McKinsey senior partner: Why AI’s easiest wins are misleading CEOs

McKinsey found AI adoption was nearly flat at 89% of organizations versus 88% a year earlier, while only 6% qualified as high performers generating at least 5% of EBIT from AI. Although AI improved customer-support productivity by 15% and software-developer task completion by roughly 26% in cited studies, only 37% of organizations report any positive earnings effect. Organizations that redesigned workflows were 5.3x more likely to report enterprise-level AI value (32% versus 6%), underscoring that process redesign and accountable ownership—not tool deployment alone—drive returns.

Analysis

The investable distinction is shifting from AI seat deployment to measurable workflow ownership. Near term, this challenges the market’s tendency to capitalize broad enterprise AI adoption into software valuations: vendors with usage-driven pricing but weak evidence of customer ROI face longer sales cycles, elevated churn risk at renewal, and procurement scrutiny. The most exposed cohort is horizontal copilots and generic application-layer vendors; infrastructure demand remains more durable because experimentation still consumes compute even when pilots fail to scale.

Over the next 1-3 months, prioritize companies where AI is embedded in a workflow with closed-loop measurement, proprietary data, and a clear economic buyer. CRM (Salesforce), service management (ServiceNow), contact-center platforms (Genesys/privately held; NICE), and vertical software providers can monetize AI when it either raises throughput or reduces regulated error rates within an existing system of record. The second-order winner is implementation and data-governance spend: Accenture, Cognizant, IBM and Snowflake/Databricks-linked ecosystems may capture more of the budget than standalone model vendors as customers discover integration—not model access—is the bottleneck.

The 6-18 month upside is concentrated in firms able to redeploy released labor into revenue capacity rather than merely report theoretical productivity. Financial services, insurance, and healthcare offer the largest dollar pools but also the slowest conversion because model-risk controls, auditability, and fragmented decision rights constrain automation. The contrarian view is that weak enterprise ROI does not invalidate the AI capex cycle; it extends it, shifting value from first-wave model and chip beneficiaries toward systems-of-record, workflow redesign, and governance vendors. This thesis fails if enterprise software renewals show broad AI SKU attach-rate acceleration without associated services/data spend, or if hyperscalers materially reduce AI infrastructure capex due to utilization shortfalls.

Key Decisions for Investors

• Prefer a 6-12 month long CRM or NOW versus a short basket of high-multiple, subscale horizontal AI application vendors: systems-of-record have distribution and workflow data, while generic copilots face feature commoditization. Size only after checking AI attach rates, net retention, and deferred-revenue commentary in the next earnings cycle.
• Accumulate ACN and IBM on AI-related guidance resets rather than chase infrastructure beta. Risk/reward depends on bookings converting to revenue within 2-4 quarters; exit if consulting utilization falls while AI bookings rise, indicating low-margin pilot work rather than scaled transformation.
• Watch NICE as a cleaner public proxy for proven contact-center automation. Initiate only if management demonstrates AI revenue contribution alongside stable gross margin and customer retention; the key risk is hyperscaler/platform bundling compressing contact-center software pricing.
• Avoid treating broad AI adoption announcements as a catalyst for enterprise-software multiple expansion over the next quarter. Require evidence of paid production deployments, reduced customer acquisition friction, or higher net revenue retention; absent those metrics, favor quality software with durable non-AI growth.