Enterprise AI Transformation Programmes (2025–2026)

Narrative

The single most contested data point of the period is MIT NANDA's claim, popularised by Fortune in August 2025, that about 5% of AI pilot programs achieve rapid revenue acceleration; the vast majority stall, delivering little to no measurable impact on P&L. The study's methodology, based on 52 executive interviews, surveys of 153 leaders, and analysis of 300 public AI deployments, has drawn sustained pushback: critics note the report has a narrow definition of success and that a closer look at the study's methodology reveals a very different picture once measurement criteria are examined. McKinsey's competing 2025 State of AI survey, covering over 1,900 respondents, complicates the binary failure narrative further: nearly nine out of ten survey respondents say their organizations are regularly using AI, though most have yet to scale the technologies, with the share using AI in at least one function increasing since last year, and management practices span six dimensions essential to capturing value from AI: strategy, talent, operating model, technology, data, and adoption and scaling, all correlating positively with value attributable to AI.

Bain's independently run quarterly tracker offers a third data point that neither confirms catastrophic failure nor unambiguous success. Its 951-company Automation and AI Pathfinder survey found while 37% targeted cost reductions of 11% to 20%, nearly 40% of those who measured outcomes landed in the 0% to 10% bucket instead, yet 90% of those same companies are now increasing their budgets again for agentic AI, funded circularly: 44% of companies cited savings from prior automation programs as the largest funding source, which sounds like discipline but is a circular bet with a structural leak. Bain's separate executive tracker paints a more optimistic production picture, finding executives say that 80% of generative AI use cases met or exceeded expectations, but only 23% can tie initiatives to new revenue or lower costs, and that adoption growth across domains is far more rapid in just three years than anything seen in previous technology waves. Gartner's parallel warning that over 40% of agentic AI projects will be canceled by the end of 2027, due to escalating costs, unclear business value or inadequate risk controls has itself become a recycled statistic: a Forbes piece a year later noted the coverage routinely presents the figure as new, while the underlying cause is management discipline, not model capability, since Gartner's June 2025 analysis named escalating costs, unclear business value, and inadequate risk controls, with model capability notably absent from the list.

On delivery approaches, the ThoughtWorks Technology Radar's April 2026 edition (volume 34) marks a shift from proliferation to consolidation, warning that as agentic systems make it easier to create code quickly, traditional and established practices that ensure discipline and rigor are more vital than ever, with a significant need for humans to proactively implement appropriate practices and technical harnesses, and observing a return to well-established techniques like zero trust architecture, DORA metrics, and testability to manage complexity. Forrester's 2026 predictions echo this convergence of speed and control, forecasting that enterprises will delay 25% of AI spend into 2027, as only 15% of AI decision-makers reported an EBITDA lift, while its June 2026 State of Agentic AI report finds three-quarters of enterprises are adopting agentic AI, but only a small minority have it running in meaningful production beyond 'agentish' chatbots, and recommends organisations treat every agent as a governed identity, giving it unique credentials, least privilege, full logging, and a named owner who manages its lifecycle.

Token economics and governance frameworks are converging around similar tensions between velocity and control. a16z's enterprise data shows model diversification accelerating cost complexity:

Yipit's panel data of ~1,000 mid-market and enterprise companies shows adoption rates that closely mirror our findings, with OpenAI around 85% and Anthropic near 55% and rising, though OAI still commands a majority wallet share at ~56%, and 81% now use three or more model families in testing or production, up from 68% less than a year ago. On regulation, the EU AI Act's phased implementation has itself become a case study in control-versus-velocity trade-offs: after industry pressure, the European Commission published the Digital Omnibus on AI on 19 November 2025, proposing to defer the high-risk compliance deadline from 2 August 2026 to 2 December 2027, while standards bodies caution that the regulatory context has grown more complex, with the Commission proposing in its Digital Omnibus package to delay Annex III compliance deadlines to December 2027, citing the late arrival of harmonized standards. On the NIST/ISO governance stack, practitioners increasingly frame the two as complementary rather than competing:

NIST AI RMF suits organisations whose audience is the engineering organization or whose regulatory exposure is U.S.-centric, while ISO 42001 suits organisations whose audience includes procurement, customers, or regulators seeking third-party assurance, since ISO 42001 is the path to a certificate and NIST AI RMF is the path to a self-attestation document.