Process Architecture in AI-Augmented Service Operations: From Horizontal Adoption to Vertical Redesign

We consider a firm operating a sequential service process, such as mortgage origination or M&amp;A due diligence, where work passes through specialist workers at successive stages. As AI improves, the firm faces a choice between two modes of adoption. Under horizontal adoption, AI is deployed within the existing process: workers become more productive, but the division of labor and handoffs between specialists remain unchanged. Under vertical redesign, AI enables the firm to reorganize the process by merging tasks, eliminating handoffs, and expanding workers' scope. We study how this choice depends on the task structure of the process, and derive implications for staffing, pricing, and welfare. <div> <br> </div> <div> Tasks differ in their context-engineering burden, κ: low-κ tasks are more structured and benefit more from AI, while high-κ tasks rely more on tacit expertise and improve less. Under horizontal adoption, this heterogeneity creates growing imbalance: lowκ tasks accelerate while high-κ tasks become bottlenecks. Staffing may initially rise, especially at the bottleneck, but eventually declines as productivity gains dominate.&nbsp; </div> <div> <br> </div> <div> Vertical redesign is different. Merging tasks removes handoffs and shortens flow times, but also broadens workers' roles. A key asymmetry emerges: horizontal adoption mainly reduces costs, whereas vertical redesign mainly improves service. Redesign becomes attractive when AI is capable enough to support broader roles and when customers value speed enough for the faster process to generate additional demand. Because firms do not fully capture the consumer surplus created by faster service, private incentives favor delaying reorganization until AI is more capable than is socially optimal. </div>