Why knowing about the deviation in advance isn’t enough, and what it means to build a digitized logistics operation that takes action with the most innovative AI-native tools available. By Chris McDonald, former SVP & Global Head of Technical Operations, Kite (a Gilead Company), and Henry Ames, cold-chain & logistics-orchestration veteran (Sensitech, TraceLink, PDA).

Executive Summary

While your control tower can likely see the problem, it still cannot fix it. A decade of logistics platforms promised to transform healthcare supply chains, and most delivered the same thing: a better dashboard. When a cell therapy shipment misses its connecting flight, the dashboard shows the delay, but an individual still calls the courier, waits on hold, reroutes the product, manages the clinical site, and documents the deviation, all while the viability window closes. Visibility was the promise. Phone calls, emails, and manual work are still the reality.

The gap between complexity and capacity is widening fast. As of 2025, Biologics have overtaken small molecules in terms of value, and this growth is expected to continue. Biologics are projected to account for 57% of global prescription drug sales by 2030¹. The IQVIA Institute for Human Data Science estimated that the biopharma industry loses approximately $35 billion annually as a result of failures in temperature-controlled logistics from lost product, clinical trial loss and replacement costs, to wasted logistics costs and the costs of root-cause analysis.² ³ Yet despite these meaningful industry challenges, exception management still scales with headcount: 20% more shipments demand 20% more coordinators, in a market where experienced ones are scarce.

A new architecture closes that gap. AI-native platforms provide capabilities that enable the system to be the first responder: voice agents that call courier operations lines, email agents that notify sites and recipients, no-code workflows that turn SOPs from MS Word documents into executed, audited sequences with every action logged to a 21 CFR Part 11-compliant chain of custody. While the on-time delivery metric remains a primary goal, it is the zero-touch exception resolution rate that captures operational efficiency ultimately supporting on-time delivery performance. Before AI, the best-run operations resolved 10-15% of incidents without a human-in-the-loop.⁴ Today, deployments of AI-native operating platforms are achieving 50%-70% auto-resolution of alerts, 74% shorter delays of shipments in-transit, and quality reviews decreased by three days.⁴

So, the right question for any platform isn’t “Does it show me everything?” It’s “What percentage of exceptions does it correctly resolve without needing a human-in-the loop?” If the answer is zero, you are buying a better dashboard. If it is measurable and growing, you are looking at something new. This paper will highlight the capabilities and benefits associated with the adoption of AI-native logistics operations platforms.

Introduction

The past decade has produced a generation of “logistics control towers” that promised to transform pharmaceutical and healthcare supply chain operations. Most have delivered a better dashboard with improved visibility, some have delivered predictive risk models that surface potential deviations. What they have not delivered is fewer phone calls, fewer escalations, and fewer missed deliveries that would result in increased operational efficiency and delighted customers.

The problem is structural. The dominant architecture of these platforms was built to aggregate data from carrier portals, sensor platforms, and ERPs into a single screen. That is useful. But aggregation and action are two different things. A consolidated view of a flight delay still requires a human to collect relevant information related to possible options, weigh those options based on risk, time, cost, etc… and then decide what to do. And, once a decision has been made, they still need to make the call, send the email, and log the outcome.

In the meantime, shipment volumes in healthcare and life sciences are generally growing faster than the operations teams that manage them. Furthermore, the global regulatory and standards-based guidance covering the storage, handling, and distribution of medicinal products continue to increase. As previously noted, Biologics, which require cold chain logistics, are projected to account for 57% of pharmaceutical sales by 2030.¹ Additionally, many new products require specialized manufacturing and distribution activities. For example, cell and gene therapies (CGT) and radiopharmaceuticals represent unique challenges due to their short shelf-lives and strict quality control timelines. The result is a widening gap between the growing complexity of the logistics operations required for these advanced therapies and the capacity of the teams responsible for getting it to the appropriate location on-time.

A new generation of AI-native logistics platforms is closing that gap, built on technology that didn’t exist two years ago. The first change is in design: these platforms bring “Uber-like” intuitiveness to a life sciences vertical that has long resisted change. The more important change is operational: they augment teams by automating the manual work that follows an alert.

Part I · The Problem

When the dashboard sees everything and fixes nothing

The Control Tower Promise vs. Reality

The term “control tower” entered healthcare logistics roughly fifteen years ago, borrowed from aerospace and general freight. The concept was straightforward: instead of checking five carrier portals and two sensor dashboards to understand the status of your shipments, data would be aggregated into a consolidated view.

For a small operations team managing a few hundred shipments a month, this was transformative. For a team managing tens of thousands of shipments across multiple couriers, various sensor technology and vendors, clinical trial sites, and patients, the consolidated view has become the floor, not the ceiling. The problem isn’t visibility. It is how to effectively respond to the alerts that actually matter in a timely manner - resulting in a positive impact.

Consider a scenario familiar to anyone who manages clinical supply logistics. A cell therapy shipment is en route from a manufacturing site in Los Angeles to a clinical site in Kansas City. The courier hasn’t updated the tracking portal since pickup. The flight that was supposed to carry the package has been delayed. The clinical team at the receiving site has a patient scheduled. The logistics coordinator, checking the control tower dashboard, can see the delay. What happens next is entirely manual: someone calls the courier’s operations line, waits on hold, escalates, potentially sources an alternative solution, calls the site to manage expectations, and documents the deviation in the quality system. This is all happening while the product’s viability window continues to close.

The dashboard didn’t solve the problem. It just made the problem visible.

The Aggregation Trap

The dominant architecture of first-generation logistics platforms was built around data aggregation: pull tracking events from FedEx, UPS, and specialty couriers; pull temperature readings from sensor platforms; pull flight status from aviation data feeds; display it all in one place. This is technically impressive and operationally useful. But it is not intelligence, and it is not action.

The aggregation trap is the assumption that giving people more information automatically leads to better outcomes. In practice, more information without automated resolution creates three compounding problems.

Alert fatigue

When every shipment event triggers an alert, the minor temperature blip, the routine scan delay, the benign carrier code that only looks alarming, operations teams learn to ignore the feed. A system that fires hundreds of alerts a day buries its real exceptions in its own noise, and the visibility meant to surface problems ends up hiding them.

The human-in-the-loop bottleneck

Every exception that does require action still requires a person to interpret the situation, decide on a course of action, execute communications with carriers and recipients, and document the outcome. At Uber Freight, where one of Zoomlogi’s co-founders oversaw logistics operations at scale, the metric that distinguished high-performing teams was the percentage of exceptions resolved without human intervention (i.e., zero-touch rate). Pre-AI, even the most sophisticated operations achieved single-digit zero-touch rates on complex exceptions. Everything else was manual.

The headcount scaling problem

Manual exception management scales linearly with volume. Add 20% more shipments and you need 20% more coordinators to handle the exceptions, or you accept degrading service quality. In a labor market where experienced logistics coordinators are scarce and expensive, this creates an operational ceiling that no dashboard resolves.

The Pressure Is Getting Worse

The gap between complexity and capacity is not static, and it is accelerating. Three structural forces are converging to make it more acute.

The biologics boom

Biopharmaceuticals are growing at roughly 9%⁵ annually and are projected to account for 57% of global pharmaceutical sales by 2030⁶, up from under 10% in the early 2000s.⁷ Unlike traditional small-molecule drugs, biologics are temperature-sensitive, time-sensitive, and often patient-specific. A shipping failure is not a reorder. It is a patient whose surgery is delayed without their product, a clinical trial that misses an enrollment window, or a sample that needs to be recollected.

The cell and gene therapy inflection

Cell and gene therapies represent one of the most logistically challenging categories. Autologous therapies, where a specific patient’s cells are collected, modified, and returned to that same patient, have zero tolerance for delay. There is no backup product. A failed shipment is not a logistics failure; it is a patient safety event.

One operations leader at a Phase 2 autologous cell therapy company described a real incident: a patient was already in the operating room for a tumor resection when the company discovered that the cryogenic container had not made the connecting flight from Los Angeles. The team had to locate a pre-conditioned shipper from a nearby warehouse and coordinate an emergency alternative while the patient was on the table. “If we don’t come through,” the operations lead said, “that’s a patient that went in the operating room doing surgery for nothing.”

The rural healthcare access expansion

The geography of healthcare delivery is also shifting. Specialty medications, diagnostic kits, and clinical trial materials are increasingly being shipped to patients at home and to clinic sites in rural markets that lack the logistics infrastructure of urban academic medical centers. This expands the surface area of last-mile complexity precisely at the moment when the products being shipped demand more precision.

Part II · What Changed

Architecture decides whether AI can act

AI-Native vs. AI-Bolted-On: Why Architecture Matters

The term “AI-powered” has become nearly universal in logistics software marketing. It describes everything from basic predictive ETAs, to large language model assistants, to rule-based automation rebranded with a more marketable name. For logistics leaders making purchasing decisions, the distinction that matters is not whether a platform uses AI. It is whether the platform was architecturally designed around AI from the beginning, or whether AI capabilities were added to an existing visibility product.

The distinction has both direct operational and compliance implications.

AI-bolted-on: the upgrade problem

Most first-generation logistics platforms were built to display data. Their core data model, event processing logic, and user experience were all designed with a human reviewer at the center. AI features added later were typically a predictive ETA layer, a risk scoring overlay, or a chatbot interface sitting on top of an architecture that was never designed to support autonomous action. The problem is not that these features are bad. Predictive ETAs that are more accurate than carrier-published estimates are valuable. Risk scores that flag shipments before an exception becomes critical can help a coordinator prioritize their queue. But because the underlying platform was designed for display, not for action, the AI layer can identify a problem and surface it, but the resolution still runs through the same manual workflows it always did. The AI got better at telling you what was wrong. The process of fixing it did not change. This problem is further compounded when legacy visibility vendors attempt to solve it through acquisitions.

Buying an AI company and integrating it into an existing platform does not change the underlying architecture, but instead it just adds a layer on top of it. The seams show because data models built for display don’t natively support autonomous action, audit trails designed around human workflows can’t seamlessly log AI-generated decisions, and the acquired product must be re-engineered to fit an architecture it was never designed for. The result is a visibility platform with expanded marketing, not a fundamentally different operational capability. There is no acquisition path to AI-native, the capability has to be fused into the product from the start.

AI-native: action as the default

An AI-native platform is built with the assumption that the system, not the human, should be the first responder to any exception. This changes the data model, the integration layer, and the user experience.

In the data model, every shipment event is classified not just by what happened but by what the appropriate response is, and whether that response can be executed autonomously. A flight delay that adds 45-minutes to a 48-hour transit for a refrigerated product packed in an insulated shipper with a pre-qualified shipment duration of 96-hours is a low-priority notification. The same delay for an autologous cell therapy with a 24-hour viability window is an immediate escalation requiring carrier coordination, recipient-site notification, and QA documentation. An AI-native platform encodes these decisions into the system rather than leaving them to individual coordinators.

It is worth noting that further delays would result in additional escalation paths that could include clinical site coordinators, infusion scheduling adjustment, and confirmed chain of custody, to name a few.

In the integration layer, AI-native platforms go beyond read-only API access to carrier and sensor data. They maintain active communication channels: voice AI that can call courier operations lines, handle inbound inquiries from drivers who are having difficulty finding a delivery location, email agents that can send and interpret responses, webhook connections that log outcomes back into the chain of custody automatically for a full audit-trail. The ultimate goal is not to simply give a human a better picture to act on. The goal is to build a system that emphasizes a human-centric, risk-based approach, human-in-the-loop oversight, data integrity, and AI lifecycle management in line with the FDA and EMA’s jointly published guidance on Good AI Practice in Drug Development.

In the user experience, the primary interface is not a dashboard of shipments, it is a queue of exceptions that require human judgment based on an individual company’s risk-based approach. For example, a routine call based on a customer SOP to confirm the pickup has happened can be automated with AI. However, a re-routing decision for a cell and gene therapy shipment where the viability window clock is winding down, requires human approval.

The question shifts from “What is the status of my shipments?” to “What do I actually need to decide today?”

AI-native platforms take this further by continuously monitoring upstream risk signals like weather events, labor actions, and carrier operational status, to identify at-risk shipments and trigger re-routing before an exception ever occurs.

The Compliance Dimension

For pharmaceutical and biotech operations, AI architecture is not just an operational question, but also a compliance question. Under 21 CFR Part 11 and EU Annex 11, any system that generates or maintains electronic records related to GMP-regulated activities must meet specific requirements for audit trails, access controls, and data integrity. This includes chain-of-custody records for clinical trial materials and temperature excursion documentation.

A platform that uses AI to autonomously execute communications with carriers, recipients, and QA teams must ensure that every AI-generated action is logged with full traceability: what was communicated, to whom, when, and what the response was.

In a bolted-on architecture, where the AI layer sits above the record-keeping system, this traceability is difficult to guarantee. In a purpose-built architecture, every AI action is natively threaded into the same audit trail as human actions because the system was designed from the start with the assumption that AI would be doing the work.

One operations leader at a global pharma 3PL described the value: “What’s valuable here is the completeness of the information with chain of custody, temperature data, courier events, and customer messaging, all in one place.” The value is not just visibility, it is defensible documentation.

Digitization of the Standard Operating Procedure (SOP)

Every logistics team has a version of the same artifact: a SOP document, written in Microsoft Word, that describes what to do when a specific exception occurs. What to do when a carrier shows “Delivery Exception.” What to do when a temperature excursion alert fires. Who to call when a site says the shipment hasn’t arrived, and what to do if that contact doesn’t answer the phone. These documents represent years of accumulated institutional knowledge, and they have two consistent failure modes.

The first is inconsistent execution. SOPs describe what to do; they do not execute it. When a new coordinator joins the team, or when the experienced coordinator is out and someone else handles the queue, steps get skipped, condensed, or misapplied. Exception resolution becomes dependent on who is working that day, which is the definition of a fragile process.

The second is the documentation lag. The exception happened, the coordinator handled it, the shipment was recovered, and the outcome was logged in a ticket, an email chain, a Microsoft Teams thread, and possibly a spreadsheet. The SOP was never updated to reflect what actually worked. The knowledge stays tribal.

No-code workflow builders: the SOP becomes executable

A new class of no-code workflow tools in logistics platforms allows operations teams to translate their SOPs directly into automated sequences that execute without human intervention (i.e., if this, then that logic). Rather than describing what to do in a document: if a flight delay alert fires with fewer than four hours remaining in the product’s stability window, automatically call the courier’s operations line using the AI voice agent; simultaneously notify the receiving site; log the alert and all subsequent communications in the chain of custody; escalate to the on-call coordinator only if the courier cannot commit to a resolution within 30 minutes.

The significant shift is that these workflows do not require software engineers. Operations leaders familiar with basic Microsoft tools can configure them directly. One logistics leader described early results: “The efficiency gains are massive.”

These digitized workflows also become executable at scale in a consistent manner that was not historically realistically possible. When the workflow fires and completes - or escalates because it could not complete autonomously, every action is logged, timestamped, and attached to the shipment record. The SOP is not just described; it is executed with full auditability.

Part III · The Benchmark

Beyond on-time delivery: measuring zero-touch resolution

For most of the past decade, healthcare logistics performance has been measured primarily through on-time delivery rate - the percentage of shipments that arrive within the committed timeframe. This is a useful metric. But it is a lagging indicator that tells you what happened, not how efficiently your team handled what happened. On-time delivery rate also obscures the hidden cost of exception management. A shipment that was technically “on time” may have required two hours of coordinator time, three calls to the courier, and a documented deviation, none of which are visible in an OTD report. Two operations teams could have identical OTD rates and radically different operational costs, service consistency, and compliance risk profiles.

The metric that captures the difference is the zero-touch exception resolution rate: the percentage of shipment exceptions that are identified, actioned, resolved, and documented without requiring any human intervention.

Why this metric was not trackable before

In first-generation logistics platforms, zero-touch rate was meaningless. The platform could not resolve exceptions, so every exception required human action by definition. Operations leaders optimized for response time instead: how fast the coordinator noticed the alert and acted. AI-native platforms change the denominator. When the platform can autonomously resolve a significant percentage of exception types, for example incorrect addresses, pickup confirmation calls, delay notifications to recipients, temperature excursion documentation, carrier escalations for overdue deliveries, the question is no longer how quickly humans respond. It is which exceptions actually need a human, and why.

At Uber Freight, which built one of the most sophisticated freight operations platforms available before the current generation of AI tools, zero-touch rate on routine exceptions was a primary operational KPI. Even with significant automation investment, the pre-AI ceiling was roughly 10–15% for complex exception categories. Voice AI that can conduct live conversations with courier operations lines, interpret responses, and take follow-on action has fundamentally changed what is achievable. Early results in pharmaceutical logistics suggest that zero-touch rates of 50% or higher are achievable for the most common exception types.

What to Measure: A Practical Framework

For logistics leaders evaluating their current operations, or evaluating a platform that claims to improve them, the following metrics provide a more complete picture than on-time delivery rate alone.

1. Zero-touch exception resolution rate — Of the exceptions your team handles monthly, what percentage resolved without any human action? A meaningful AI-native platform can tell you this and show which exception types resolve autonomously versus escalate. If a vendor can’t give you a zero-touch rate by exception type, the AI layer is likely cosmetic.

2. Time-to-resolution by exception type — For exceptions needing human intervention, how long from alert to close? This exposes where bottlenecks lie: carrier communication, QA documentation, or recipient coordination. AI-assisted resolution should compress this even where judgment is required: AI handles communication and presents a risk-based assessment of options, while the human decides.

3. Coordinator hours per exception — How many hours does your team spend per exception, fully loaded - response, carrier and recipient communications, analysis, documentation, follow-up? Most teams can’t say, because the work is scattered across email, phone, and spreadsheets. If you can’t measure cost per exception, you can’t tell whether a platform is improving it.

4. Chain-of-custody completeness rate — For regulated shipments, what percentage have a complete, auditable record of every event - carrier updates, temperature readings, site and recipient communications, documented outcomes? An incomplete chain of custody is both a compliance and service risk, and signals your tools aren’t capturing the full record automatically.

Setting a realistic baseline

Before evaluating any platform, organizations should establish their current baseline on these metrics. Across the pharmaceutical logistics teams we have worked with, the most common findings are:

These baselines are not unusual or indicative of poor management. They are the predictable output of operations teams using tools designed for aggregation, not action.

Conclusion

The right question to ask

The healthcare and life sciences industry is at an inflection point in logistics technology. The first generation of control tower platforms delivered genuine value by consolidating disparate data sources into a unified view. But for operations teams managing increasingly complex shipments at increasing volume, visibility is no longer the constraint. AI-native logistics platforms that can autonomously execute exception resolution - not just surface alerts, but make calls, send updates, log outcomes, and escalate only what actually requires judgment - represent a structural change in what is possible. The early results are meaningful: teams reclaiming hours previously consumed by carrier hold queues; AI agents recovering specialty pharmacy shipments in hours that would have taken more than two days manually; invoice audit automation recovering freight spend that no one had the bandwidth to pursue; claims approval rates nearly tripling through automated documentation.

The right question for logistics leaders evaluating any platform is not “Does it show me everything?” It is: “What percentage of exceptions not requiring escalation for human judgment does it resolve without my team’s involvement?” And, for those decisions that require a human-in-the-loop, how much efficiency is gained in the resolution process because the AI-native platform has not only identified and escalated the issue, but also provided several risk-weighted potential options.

If the answer is zero, you are looking at a better dashboard. If the answer is measurable and growing, you are looking at something new.

About & References

Chris McDonald

SVP & Global Head of Technical Operations · Kite, a Gilead Company

Chris McDonald brings decades of experience scaling the world’s most complex pharmaceutical supply chains. As SVP and Global Head of Technical Operations at Kite, a Gilead Company, he oversaw the commercial delivery of Yescarta and Tecartus, two of the world’s first CAR-T therapies, managing global manufacturing, quality, process development, and supply chain across six worldwide sites and more than 3,000 employees. He has held senior leadership roles at Amgen, Novartis, and AstraZeneca, and currently serves on the Advisory Board of Cellares and the Board of Directors of Nucleus RadioPharma.

Henry Ames

Cold-chain & logistics-orchestration veteran · Sensitech, TraceLink, PDA

Henry Ames has built his career at the intersection of pharmaceutical logistics and regulatory compliance. As General Manager at Sensitech, a leading IoT provider in healthcare and pharma, he led the life sciences business and contributed directly to the cold-chain regulation now governing the industry. He later served as General Manager at TraceLink, a leading platform for logistics orchestration and drug serialization compliance, and currently serves as Co-Chair of the Parenteral Drug Association’s Supply Chain Management Interest Group. Henry is also a Contributor to IATA’s Cell and Gene Therapies Focus Unit and an Observer in its Healthcare Cargo Working Group.

References

1. 2025 World Preview: Pharma Growth Steady Amid Turbulent Seas and Rising China. evaluate.com/thought-leadership/2025-world-preview

2. IQVIA Institute for Human Data Science. The Global Use of Medicines 2023: Outlook to 2027. Cold chain cost estimate from IQVIA commissioned research, 2019.

3. 2019 Biopharma Cold Chain Logistics Survey (Peli BioThermal). 2019 Biopharma Cold-Chain Logistics Survey, HubSpot.

4. Prospect quotes and operational statistics sourced from anonymized Zoomlogi customer conversations (2025–2026).

5. ResearchAndMarkets / GlobeNewswire (2025). Biopharmaceuticals Global Overview Report 2025.

6. Evaluate Pharma World Preview 2024/2025. evaluate.com 2030 forecasts for global pharmaceutical market.

7. Generics and Biosimilars Initiative (GaBI Online). Biologicals sales quadruple from 2002 to 2017.