1IiThe GenAl DivideSTATEOFAINBUSINESS2025MITNANDAAdityaChallapallyChrisPeaseRameshRaskarPradyumnaChariJulv 2025
NOTESPreliminary Findings from AlImplementation Research from Project NANDAReviewers: Pradyumna Chari, Prject NANDA,Methodology: This report is based on a multi-method research design that includesasystematicreviewofover300publicly disclosed Alinitiatives.structuredinterviews with representativesfrom 52 organizations, and survey responses from153seniorleaderscollectedacrossfourmajorindustryconferences,Disclaimer.Theviewsexpressed in thisreport aresolelythose ofthe authors andreviewersanddonotreflectthe positions of anvaffliated emploversConfidentiality Note: All companv-specific data and auotes have beenanonymizedtomaintaincompliancewithcorporatedisclosurepoliciesandconfidentiality agreements, ensure neutrality, and prevent anv perceotion ofcommercial advancement or opinion
1ConTENTS1.ExecutiveSummary2.The WrongSideofthe GenAlDivide:HighAdoptin.Low Transformation3CrossinatheGenAlDivide:HowtheBestBuildersSucceed5Crossing theGenADivide.Howthe BestBluyersSuced6. Conclusion: BridgingtheGenAlDivide2EXECUTIVESUMMARYDespite $30-40 bllion in enterprise investment into GenAl, this report uncovers a surprisingresult in that gs%oforaanizations are aettina zero return. The outcomesaresostarklydivided across both buyers (enterprises, mid-market, SMBs) and builders (startups, vendorsconsultancieslthat we callit theGenAl Divide.Just 5% of integrated Alpilots areextractingmillions in value, while the vast majority remain stuck with no measurable P&L impact.Thisdivide does not seem to be driven by model quality or requlation, but seemsto bedetermined by approachToobslkeChatGPT and Copilot are widely adopted Over 8o poercent of oranizations have40dercent reoortdeolovment.Butthesetoolsprimarilyenhance individual productivity, not P&Lperformance.Meanwhile.enterorisegrade systems, custom or vendor-sold, are being quiety rgected Socty percentoforganizations evaluated such tools, but only 20 percent reached pilot stace and justspercentreached production. Mostfalldue tobritte workflows lack of contextualleamingand misalianment with dav-to-davoperationsFrom our interviews. survevs. and analysis of 300 public implementations. four patternsemeraed thatdefinetheGonAl Dividla Limited disruption: Only 2 of 8 maijor sectors show me aningful structural cthangeEnterprise paradox: Big firms lead in pilot volume but laq in scale-upInvestmentbias:Budaetsfavorvisible. top-line functionsoverhiah-ROI back officeThe core barrier to scaling is not infrastructure, regulation, ortalent. It is learning. MostGenAlsvstemsdonotretainfeedback, adapt tocontext, or improve overtimeAsmallarouoofvendorsandbuyersare achievingfasterprogress by addressing theselimitations directly. Buyers who succeed demand process-specific customization andevaluate tools based on business outcomes rather than software benchmarks. They expedpg.a
systems thatintegrate withexisting processes and improve over time. Vendors meetingtheseexpectations are sec...
