Outlast the Hype with AI That Lasts: The Next Competitive Edge Delivers ROI at Scale (and Keeps Getting Smarter) By Jocelyn Sexton, GAP’s VP of Marketing
AI IS NO LONGER A PROJECT — IT’S AN EVOLVING SYSTEM Artificialintelligencehasmovedbeyondnovelty andreachedapivotalinflectionpoint.Whatwas onceaboldexperimenthasnowbecomea businessimperative.Formanyorganizations,it nowfunctionsasalayerofinfrastructurethat mustbedesignedwiththesamerigorasany enterprisesystem.Andifbuilthurriedlyor withoutgovernance,itquicklybecomes unstableanddifficulttomaintain.
LeadersataNovember2025Technology
ExecutivesRoundtableinNewYorkCityhosted byGrowthAccelerationPartners(GAP)agreed thatthepathforwardrequirestreatingAIasan adaptivearchitecture,notanisolatedinitiative.
Attheevent,CTOs,engineeringexecutivesand datagovernanceleadersgatheredtoexplore threebigquestions:
1. The Problem — What’s holding back AI investments?
2. The Challenge — What does it take to build AI that lasts?
3. The Solution — How do we engineer adaptive systems that continue returning value over time?
Throughoutthediscussion,participants describedthesametension—theneedtoshow resultsquicklywhilealsobuildingfoundations sturdyenoughtoevolve.Theconversation revealedashiftfrom“HowfastcanwebuildAI?” to“HowdowebuildAIthatdeliversmeaningful valuenotjustinmonthone,butinyearthree?”
This article captures patterns from GAP’s Roundtable event: what works, what doesn't, and how to build AI systems that are scalable and sustainable. It explores how organizations can build adaptive, production-grade AI that delivers continuous ROI through governance, scalability and feedback. It also reframes ROI beyond cost savings, emphasizing trust, adaptability and knowledge retention as measures of success.
THE PROBLEM THE ROI DILEMMA: HYPE, PRESSURE AND MISALIGNED EXPECTATIONS ThehypephaseofAIisover.Whatremainsis anexpectationthatAIwilldeliverbusiness impact,yetmanyorganizationsstillstruggleto provesustainedROI.Quickwinslike automatingmanualtasksorreducingticket backlogsarehelpful,buttheyrarelyproducea durableadvantage.
Leadersdescribedfeelingtornbetweenproving AI’svaluenowandensuringtheworkwon’tfall apartlater.Oneexecutivesummarizedthe challenge:
“It’s about what we can benefit from today, with an eye toward what lasts.” Whilesomeindustriesreportencouraging progress shorterbuyingcycles,faster workflows,reducederrors—othersrevealstark gapsinmaturity.Aparticipantpointedtoa surveyshowingfewerthan1%ofenterprises scoringabove50ona100-pointAImaturity scale.
AIdemandsdiscipline,governanceanda forward-lookingviewofstrategicvalue.One technologyleaderdescribeditthisway:“The firstwaveofAIprojectsoftenfeltlike adrenaline—quick,exciting,butunsustainable. Thesecondwavemustfeellikearchitecture.”
Also,ROIinAIdoesn’talwayslooklikeasimple costreduction.Sometimesit’stimereclaimed, talentunlockedorknowledgepreserved.These softermetricsrelatedtoefficiency,trustand adaptabilitycanbecomeanadditionaltrue currencyoftransformation.
Traditionalmetricsstruggleto captureAI’strueimpact.Costsaving istheeasiestROItomeasure,but long-termimpactshowsupinthree ways:
Trust –
Trust: Do people rely on the system’s output?
Adaptability –
Adaptability: Can the system evolve with the business?
Knowledge Retention –
Knowledge Retention: Are we reducing single-expert dependency?
OneparticipantsharedhowAI reduced“retirementrisk”fortheir legacysystems.“Oursenior engineersusedtobetheonlyones whounderstoodcertain components.Nowanyonecanquery AIforcontext.Itreduced dependenceonsingleexpertsby 30%,”hesaid.
Anothernotedthatimprovedtrust oftenleadsdirectlytoimproved efficiency.WhenAIsystems consistentlyprocessclaimsoremails withhighaccuracy,“employeesget tofocusonthecomplexworkthey werehiredfor,”theleadersaid.
Thesesofterreturns—shared knowledge,increasedtrustandmore engagedteams—areincreasingly recognizedascoretolong-term transformation.
“Value creation is what we can measure today,” the participant said. “Value building is what we’re doing for tomorrow
— cultivating trust in the results and building a healthy backlog of ideas that feed future innovation.”
Curiosity,fearandthepaceofchangewere bigtopicsduringthediscussion.Despiteits technicalcomplexity,AI’stoughest challengeremainshuman.Leaders repeatedlyreturnedtothesametheme changemanagementistherealbarrier.
“Peopleweregenuinelyfrightened,”one R&Dleaderadmitted.“WeintroducedAIto help,butformanyitfeltlikeathreat.It takespatience…holdinghands,notjust writingpolicies.”
Mandateslike“useAIorriskbeingleft behind”rarelyinspireadoption.Instead, theycreateanxietyandresistance.The mostsuccessfulorganizationscultivatea cultureofcuriosityandexperimentation ratherthancompliance.
PATTERNS IN SUCCESSFUL AI CULTURES INCLUDE: TheRoundtableparticipantsagreedweall needleaderswhosharestories, demonstrateresultsandsparkenthusiasm acrossdepartments.It’sagrassroots approachthattransformscomplianceinto creativity.Theleaderssucceedinginthis spacesharerealexamples,celebratequick winsandmakeAIfeelaccessible.They treatadoptionasaculturaljourney,nota compliancerequirement.
Curiosity,onceconsideredasoftskill,has becomeahardrequirement.Thosewilling toaskbetterquestions—aboutdata, outcomesandprocesses aretheones shapingthenextphaseofbusiness intelligence.Asautomationabsorbs repetitivework,curiositybecomesthenew jobsecurity.
“The biggest risk isn’t that people can’t learn AI,” one participant observed. “It’s that they stop imagining what’s possible.”
THE CHALLENGE GOVERNANCE: THE FOUNDATION FOR AI THAT LASTS Asenterprisesmovebeyondproofsofconcept, governancebecomesthedefiningdisciplineof sustainableAI.It’snolongeroptional.It’s necessaryduetorisk,complianceandtheneed forreliability.
AIgovernanceisemergingasa multidisciplinaryeffortspanninglegal,risk,HR andtechnology.Somecompaniesnowoperate cross-functionalAIcouncilsthatoversee everythingfrommodelregistrationtoethics reviewsandlicensemanagement.Manyhave introducedsandboxenvironmentsforsafe experimentation,ensuringfreedomwithout compromisingcompliance.Otherrobust governanceframeworksnowincludepolicies fordatausageandmodelregistration.These systemsdon’texisttoslowinnovation,butto ensureithappenswithaccountability.
“Governancedefineshowwebuild,”saidone moderator.“Feedbackdefineshowweevolve. Together,theytransformdisconnectedprojects intocohesiveecosystemsthatcanadaptasfast asthetechnologyitself.”
Continuousmonitoring,automatedretraining anduser-in-the-loopfeedbackarethe operationalbackboneofanyproduction-grade system.Youalreadyknowdriftisn’ttheoretical. It’sinevitable,andevenminordeviations compoundquicklywhenmodelsareembedded inpricing,routing,riskscoringorclinical workflows.
Theorganizationsthatstayaheadtreatdriftasanengineeringproblem,notanafterhoursdashboardexercise.Theyinstrumentmodelsthesamewaytheyinstrument distributedsystems withtelemetry,alertingthresholdsandauditablepathwaysback tothedataandfeaturesthatshapedeachprediction.
Thegoalisn’tjusttokeepmodelsonline;it’stokeepthemalignedwithreal-world behavior.Thatrequiresaninterplaybetweenautomatedpipelinesandhuman oversightthatcanadjudicateedgecasesandhigh-impactexceptions.Whenthis feedbackloopishealthy,thesystembecomesadaptiveinthetruestsense continuouslycorrecting,recalibratingandimproving.ThisishowAImovesfroma point-in-timedeploymenttoalivingarchitecturethatadjustsasfastastheconditions arounditchange.
GAP’sCTOPaulBrownellalsoemphasizedtheriskofconfusingspeedwithvalue. “Measuringsuccessofoutputs ofmodelaccuracy isreallymeasuringwhetherAI isrobustandenterprise-quality,”hesaid.“It'snotreallyaboutmeasuringthevalueAI givestothebusiness;it’showvaluableAIistoaccelerateyourcarfasterifitispointed inthewrongdirection?Theengineering,designandoperationofAIsystemsmust focusonsuccessfuloutputs.Becausewhocareshowmuchfasteryoucangenerate inadequateresults?”
Brownell’spointunderscoredashiftinmind-set fromcelebratingoutputvolumeto prioritizingquality,contextandrelevance.
THE SOLUTION BLUEPRINTS FOR ADAPTIVE AI: PRAGMATIC, SCALABLE SYSTEMS ParticipantsagreedthatsustainableAIbeginswithsimplicity. Anditmaysurprisesometoknowtheleadersatthe Roundtableoftenrepeatedanow-familiarlinefromtheevent:
AI is not the solution to every problem or opportunity. Themosteffectiveorganizationsaretakingameasured approachtoscalability.They’relearningthatsustainableAI startswithsimplicity.“Ifyoucandoitonpaper,don’tuseAI,” oneparticipantsaid,capturingtheemergingmindset.
AIshouldenhancesoundengineeringjudgment,notreplace it.Ifanoldermethodsolvestheproblemwell,useit.IfAIis required,useonlywhat’snecessary.Overbuildingmakes systemsexpensiveandfragile.
Manyteamsarenowdiscoveringthat models outperformsprawlingnetworksoflooselygovernedones.As oneattendeenoted,“Maybewedon’tneed10models.Maybe fourorfive—donewell—areenough.”Scaling,inthissense, isnotaboutmultiplyingmodels,butmultiplyingimpact. fewer, better
PRINCIPLES OF SUSTAINABLE SCALABILITY: Understand data lineage its age, version and governance tier
Set thresholds and checkpoints to prevent scope drift
Use small, expert teams to validate ideas before scaling
Bring in external experts to accelerate discovery and reveal blind spots
Thismeasuredapproachprotectsteamsfrom racingaheadwithoutclarityordiscipline.
Oneparticipantdescribeditwell:
“The basics matter. We can’t lose high-level discipline just because the tools make it easy to move fast. So we must stay fast, but not reckless.”
Additionally,scopecreepisconstantinAI. Objectivesshiftasdata,usersandconstraints evolve,andtheoriginalproblemstatementoften becomesstalebeforethesystemships.The answerisn’ttoeliminatescopechanges—it’sto managethemwithdiscipline.Asoneparticipant putit,“Itwillkeepchanging.Thegoalisn’tto preventit,buttocontrolitintelligently.”
Theteamsnavigatingthiswellanchor everythinginclarityofpurpose.Theyrelyon small,senior“SWATunits”thatvalidateideas quickly,killweakassumptionsearlyandonly scalewhatprovesout.It’samoveawayfrom big-projectthinkingandtowardoutcome-first engineering—makingsureeveryone understandswhysomethingshouldexistbefore committingresourcestohowitgetsbuilt.
Externalexpertisecanacceleratethismaturity curve.Therightpartnerssurfaceblindspots, pressure-testreadinessandforcehard conversationsaboutdataquality,architectural debtandrealvaluedrivers.Wheninternaland externalperspectivesalign,progress compounds.
Underneathallofthisistherealmoat:talent. Toolsareubiquitous,buttheabilitytoaskthe rightquestions,simplifythecomplexandpush fordisciplinedexecutionisnot.Asoneleader said,“Thebiggestmoatistalentplusambition.” Curiosityandintentarewhatseparatedurable AIsystemsfromnoise.
Production-grade requires a lifecycle of transparency, monitoring and human judgment.
WhenleadersattheRoundtabletalkedabout“production-gradeAI,”theyweren’t describingatechnicalmilestone.Theyweredescribinganoperationallifecycle. Production-gradeAIisexplainable,monitoredandcontinuallyretrained.Itsurfaces biasratherthanhidingit.Itincludesguardrailsthatpreventmisuseanddrift.
“Explainability doesn’t fix bias — it reveals it,”
“Explainability doesn’t fix bias it reveals it,” one expert said. “That visibility is what builds trust.”
Thegroupdiscussedrisingconcernsaboutadversarialbehaviorandmisinformation embeddedintrainingdata.Oneparticipantnoted,“Badactorsarealreadyteaching modelsthewronglessons.Feedbackloopscanbegamed.”Robustgovernance,data lineageandtheabilitytodeactivatecompromisedpipelinesquicklyarecritical safeguards.
Thisreinforcedtheimportanceofgovernancestructuresthatarenotjuststrong,but adaptable.Ultimately,production-gradeAIisnotafinishlinebutafeedbackloop.It’sa systemdesignedtoimprovethroughuse,tolearnresponsiblyandtostayaccountable.
A PRACTICAL, REPEATABLE FRAMEWORK EMERGED FROM THE ROUNDTABLE: Frame the why, then size the fix
Frame the why, then size the fix.
Define success without AI If a paper process or a SQL view solves it, do that first — then add models where they create disproportionate value.
Architect for context.
Architect for context Version data and models Stamp outputs with lineage, policy tier and freshness so users understand limits.
Design for change.
Design for change. Expect scope to evolve Limit model count Wire monitoring for drift, abuse and business KPIs not just model metrics.
Close the loop with humans.
Close the loop with humans. Put SMEs in the review gates that matter Treat their time as a lever, not a cost
Accelerate with allies.
Accelerate with allies. Bring in external specialists to compress discovery not to own your strategy Keep governance and IP tight
Invest in the posture, not just the platform.
Invest in the posture, not just the platform. Hire for curiosity. Reward simplification. Celebrate teams that decommission models when the simple answer wins
a growing recognition that not every problem needs an AI solution. experiment quickly, and implement thoughtfully.”
THE SOLUTION AIisevolvingintoaninterconnected ecosystemofexperimentation,feedback andcontinuouslearning.Theorganizations positionedtoleadthisfuturesharea commonDNA:
Vision with discipline —
ambition paired with intentional guardrails
structures that protect creativity
Technology with humanity —
Governance with flexibility — systems that amplify imagination, not replace it
NeartheendoftheRoundtable,one attendeesummeditupbeautifully:
“AI isn’t here to replace what people do — it’s here to expand what they can imagine.”
Theorganizationsthatwilllastaretheones thatholdbothtruthsatonce:buildsystems thatlearnandadapt,andbuildculturesthat dothesame.
Thecompaniesthatthrivewillbe thosebuildingsystemsandcultures thatevolvetogether responsibly, sustainablyandwithpurpose.
Thehypewillfade,butthe intelligenceecosystemsorganizations buildtodaywilldefinethenext competitiveedge.TrueAI transformationdoesn’tcomefrom scalingmodelsfaster,butfrom engineeringtransparency, accountabilityandsimplicityinto everylayerofthesystem.
AtGrowthAccelerationPartners,we helpenterprisesnavigatethisshift withframeworksthatsupport adaptive,responsibleAI acceleratingexperimentationwhile ensuringenduringimpact.Bybuilding systemsthatlearnwitheveryiteration, organizationscandelivercontinuous ROIandcreateintelligencedesigned tolast.
