Deloitte Proposes AI Digital Staff as a Solution to Production Bottlenecks in Advanced Therapies
Growth Too High to Manage with Staff Alone
Deloitte Consulting LLP diagnosed on August 31, 2026, that the labor-centric operations of advanced therapy (AT) manufacturers have reached their limits. Over 1,000 ATs are in development, and the number of patients in the U.S. receiving these therapies is expected to grow from approximately 12,000 in 2020 to over 340,000 by 2030. During the same period, the cumulative product and indication approvals in the U.S. are projected to reach around 60, rapidly increasing per-patient production and quality control tasks. Particularly, autologous cell therapies have no inventory or alternative batches, so labor shortages directly lead to treatment delays and patient safety risks.
Operational Redesign with Digital Staff
Deloitte's proposed digital staff refers to AI agents that read data from order, slot, and quality control systems, synthesize signals, and perform tasks within defined parameters. Humans retain judgment and accountability, while agents handle repetitive, system-integrated tasks such as slot scheduling, order coordination, and deviation reporting. Salesforce (CRM) launched Headless 360 in April 2026, enabling platform functions to be called via APIs and protocol-based tools, and ServiceNow (NOW) officially provided a universal AI agent MCP server in May. This means advanced therapy companies can introduce an agent layer without replacing their existing enterprise resource planning (ERP) and quality management systems (QMS).
Control Plane and Data Infrastructure Are Key
The proposed model consists of a control plane that oversees record systems, human decision-makers, and agents, and an integrated data fabric that connects them. The control plane sets agent-specific permissions and approval procedures, while the data fabric maintains chain of identity and end-to-end audit trails. In a GxP environment, regulatory acceptance hinges not on automation rates but on the ability to reproduce who acted and on what basis. Therefore, a realistic implementation path is to validate single high-risk tasks first and then expand to supply chain, commercial operations, and manufacturing science and technology (MSAT).
Operational Competition in Commercial Therapies
Representative commercial autologous CAR-T therapies, Novartis (NVS)'s Kymriah (tisagenlecleucel, CD19-targeted) and Kite Pharma, a subsidiary of Gilead Sciences (GILD)'s Yescarta (axicabtagene ciloleucel, CD19-targeted), received FDA approvals on August 30, 2017, and October 18, 2017, respectively. Both products are in the Marketed stage and are homologous indication competitors requiring individual tracking from patient collection to manufacturing, release testing, cold logistics, and reinfusion. The global cell and gene therapy manufacturing market is projected to grow from $17.5 billion in 2025 to $75.1 billion in 2033, making operational scalability directly tied to product competitiveness. The economic value of AI adoption is formed not by simple labor cost savings but by increasing manufacturing slot utilization, deviation prevention, and treatment completion rates.
As the advanced therapy manufacturing market expands from $17.5 billion in 2025 to $75.1 billion in 2033, and the number of U.S. patients treated exceeds 340,000 by 2030, labor-based operations will constrain commercialization speed. From an investment perspective, Deloitte Consulting LLP's advisory demand and the adoption rates of Salesforce (CRM) and ServiceNow (NOW)'s regulated industry agent and workflow platforms are short-term benefit indicators. For research and manufacturing organizations, products like Kymriah and Yescarta, which are Marketed-stage CD19 CAR-T therapies requiring patient-specific batches and chain of identity, mean that slot coordination, deviation handling, and audit trails can be automated. In the medium to long term, the approximately 1,000 development pipelines and the 60 product and indication approvals expected in the U.S. by 2030 will widen the cost and throughput gap between manufacturers who preemptively build control planes and data fabrics and latecomers.