Yale’s Chief Executive Leadership Institute published the final installment of a four-part research series on Fortune today analyzing agentic AI deployment strategies across 13 industries. The central finding: the deployments generating the most durable returns are the ones customers never see.
The framework, built from conversations with senior technology leaders across those industries, classifies agentic AI deployments into three proximity categories based on how directly they affect the customer experience: direct, mediated, and background. Background operations consistently delivered the highest ROI with the lowest risk to customer relationships.
The Numbers Behind Background Deployments
The case studies are specific. C.H. Robinson’s 30-agent system handles over 318,000 tracking updates per month and responds to 100% of inbound carrier requests, up from 60% before automation, according to Fortune. The company now processes roughly one-third more freight with roughly one-third fewer employees than in 2019.
Unilever is targeting $800 million in savings by 2026 from its agentic AI program, anchored by a digital twin of its global supply chain that simulates disruptions and autonomously triggers logistics responses. Cohere Health reports 90% of prior authorizations now automated, with 96% approved in seconds and 47% administrative cost savings.
McKinsey projects $450 to $650 billion in additional annual revenue across advanced industries by 2030 from agentic AI, with 30 to 50% cost reductions through operational automation.
Customer-Facing Deployments: High Visibility, High Risk
The contrast with direct-proximity deployments is stark. The 2025 National Customer Rage Survey found that 88% of e-commerce customers who believed they had interacted with AI viewed the experience unfavorably, according to Fortune.
Consumer complaints filed with the CFPB nearly doubled over two years, from about 770,000 before ChatGPT’s public launch to over 1.5 million after, with the increase concentrated at firms with high AI exposure. Low-exposure areas that retained human-in-the-loop operations, such as mortgage and student loans, saw flat complaint volumes.
The framework’s opening example: Cursor’s AI customer support agent “Sam” fabricated a licensing policy in April 2025 that triggered cancellations and complaints on Hacker News before the company discovered the error.
The Middle Path: Mediated Deployments
The framework’s mediated category covers deployments where agents work alongside human employees. Thomson Reuters’ CoCounsel, deployed across 20,000+ law firms including the majority of the Am Law 100, returns an estimated 110% over three years. EY’s EY.ai platform deploys 150 AI agents supporting 80,000 tax professionals processing more than three million tax deliverables annually.
TELUS reports 57,000 employees using AI tools save approximately 40 minutes per customer interaction. The risk: because the agent is invisible to the customer, failures get attributed to the human or the firm, making recovery harder than for a visible chatbot failure.
Gartner’s Cancellation Forecast
Gartner projects that over 40% of current agentic AI projects will be canceled by the end of 2027, not because the technology failed, but because firms deployed without the governance scaffolding to catch errors before they compounded.
The Deployment Sequence
The framework’s prescription is explicit: deploy quickly where the agent is invisible, deploy deliberately where customers can feel the system, pause deployment where harm from a single failure is high and reversibility is low. The research concludes that the firms leading durable returns today are not those with the most sophisticated models. They are the firms that mapped where to deploy before deciding what to deploy.
This is the fourth and final installment of Yale CELI’s series. Part one covered job displacement patterns, part two examined data infrastructure readiness, and part three addressed governance frameworks.