The Secret Sauce Behind Why Thousands of Employees Rely on AI to Make High-Stakes Decisions

In a world where billions of packages are scanned every day across networks, competitive advantage is determined by speed and accuracy. Pavan Mullapudi is leading a new era of decision intelligence at Amazon, where even a few minutes of delay can result in millions of dollars in expenses and unhappy customers. This new era of decision intelligence is driven by causal AI rather than intuition.

An experienced leader in Amazon's Logistics division, Pavan has laid the foundation for what insiders now call the company's “attribution system,” a state‑of‑the‑art inference engine that not only identifies issues like missed delivery estimates, but also determines the causes of them. The way thousands of workers, ranging from finance heads to sort‑center managers, prioritize their next steps is changing as a result of the transition from dashboards that only show the surface to statistically sound, root‑cause intelligence.

A probabilistic graph model that analyzes a package's path across various processes through several nodes and identifies the most significant upstream triggers for downstream disruptions is at the core of his innovation. This model has moved from theory to an operational pilot—already used by a team to identify root causes of delivery misses and now being expanded to additional areas. It reduced the time needed for root‑cause investigations by 60 % in one pilot, saving hundreds of analyst hours each month. Networks are projected to experience a 22 % decrease in missed Delivery Estimate Accuracy (DEA) within two weeks of focused interventions and annual savings from this transformation are pegged at $10 million, thanks to reduced overtime, re‑handling, and experience penalties.

But Pavan’s impact doesn’t stop at detection. As a techno‑functional innovator, he also led the development of a Gen‑AI strategy that automates the generation of SOPs and defect narratives. This retrieval‑augmented LLM framework reduced SOP generation time from days to under an hour per station, speeding up decision‑making cycles across the board.

Beyond the numbers, perhaps his most nuanced success has been cultural. When Pavan first introduced causal models to site leaders, skepticism was high. Operations teams, used to on‑the‑ground intuition, were wary of a science group prescribing fixes from afar. He changed the narrative by co‑creating solutions with them, pairing every causal claim with statistical confidence intervals and real‑world dollar impacts. His federated schema, unifying data across disparate systems, made it possible to deploy these tools without disrupting localized workflows.

This blend of science leadership, field empathy, and product acumen has culminated in two major Amazon Science publications. His first, “Beyond Mere Automation: A Techno‑Functional Framework for Re‑Imagining Gen‑AI in Supply‑Chain Operations,” redefines how generative AI can accelerate SOP consistency and defect response. The second, “An End‑to‑End Causal Modeling Framework for Advanced Attribution in Supply‑Chain Operations,” accepted for KDD 2025, showcases empirical results and provides a replicable blueprint for global operations.

According to Pavan, federated causal learning and Gen‑AI‑powered digital twins—which allow planners to query, simulate, and validate strategies in natural language—are the next frontier. He notes, “Descriptive dashboards are no longer enough. Your model will never be used if it cannot convert its lift into money saved or time saved. Explainable AI with practical applications is the way of the future.”

Pavan Mullapudi's work provides not only a glimpse of what is possible but also a tested route to sustainable, AI‑driven operations as global supply chains become more intricate and decentralized. His causal intelligence engine has revolutionized efficiency at scale by enabling thousands of employees to make decisions not only more quickly but also more intelligently.