automationtechnologies.

?Prioritizedataqualityasthefoundationfor

successfulagentimplementation.High-qualitydatainputensuresagentscanmakeinformed

decisionsanddelivervaluableoutcomes.

Sponsoredby:9GEP5

procureconInsIGHTS

procurementsupercharged:whathappens

whenagenticaiorchestratessource-to-pay

transformingsource-to-paywithmulti-agentsystems

?RequisitionAnalyzingAgent

?SupplierOnboardingAgent

?ProfileEnrichmentAgent

?PerformanceMonitorAgent

?CommunicationSummarizeAgent

?SupplierDiscoveryAgent

?AutoNegotiationAgent

Autonomous

?BidResponseAnalysisAgent

Sourcing

C

Intelligent

Category

Management

Management/Collaboration

Supplier

P?

SUPERAGENT

?ExpiryMonitoringAgent

?MetadataExtractionAgent

?ObligationExtractionAgent

?Post-SignaturePerformanceTrackerAgent

?MarketInsightsAgent?SpendAnalysisAgent

?StrategyBuilderAgent

?StrategyTrackerAgent

1:ORCHESTRATION

Risk

Management

Contract

2:TASKAUTOMATION

Lifecycle

Management

?OrchestrationAgent

?ItemRecommendationAgent

?OrderTrackingAgent

?ReceivingAgent

?SupplierScreeningAgent

Procure-to-Pay

(P2P)

?RiskClauseExtractorAgent

?TransactionAnomalyAgent

?RegulatoryComplianceAgent

Multi-agentsystemsrepresentadeparturefromsingle-

pointsolutionstowardorchestratednetworksofspecializedagents.Unliketraditionalprocurementplatformsrequiringuserstonavigatebetweendifferentmodules,agenticAI

createsseamlesswork?owswhereacentralorchestratormanagestheentireprocess.

Thisorchestratoractslikeanautonomousdriver,

coordinatingvariousspecializedagentswhilemaintaininghumanoversightwhereneeded.

theimportanceofhuman

intelligenceinagenticprocurement

Thehumanelementremainscrucialevenasautomation

increases.Procurementprofessionalscanberesponsible

forthinkingstrategically,sourcingethically,understandingnuance,andmaking?naljudgmentcalls,whileagentsexcelatspeed,scale,contextawareness,andpatternrecognition.

Thisdivisionoflaborallowshumanstofocuson

high-valuestrategicactivitieswhileagentshandleroutineoperationaltasks.

ThebalancebetweenhumanandAIresponsibilities

willcontinueshiftingastechnologyadvances.Currentimplementationsshowagentscaninterprethuman

emotionsandreactionsduringnegotiations,adjustingstrategiesbasedonsupplierresponsiveness.

FuturedevelopmentsmayexpandAIcapabilitiesinareas

currentlydominatedbyhumanintelligence,thoughhumanoversightwillremainessentialforethicalandstrategic

decision-making.

Sponsoredby:9GEP6

procureconInsIGHTS

procurementsupercharged:whathappens

whenagenticaiorchestratessource-to-pay

HumanRequest

HumanReview

UIAGENT

ORCHESTRATOR

AGENT

SELF-REFLECTION

SOURCINGNEGOTIATIONSCONTRACTSCOMPLIANCE

AGENTAGENTAGENTAGENT

howamulti-agentsystemworks

Multi-agentsystemsalsoaddressacommonprocurementchallenge:organizationalsilos.Agentscansharelearningsacrossregions,categories,andbusinessunits,creating

uni?edglobalapproaches.

Anagentthatsuccessfullyonboardsasupplierinonelocationcanapplythoselessonstosimilarsituations

worldwide,promotingconsistencyandbestpracticesacrosstheorganization.

keysuggestions

?Designagentimplementationsthatpreserve

humancontroloverstrategicandethicaldecisions.ThemosteffectivesystemscombineAIef?ciency

withhumanjudgmentforoptimaloutcomes.

?Implementorchestrationagentstocoordinate

work?owsacrossmultiplespecializedagents.Thisapproacheliminatessilosandcreatesseamlessend-to-endprocurementexperiences.

“Whentheorchestratoragentisactivated,itdetermineswhethersomethingisasourcingrequest,whetheritneedstonegotiatewithasupplieronthecompany’sbehalf,orifitshouldmoveintocreatingacontract.Italsochecksifeverythingmeetscompanyrulesandhowpeopleareinvolved.Theagentcontinuously

reviewsitsownactionsbeforemovingforward,makingtheprocessasindependentaspossible.”

ChrisGovers

GEP

Sponsoredby:9GEP7

procureconInsIGHTS

procurementsupercharged:whathappens

whenagenticaiorchestratessource-to-pay

thereal-worldapplications

ofagenticaiinsource-to-pay

CurrentagenticAIimplementationsinprocurementfocusonthreeprimaryareasthatdeliverimmediatevalueto

organizations.Theorchestrationagentservesasthecentralcommandcenter,interpretinguserrequestsanddirectingappropriatespecializedagentstocompletetasks.

Whenausersimplystates“Ineedasustainabilityconsultant,”theorchestrationagentanalyzesthisrequest,checksagainstexistingcatalogs,andeithercreatesarequisitionortriggersasourcingeventbasedonprede?nedbusinessrules.

autonegotiationandbidanalysis

Negotiationagentsdemonstratesophisticatedcapabilitiesbydetectingsupplieremotionsandadjustingstrategies

accordingly.Theseagentscanapplytargetpricingbasedoninitialbids,thenengagesuppliersinfollow-upnegotiationstoachievebetterterms.

Thebidresponseanalysisagentcomplementsthisbyevaluatingresponsesline-by-lineandmakingawardrecommendationsbasedoncomprehensivecriteria,includingprice,quality,andsupplierperformance.

ThenegotiationprocessshowcasestheReActframeworkinaction.Agentsreasonthroughoptimalnegotiation

approachesbasedonsupplierbehaviorpatterns,thenactbysendingtargetedcommunications.Ifasupplierappearsreceptivebasedontheirresponsepatterns,theagent

mightpursuemoreaggressivepricingdiscussions.

Conversely,ifasupplierseemsresistant,theagentadjustsitsapproachtomaintainrelationshipintegritywhilestill

seekingfavorableterms.

Sponsoredby:9GEP8

procureconInsIGHTS

procurementsupercharged:whathappens

whenagenticaiorchestratessource-to-pay

supplieronboardingandproHleenhancement

Supplieronboardingrepresentsanotherhigh-impact

applicationwhereagentssigni?cantlyreducemanualeffort.

Whencreatinganewsupplierpro?le,agentsautomaticallygatherinformationfrompubliclyavailablesources,

includingcorporateregistries,existingGEPsupplier

databases,andmarketintelligenceplatforms.Thispre-

populationeliminatesmostmanualdataentryforsupplierswhileensuringpro?lecompletenessandaccuracy.

Thesupplierpro?leenhancementagentgoesbeyondbasicdatacollection.Itguidessuppliersthroughstep-by-step

onboardingprocessescustomizedtoeachorganization’srequirements.Thiscon?gurabilityensurescompliance

withspeci?ccompanypolicieswhilecreatingconsistentexperiencesacrossallsupplierrelationships.

Thesystembene?tsboththebuyingorganizationandsuppliersbystreamliningwhattraditionallyrepresentsacumbersome,manualprocess.

keysuggestions

?StartagenticAIimplementationwithhigh-volume,repetitiveprocesseslikesupplieronboarding.

TheseareasprovideclearROIwhilebuilding

organizationalcon?denceinagentcapabilities.

?Focusonorchestrationcapabilitiesthatconnectmultipleprocurementfunctionsseamlessly.End-to-endautomationdeliversmorevaluethanpointsolutionstargetingindividualtasks.

“InourAI-generatedsummaryofasourcing

event,twosupplierssubmittedtheirbids,andthesummaryshowswhichbidsofferthebestsavings.Thesystemalsosuggeststhenext

beststepsbecausetheorchestrationagentunderstandstheoverallgoal:togetthebestpriceforthebestproducts.”

AndrewTumath

GEP

Sponsoredby:9GEP9

procureconInsIGHTS

procurementsupercharged:whathappens

whenagenticaiorchestratessource-to-pay

thefutureofai-drivenprocurement

Thefutureofprocurementliesinincreasinglysophisticatedagenttypesthathandleprogressivelycomplex

responsibilities.Fivedistinctagentcategorieswillshapethisevolution:

?Re?exAgents:handlebasicprocurementinquiriesandautomatedresponsessimilartocurrentRPAbots.

?Model-basedAgents:useapplicationdatatomonitorsupplierrisksanddetectspendinganomalies.

?Goal-basedAgents:focusonspeci?coutcomeslikestrategicsourcingeventsorsupplierdiscoverybasedonde?nedcriteria.

?Utility-basedAgents:makesophisticateddecisions

consideringmultiplevariablesandconstraints,suchasoptimizingsourcingdecisionsacrosscost,quality,andriskfactors.

?LearningAgents:forcontinuouslyimprovingperformancethroughaccumulatedexperience.

Learningagentsrepresentthemostadvancedcategory,buildingcollectiveintelligencethatimprovesovertime.

Theseagentsanalyzenegotiationpatterns,supplier

performancetrends,andmarketdynamicstoidentifyopportunitiesandrisks.

Theycandetectfraudulentactivities,predictmarket

changes,andrecommendstrategicimprovementsbasedonaccumulatedexperienceacrosstheentireplatform.

Sponsoredby:9GEP10

procureconInsIGHTS

procurementsupercharged:whathappens

whenagenticaiorchestratessource-to-pay

thenextevolutionin

BEFORE?TRADITIONALPROCUREMENT

ai-drivenprocurement

?SlowerProcurementCycles

?LowerEfficiency&Productivity

?IncreasedRisk&Gut-BasedDecision-Making

?LimitedScalability&Short-TermViability

?DelayedKnowledgeProcessing

?Fragmented&InefficientUserExperience

Thevisionextendsbeyondcurrentcapabilitiestowardtrulyautonomousserviceecosystems.

Notably,thisevolutionwon’teliminatehumanrolesbut

ratherreshapethemtowardhigher-valuestrategicactivities.Thesesystemswillanticipateuserneedsbeforethey’re

explicitlystated,proactivelymanagingsupplierrelationshipsandmarketopportunities.

AFTER?NEW-AGEAGENTICAI

?FasterProcurementCycles

?HigherEfficiency&Productivity

?ImprovedRisk&Data-DrivenDecision-Making

?Future-ProofScalability

?AcceleratedKnowledgeProcessing

?SeamlessUserExperience

Organizationsmustprepareforthistransitionbyidentifyingrepetitivetaskssuitableforagentdelegationwhile

maintainingstrategicoversight.AItechnologyfaces

ongoingchallengeslikebias,hallucinations,andincorrectdecision-making.However,itstrajectoryinprocurement

suggeststhefunctionwillbecomeincreasinglyautomated.

keysuggestions

?Identifyyourtop?verepetitivetasksthatcouldbedelegatedtoagentstoday.Thisexercisehelpsprioritizeimplementationareaswiththehighestpotentialimpact.

?Planfortheevolutionofagentcapabilitiesratherthanviewingcurrentimplementationsasstaticsolutions.Technologyadvancementwillcontinuouslyexpandwhatagentscanaccomplishautonomously.

“Wedon’twanttosuggestthatAIagentscan?xeverythingimmediately.That’ssimplynottrue.

AI—includingagenti