Two-thirds of organizations report productivity gains from AI. Only one in five is seeing revenue growth [1]. That gap tells you everything about where AI transformation actually stalls.
Deloitte surveyed 3,235 leaders for their State of AI in the Enterprise 2026 report. Seventy-four percent hope to grow revenue through AI. Twenty percent are doing it [1]. The remaining 54 percentage points represent the gap between AI capability and the organization's capacity to absorb it. That distance has a name in logistics, telecommunications, and now enterprise transformation: the last mile.
The last mile is not a technology problem. It is an operating model problem.
Pilot Rich, Transformation Poor
Lakhani, Spataro, and Stave describe the pattern precisely in their March 2026 HBR analysis. Organizations launch hundreds of AI pilots. Many succeed. Gains fail to translate into standard operations [2]. The authors call this being “pilot rich but transformation poor,” a phrase that captures the structural dysfunction with uncomfortable accuracy.
One global investment bank reported 250 or more LLM applications in production. A global payments network achieved 99% employee adoption of Copilot. An asset-servicing firm currently runs over 100 agents and plans to scale to tens of thousands [2]. These are not technology failures. These are organizations where individual productivity improvements vanish before reaching the balance sheet.
Gartner predicts 40% of enterprise applications will include task-specific AI agents by the end of 2026, up from less than 5% in early 2025 [3]. In the same breath, Gartner forecasts 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 [4]. Read those two predictions together. The technology is arriving. The organizational capacity to use it is not.
Seven Obstacles, One Root Cause
The HBR analysis identifies seven structural frictions preventing AI from crossing the last mile [2]. Each one is organizational, not technical.
Proliferation of pilots without integration pathways. Productivity gaps in which individual efficiency gains are absorbed into low-value activities rather than showing up on income statements. Process debt from decades of exception-laden workflows that resist automation. Tribal knowledge hoarding, where employees fear externalizing expertise because it diminishes their organizational identity. Governance collapses when traditional oversight models face autonomous agents operating at machine speed. Architectural complexity across multi-vendor AI systems that cannot communicate reliably. And the efficiency trap, where framing AI purely as cost reduction triggers defensive behavior and constrains the scope of transformation.
Notice the pattern. Not one of these obstacles is about model quality, data availability, or compute capacity. Every single one is about how the organization is designed: how decisions flow, how knowledge is captured, how governance scales.
In my article on Organizational Complexity as Decision Debt, I showed that two-thirds of leaders admit their organizations are too complex and inefficient to execute effectively. That complexity is what creates the last mile. AI doesn’t simplify it. AI accelerates it.
The Operating Model Gap
Deloitte’s readiness numbers tell the structural story. Talent readiness sits at 20%. Governance readiness trails at 30%. Only 34% of organizations are redesigning products, services, or business models around AI [1]. Three-quarters plan to deploy agentic AI within two years, yet only 21% report having a mature governance model for autonomous agents [1].
Those numbers reveal a specific failure mode. Organizations built to ship features through sequential approval chains cannot absorb AI agents that operate across functional boundaries at speeds that make weekly governance meetings irrelevant. The operating model was designed for a different cadence.
Clayton Christensen’s RPV framework explains why this is self-reinforcing. In my January article on that framework, I showed how organizational capabilities migrate from Resources (people, tools) to Processes (coordination patterns) to Values (prioritization criteria). AI adoption gets treated as a Resources problem: hire talent, buy tools, fund pilots. The failure happens at the Processes and Values level. Existing coordination patterns reject the transplant because they were optimized for human-speed, sequential workflows.
The last mile is the distance between resource acquisition and process redesign.
Crossing the Last Mile
Lakhani, Spataro, and Stave propose seven counter-measures, and the most instructive is clean-sheet process redesign [2]. Not optimizing existing workflows with AI bolted on. Rebuilding workflows from scratch with AI agents as first-class participants.
This connects directly to decision infrastructure. In my February article on Decision Velocity Infrastructure, I argued that the speed and quality of decisions determine organizational throughput. The last mile collapses when three conditions are met.
First, decision rights get redesigned for agent speed. Traditional governance assumes human review at every checkpoint. Agent governance requires defining boundaries within which autonomous systems operate, with human oversight at the boundary, not at every step. Second, knowledge gets externalized systematically. The tribal knowledge obstacle disappears when organizations treat knowledge capture not as documentation but as strategic infrastructure. Codified judgment becomes a reusable asset instead of an individual’s job security. Third, value framing shifts from cost reduction to capability creation. The efficiency trap springs shut when AI is positioned as a way to do the same things cheaper. It opens when AI enables work patterns that were structurally impossible before.
These are not technology implementations. They are operating model redesigns. Product Operating Models, specifically, because they orient the organization around outcomes rather than outputs, around customer problems rather than feature delivery, and around adaptive governance rather than sequential approval.
The Structural Question
Gartner says only about 130 of the thousands of agentic AI vendors offer genuine capabilities. The rest are engaged in “agent washing,” rebranding existing products without substantive change [4]. That vendor reality mirrors the organizational reality. Many companies are relabeling existing practices as “AI-enabled” without redesigning the structures through which those practices run.
The last mile will not be crossed by better models, bigger budgets, or more pilots. It will be crossed by organizations willing to redesign how decisions flow, how knowledge compounds, and how governance adapts to agent speed. The technology works. The operating model does not. Closing that gap is the transformation.
Ralph Jocham is Europe’s first Professional Scrum Trainer, co-author of “Professional Product Owner,” and contributor to the Scrum Guide Expansion Pack. As an ICF ACC certified coach, he works with organizations to build Product Operating Models where strategic clarity, operational excellence, and adaptive learning create measurable competitive advantage. Learn more at effective agile.
References
[1] Deloitte, “State of AI in the Enterprise 2026: The Untapped Edge,” Deloitte, 2026. https://www.deloitte.com/us/en/what-we-do/capabilities/applied-artificial-intelligence/content/state-of-ai-in-the-enterprise.html
[2] Lakhani, K.R., Spataro, J., and Stave, J., “The Last Mile Problem Slowing AI Transformation,” Harvard Business Review, March 2026. https://hbr.org/2026/03/the-last-mile-problem-slowing-ai-transformation
[3] Gartner, “Gartner Predicts 40% of Enterprise Apps Will Feature Task-Specific AI Agents by 2026,” Gartner Newsroom, August 2025. https://www.gartner.com/en/newsroom/press-releases/2025-08-26-gartner-predicts-40-percent-of-enterprise-apps-will-feature-task-specific-ai-agents-by-2026-up-from-less-than-5-percent-in-2025
[4] Gartner, “Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027,” Gartner Newsroom, June 2025. https://www.gartner.com/en/newsroom/press-releases/2025-06-25-gartner-predicts-over-40-percent-of-agentic-ai-projects-will-be-canceled-by-end-of-2027