Smarter safety stocks for a volatile world: Harnessing artificial intelligence and real-time data for resilient pharmaceutical supply chains

Lead times are shortened and stabilised through real-time release (RTRT) and continuous manufacturing (CM), and supply networks are redesigned around postponement and regional capacity. Recent data points from Pfizer, Janssen, AstraZeneca, and Roche are cited to illustrate feasibility. Quality Management System (QMS) changes are outlined so inventory is treated as an explicit risk control aligned with ICH Q9/Q10 principles for product availability.

In the last few years geopolitics through export controls, energy shocks, logistics disruption and shifting trade-blocs have pushed many pharmaceutical manufacturers – especially for drugs where safety stocks are not mandated, such as generics – toward a defensive posture: more and more safety stock, a “just in case” inventory. This response is straightforward in the short-term, but costly, ties up capital, and still fails to prevent all shortages. The question we should be asking is not “how much extra stock do we need?” but “how do we design systems so that we need less safety stock without increasing risk?”.

1. AI-driven agility where safety stock is an output, not an input

In most planning systems, safety stock is a static variable based on historical demand variation and assumed lead times. But in many pharma settings, the main causes of stockouts are not demand spikes, but quality events, yield losses, testing delays and supplier failures. The data describing these risks already sit in the PQS: deviation and Corrective And Preventative Action (CAPA) trends, Out Of Specification (OOS) and stability failures, process capability, batch release times, Quality Control (QC) capacity utilisation, and supplier audit performance. AI models can combine these with supply-chain data trends to estimate the probability of non-supply for each product, site and horizon. Safety stock now becomes the output of an always-changing risk model rather than a fixed value.

In fact, several large companies are moving in this direction. Pfizer has reported using advanced analytics on manufacturing and supply-chain data¹ to detect bottlenecks and optimise vaccine and antiviral production, increasing yield and reducing cycle time on critical steps. Pfizer explicitly stated that the cycle time of a critical step was reduced by 67%, allowing the production of 20,000 doses per batch.² AstraZeneca and Roche have experienced similar gains, with AZ integrating Bluecrux’s Binocs into 17 QC laboratories to improve operations³ and Roche eliminating labour intensive processes like manual reviews and investigations, saving 30-50% more time⁴.

As process capability, supplier performance and deviation management improve, the modelled risk declines and the quality related component of safety stock can be reduced. Inventory becomes an explicit risk control within the PQS, lining up with ICH Q9/Q10 thinking on risk-based management of product availability.

2. Real-time release and continuous manufacturing as inventory levers

A substantial amount of safety stock exists to protect against long and unpredictable lead times for manufacturing, testing and release. Real-time release testing (RTRT) and continuous manufacturing (CM) changes this and, with it, the required buffer inventory.

RTRT uses in-line or at-line analytics and control strategies to assure important quality attributes without waiting for lengthy off-line tests. CM converts stop-start batch campaigns into steady-state or semi-continuous operations with integrated process analytical technology (PAT). Both approaches can decrease end-to-end production and release time from weeks to days and reduce the time variability.

Early adopters have demonstrated these effects for commercial small-molecule products, reporting sharp reductions in test-to-release times and more predictable output. Janssen, for example, reported an 80% reduction in manufacturing and cycle testing time when the company was given FDA approval for CM for the HIV drug, Prezista⁵. Once these processes stabilise, because there is less process time and uncertainty, inventory footprints decrease. In practice, capital intensity, regulatory effort, and integration with existing batch-based packaging and QP release processes mean RTRT and CM will initially be most suited to specific product types – for example, high-volume oral solid doses or high-value, “quick-to-make” products – rather than the entire portfolio. When shown explicitly in QRM and QMS processes, they can be recognised as elements of shortage prevention and supply continuity strategy, as well as quality assurance tools.

3. What fast fashion does better and how much can pharma adopt?

In comparison to other industries, fast fashion companies like Zara operate with relatively low inventory and high agility ⁶. They achieve this by designing supply chains around short lead times, regional capacity and postponement. As a result, Design to store cycles are kept very short (approximately a fortnight compared with 5-6 months for typical retailers)⁷, production is located close to key markets, and certain decisions, like the printing and dying of greige fabric (~50% of Zara’s fabric purchase)⁸, are delayed until demand signals are clearer. Stores are replenished frequently in small lots, leading to high inventory turns and relatively low unsold stock. Zara, for example, achieves around 12 inventory turns per year versus 3-4 for typical competitors, with an unsold inventory of approx 10% compared with 17-20% industry averages ⁹.

While pharmaceutical and chemical products’ regulatory constraints are very different, some underlying design principles are transferable. One pattern is to hold more inventory in a semi-finished and to push differentiation closer to the market via regional late-stage operations. Drug substance or bulk intermediates can be produced in a small number of highly optimised sites, while packaging is distributed across regional hubs nearer to patients. Customisation in later stages¹⁰ is already discussed in pharma packaging as a way to manage small batch sizes and country-specific requirements more flexibly, although practical deployment remains limited.

Resilience here comes from postponing final product configuration and destination, allowing half-finished stock to be redirected as conditions change. Regional capacity absorbs shocks, reducing central inventories and lowering expiry/obsolescence risk. The Zara model is less viable for vaccines or complex oncology APIs with safety-stock needs and long chains, but may fit some fast, high-value products. It requires strong quality/regulatory design and procurement-enabled resilience.

Bringing it all together to build the data spine for risk-based inventory control into the QMS

To make safety stock an explicit risk control, the QMS should leverage the data already collected, deviation aging, CAPA trends, OOS rates, batch release times, QC capacity, and supplier audit scores, in a common structure. A simple model could use these indicators to generate a “quality-risk” score that planners can use alongside demand and lead-time variability. The transformation can be gradual and incremental, e.g., start with a shared dashboard that publishes three metrics: weekly median release turnaround, deviation closure time, and QC capacity utilisation. Over time, these feed an AI-driven risk score per product and site, allowing inventory buffers to shrink as process capability and quality performance improve. In effect, the QMS becomes a shortage-prevention tool, consistent with ICH Q9/Q10 while increasing overall efficiency and potential reduction in the safety stock.

Conclusion

In conclusion, Safety stock requirements will remain necessary in pharmaceutical and chemical supply chains, especially for critical products with mandated national reserves and very long lead times. Treating it as the primary defence against volatility is neither economically sustainable nor aligned with current capabilities. AI-driven integration of quality and supply-chain data can make safety stock a dynamic, risk-based output. Real-time release and continuous manufacturing shorten and stabilise lead times, structurally lowering the buffer required. And design patterns, inspired by agile sectors such as fast fashion, show how capacity, proximity and postponement can, for some portfolio segments, decrease the inventory required. All three approaches depend on collaboration between supply chain, quality, regulatory, process and procurement experts, and are best applied selectively to certain products where constraints are limited. These approaches give an attainable path to less inventory and more resilience.

This article was first published in Chemistry Today.

References and notes

¹ Singh, P., Thakur, A., & Yadav, D. (2025). AI driven Innovations in Pharmaceuticals: Optimizing Drug Discovery and Industry Operations. RSC Pharmaceutics.
² Pfizer. (2022). Pfizer leverages digital innovation to help deliver medicines to patients faster.
³ Bluecrux & Pharma Innovator Partner for QC & SC Integration. (2024, May 27). Bluecrux.
⁴ 3 AI Breakthroughs Revolutionizing Pharma Tech Ops At Roche. (2024).
⁵ O’Connor, T. (2023). CDER’s Perspective on the Continuous Manufacturing Journey (p. 1).
⁶ Vidyabharati International Interdisciplinary Research Journal. (2021, October). Retailing and supply chain management in fashion industry: best industry practices (A. S. Rathi, Ed.) [Review of Retailing and supply chain management in fashion industry: best industry practices].
⁷ Mellor, A. (2025, September 24). €36B revenue built on a 2-week supply chain. Linkedin.com.
⁸ Chu, P., Maria, A., & Saenz, J. (2005). Excellence In European Apparel Supply Chains: Zara.
⁹ Kumar, Anurag. “This Is How ZARA Leverages Analytics to Dominate Fast Fashion.” Medium, 15 Apr. 2024. Retrieved May 8, 2026.
¹⁰ Hapa Company Insight – Pharma Technology Focus | Issue 79 | February 2019. (2024, February 23).