AI In Clinical Supply: From Forecasting To Execution Decisions

A conversation with Tony Li, Supply Lead, Alexion (AZ Rare Disease)

AI is increasingly being discussed as a way to transform clinical supply planning, but the real opportunity may extend beyond better forecasting. As enrollment, IRT demand signals, inventory, manufacturing timelines, and distribution conditions change, clinical supply teams need to translate those signals into timely decisions about production, packaging, labeling, release, deployment, and replenishment.

That creates a particular challenge in clinical trials. AI can identify patterns and generate probabilistic risk assessments, but supply decisions often still require clear, governed actions. The value of AI therefore depends not only on how accurately it predicts what might happen, but on how effectively it helps teams connect those predictions to execution while maintaining quality, compliance, and patient continuity.

In this Q&A, Elizabeth Urbanek, executive editor of Clinical Supply Leader, speaks with Tony Li, clinical supply chain leader at AstraZeneca, about how AI could reshape clinical supply planning and execution, where it can create the most immediate operational value, and how organizations can use AI to support faster, better-governed decisions without replacing human accountability.

1. In today’s clinical trials, how do you see AI changing the fundamental structure of planning cycles when IRT demand signals, enrollment shifts, and supply availability are all updating continuously rather than on fixed planning cadences?

Historically, clinical supply planning has operated on a periodic cadence, with teams reviewing forecasts, enrollment, inventory, and supply requirements at defined intervals. For example, a monthly refresh may capture changes in demand and enrollment and assess their potential impact on the supply plan. However, this approach can be less effective when recruitment, screening, randomization, and supply availability are changing continuously.

AI has the potential to shift clinical supply planning from a periodic reactive process to one based on continuous sensing and proactive exception-based decision-making. AI-enabled tools could analyze structured data, including IRT demand signals, enrollment trends, screening rates, inventory levels, and shipment status, alongside relevant unstructured information from site feasibility assessments, site communications, and operational updates.

By identifying emerging patterns earlier, these tools could highlight situations such as a site recruiting faster than expected, an increase in screening activity that may translate into future demand, or a developing supply constraint at a particular depot or in a specific country. AI could then provide near-real-time alerts, explain the factors driving the potential risk, and support scenario analysis — for example, whether to adjust deployment timing, reallocate inventory, or revise the supply plan.

The fundamental change is that planning would no longer be limited to reviewing what has already happened during a monthly cycle. Instead, it could become a continuous process focused on anticipating what is likely to happen next and responding to exceptions as they emerge. Human expertise and governance would remain essential: AI can identify signals, quantify potential impacts, and recommend options, while clinical supply and operational teams retain accountability for decisions and execution.

2. Where does AI create real value inside these constraints today: improving prediction quality or helping teams re-sequence and optimize physical execution steps like packaging runs, labeling campaigns, and batch release timing?

AI can create value in both prediction and execution, but the more immediate opportunity may be in coordinating and optimizing the physical steps required to deliver clinical supply. Activities such as batch release, packaging, labeling, and drug product shipment often involve multiple functions, including quality assurance, manufacturing, demand planning, and clinical supply operations. They also depend on precise sequencing, handoffs, documentation, and adherence to approved procedures.

Today, coordinating these activities can be highly manual. Information may be distributed across several systems and teams, making it difficult to maintain a complete current view of the overall timeline. A missed dependency or delayed handoff can affect downstream activities and, ultimately, the ability to meet patient demand.

AI-enabled systems could help create a connected view of the end-to-end process. They could track milestones, identify dependencies, flag potential delays, and assess how changes in recruitment or demand may affect packaging runs, labeling campaigns, batch release, and shipment timing. They could also retrieve the relevant approved standard operating procedure for a task, provide context to the responsible team, and support the preparation of next-best actions.

This could reduce the administrative burden on teams and help minimize errors in detail-intensive activities. The greatest value may come from linking prediction to execution: identifying a potential supply risk is important, but translating that insight into a timely adjustment to the release, packaging, labeling, or shipment plan is what ultimately supports patients and study continuity.

AI would not replace established quality systems or the accountability of qualified personnel. Release decisions, procedural interpretation, and other regulated activities would remain subject to appropriate human review, governance, and approval. AI would serve as a decision support and coordination layer, helping teams manage complexity and act earlier.

3. How should AI-generated probabilistic outputs be translated into these binary operational decisions, especially when managing batch release timing, expiry risk, and patient continuity?

AI-generated outputs are probabilistic, while clinical supply decisions often need to be binary: proceed or defer, release or hold, produce or do not produce. The way to bridge that gap is through a predefined decision framework that combines AI-generated risk assessments with agreed business rules, study priorities, and decision rights.

This is particularly important because the manufacturing timeline for clinical supply can extend from nine to 18 months. Decisions about production therefore need to be made well before actual patient demand is fully visible. By the time a recruitment trend is confirmed, it may be too late to manufacture and release additional supply without creating a risk to study continuity. Conversely, producing too early or in excessive quantities can increase expiry exposure, inventory costs, and the risk of unused material.

Our teams already use scenario-based planning to reflect both current and potential clinical studies. The clinical supply agreement between clinical operations and clinical supply should define which studies are endorsed and unendorsed, as well as the expected timelines for delivering investigational medicinal product to each country. These commitments provide the foundation for translating changing demand signals into operational decisions.

4. Where do you see the most underutilized opportunity for AI in protocol feasibility, CMC readiness assessments, labeling strategy design, and packaging configuration decisions before supply execution even begins?

In my view, the greatest value comes from connecting these activities rather than optimizing each one independently. A protocol decision can affect demand, which affects manufacturing capacity, shelf-life requirements, labeling, packaging, and country deployment. AI can help model those relationships at the point when the organization still has the greatest ability to change the design.

5. Do you see a realistic role for AI in supporting decisions around depot allocation, country-specific labeling strategies, and last-mile distribution risk thresholds, or are these inherently governed decisions that AI can only inform and not execute?

AI has a realistic and potentially significant role in supporting decisions around depot allocation, country-specific labeling strategies, and last-mile distribution risk. Its greatest value may be as a global clinical supply control tower that continuously monitors multiple data sources, identifies emerging risks, and coordinates timely responses.

For example, AI-enabled systems could monitor temperature data, inventory levels at depots and clinical sites, shipment status, packaging plans, labeling requirements, and delivery timelines across regions. Operating continuously and at global scale, such systems could identify a temperature excursion or a developing inventory risk rather than relying on the manual process reported by clinical sites.

Once an event is detected, AI could assess the available information against approved stability data, predefined thresholds, and established quality procedures. It could classify the event, identify the relevant stakeholders, and recommend appropriate actions. Depending on the level of risk and the approved operating model, this could include escalating the event to quality assurance, notifying the clinical site not to use potentially affected product pending assessment, alerting the clinical supply team to arrange replacement stock, or recommending a shipment reroute.

Some operational actions, such as changing a delivery route or escalating a shipment exception, may be suitable for execution under predefined and validated rules. However, decisions about product disposition, continued use after a temperature excursion, or release of material should remain subject to the appropriate quality systems and authorization by qualified personnel. AI should support those decisions by providing a timely, traceable assessment — not by replacing the accountable decision maker. The same principle applies to depot allocation and country-specific labeling.

The key is to map these scenarios in advance and define the applicable governance, decision rights, escalation pathways, and quality criteria. With appropriate validation, data controls, audit trails, and human oversight, AI could reduce response times, improve consistency, and help prevent a local distribution issue from becoming a patient supply interruption.

6. How should AI systems be designed differently for clinical trial supply, where disruptions in batch release, labeling execution, or depot replenishment can directly interrupt patient dosing continuity versus other regulated or industrial environments?

AI systems for clinical trial supply need to be designed around patient centricity, quality, and controlled decision-making, rather than efficiency or cost optimization alone. In many industrial environments, a delayed shipment or production change may affect service levels or financial performance. In a clinical trial, however, a disruption to batch release, labeling, or depot replenishment can directly affect a patient’s ability to receive treatment according to the protocol.

Another important difference is that clinical supply systems need to be designed for rapid escalation and resilience. Rather than waiting for a scheduled review, the system should monitor continuously, identify the earliest point at which patient continuity may be at risk, and initiate the appropriate escalation pathway. It should support scenario analysis, such as determining whether an alternative depot, shipment route, label configuration, or replacement batch could protect treatment continuity.

Finally, the objective should not be to optimize a single metric, such as inventory cost or delivery speed. The system should balance patient impact, product quality, protocol requirements, supply assurance, expiry risk, capacity, and compliance. In this environment, the most effective AI is not necessarily the system that makes the most decisions autonomously. It is the one that helps the right people make timely, transparent, and well-governed decisions before a supply disruption affects a patient.

About The Expert:

Tony Li is currently Associate Director, Clinical Supply Lead, at Alexion Pharmaceuticals, Inc. He joined the global clinical supply team in 2022 and has been supporting several late phase programs, including supply planning, vendor management and global distribution to support rare disease patients in over 100 countries.

Prior to Alexion, he worked for Charles River Labs for five years as a Sr. Manager, Global Categories, leading category strategies, strategic projects, Supply Chain analytics, contract negotiations and SRM & KPI Metrics. In addition, he led logistics planning for kidney care division at Fresenius. He redesigned the direct to patient delivery model and managed the RFP projects for both linehaul and last mile freight. In the first 10 years of his career, he had various Operations roles at Trelleborg and Sinochem, both in Asia and the Greater Boston area.

He studied Supply Chain Management at MIT and is Lean Sigma Certified through Purdue University and ‘Certified Supply Chain Professional’ (CSCP) through APICS. In his spare time, he volunteers for CSCMP as VP of Programs of the New England Roundtable.