Artificial intelligence is rapidly becoming embedded in how companies make operational and commercial decisions, but some of its most consequential applications are taking place far beyond generative AI. In global logistics, where small improvements in pricing, procurement, network design, and transportation decisions can translate into significant business impact, AI has been reshaping decision-making for years.
Javier Rodriguez Del Campo built his career at the intersection of supply chain, commercial strategy, and applied artificial intelligence. Currently a Senior Product Manager at Amazon Global Logistics, he leads initiatives focused on transportation pricing and commercial decision-making across one of Amazon’s largest inbound logistics programs. His work spans machine-learning-driven pricing, transportation procurement, negotiation intelligence, and the development of tools that help business teams translate complex models into actionable decisions.
Before Amazon, Rodriguez worked at McKinsey & Company, where he advised organizations on operations and advanced analytics transformation, and later earned his MBA from Harvard Business School. Across his career, he has focused on a recurring challenge: how to move advanced analytics and AI beyond experimentation and embed them into real operating decisions.
In this interview, Javier discusses why some of the most valuable enterprise AI applications predate the generative AI boom, what it takes to move machine learning from prediction into production, why explainability matters in high-stakes commercial decisions, and how AI is reshaping the future of logistics, procurement, and pricing.
Could you introduce yourself and tell us about the work you are doing today?
I’m Javier Rodriguez Del Campo, and my career has largely been focused on the intersection of technology, strategy, and operations.
Today, I work at Amazon Global Logistics, where I lead product initiatives related to transportation optimization, pricing, and commercial decision-making. A significant part of my work involves using data science, machine learning, and AI-supported tools to help make better decisions around pricing, procurement, and negotiations. Before Amazon, I worked at McKinsey & Company, primarily on digital and Advanced Analytics strategy and operational transformation, and later completed my master’s at Harvard Business School.
What has always interested me most is how data and analytics can fundamentally reshape a business. When used well, they can change how decisions are made, how resources are allocated, and ultimately how much value an organization can create.
Could you tell us about your career journey and what drew you to the intersection of data, AI, and supply chain?
What drew me in very early was the more technical side of problem-solving. Even in consulting, I was always most interested in the problems where data, advanced analytics, and technology could materially change the answer—not just support it.
On my first projects at McKinsey, I became very curious about applied machine learning and advanced analytics because I could see the difference between using data descriptively and actually using it to improve decisions. Once I saw how much value could be created by combining business context with more sophisticated analytical methods, I never really looked back.
Since then, I have consistently gravitated toward opportunities where analytics and AI are not just interesting technically, but where they can unlock meaningful business value—whether through better pricing, procurement, operational decisions, or productivity. That combination of technical depth and measurable impact is what has continued to shape my career.
Supply Chain has been a major focus of your work. Why is supply chain particularly interesting for AI?
Supply chain is particularly interesting for AI because it is one of the richest environments in terms of data. Every movement of inventory, shipment, purchase order, route, carrier decision, delivery, and transaction generates information, creating an enormous amount of operational data that can be used to improve decisions.
At the same time, supply chain is fundamentally a complex decision-making system. Companies are constantly making interconnected decisions around forecasting, inventory, procurement, transportation, routing, pricing, and network design, often under uncertainty and changing conditions.
That combination—very rich data, high complexity, and significant economic impact—makes it an ideal environment for machine learning and advanced analytics. Even relatively small improvements can translate into meaningful value at scale through lower costs, better service levels, higher asset utilization, and stronger overall business performance.
Pricing has also been a major part of your work. What makes pricing such a compelling application for AI?
Pricing is a particularly strong use case for AI because it sits at the intersection of cost, market dynamics, customer behavior, and uncertainty. In supply chain, the right price can depend on a large number of variables such as distance, shipment characteristics, fuel, network conditions, service levels, market alternatives, and willingness to pay—and those variables are constantly changing.
AI helps make sense of that complexity at a level of granularity that would be very difficult to manage manually. But what I find most interesting is that pricing is not just a prediction problem. Estimating cost is only part of the answer. The more valuable question is: given the economics, the market, and the customer context, what should we actually charge?
What have you learned about moving machine-learning systems from experimentation into production?
The hardest problems frequently appear after the modeling work is finished.
You need reliable data pipelines. You need clear ownership. You need monitoring. You need to determine what happens when the model produces an unusual recommendation. You need processes for incorporating business feedback. You need to decide when the model should be retrained.
And most importantly, you need to integrate the output into the workflow where the decision actually occurs. A lot of AI initiatives fail because the team proves that a model works but never fully redesigns the process around it.
Production AI requires product thinking on the end-user experience just as much as data science.
What is one of the most impactful applied AI initiatives you have worked on in supply chain and operations?
While it was not the largest initiative I have worked on in terms of monetary impact, one of the most transformative—especially given the time—was the implementation of an advanced analytics solution at a large copper mining operation in Latin America.
The objective was to optimize two critical and naturally competing metrics: throughput, or how much mineral the plant could process, and recovery, or how much of the valuable mineral was actually captured in the final output. Improving one could negatively affect the other, and both were influenced by a large number of operating conditions.
The processing plant was extremely well instrumented, giving us access to more than 200 input variables to build predictive models and identify the operating conditions that maximized overall value. But the modeling was only part of the challenge. A significant amount of work went into change management and frontline adoption, making sure operators trusted the recommendations and actually incorporated them into day-to-day decisions.
The business impact was substantial, generating a multi-million-dollar improvement in output, but what made the project particularly meaningful to me was that it demonstrated, relatively early, what applied AI could look like in a very traditional industrial environment. It was a clear example of AI moving beyond analysis and becoming part of how people actually operated the business.
How should companies measure whether an AI initiative is actually successful?
I would start by asking what business decision the system is supposed to improve. Then define the economic outcome associated with improving that decision. If the use case is procurement, perhaps the outcome is lower purchased cost. If it is transportation pricing, perhaps it is better profitability, competitiveness, or retention. If it is forecasting, perhaps it is lower inventory or improved service levels. If it is a manufacturing facility, perhaps it’s on more output. Any other metrics—such as model accuracy, predictive power, complexity, or computational performance—are ultimately process metrics used to ensure that improvements in the final business outcome can be credibly linked to the work being done.
Accuracy metrics are important because they tell you whether the model works technically. But they are usually intermediate metrics. Ultimately, businesses care about whether the system changed an outcome.
Do you see AI agents becoming important in logistics?
Absolutely, although I think the most useful agents will initially operate within relatively well-defined workflows. Logistics involves thousands of repetitive but context-heavy decisions: pricing, tendering, appointment scheduling, exception management, procurement, routing, inventory positioning, carrier selection, and negotiation.
Many of these workflows combine structured data with unstructured information and require humans to synthesize information from several systems. That is exactly where agents can become valuable. I don’t think the near-term future is necessarily autonomous AI making every major supply-chain decision. I think it is more likely to be AI dramatically increasing the decision-making capacity of human operators.
One person may eventually be able to manage a level of complexity that previously required several people because the agent handles the information gathering, analysis, and preparation.
What do you think will define the next phase of AI in logistics and enterprise operations?
I think we are moving from prediction to decision. The first generation of applied machine learning was largely about producing better forecasts and predictions. The next generation will combine those predictions with context, market information, optimization, and generative AI to recommend what businesses should actually do. And eventually, in the right environments, some of those recommendations will become increasingly autonomous.
They will be the ones that can integrate AI into real workflows, earn user trust, measure business impact, and continuously improve the interaction between people and machines. That is where I think the real transformation will happen.
