The Warehouse Knew Three Days Late: Rethinking How Supply Chains See Trouble Coming

A supplier falls slightly behind and does not mention it. A port slows down. Inventory at one warehouse dips below a safe level while everyone is focused somewhere else. Each issue can look small on its own. When several happen at the same time and are noticed too late, they become the big problems customers actually feel, like empty shelves or late deliveries. By then, the easy and low cost fixes are often gone.  

For the teams running these networks, the biggest challenge is timing. Most tools are good at reporting what already happened. They are less effective at showing what is starting to go wrong. That is the gap enterprise AI researcher Nikhil Reddy Donapati is trying to close. His work describes a framework designed to spot supply chain risk as it forms, early enough that planners still have options.

A practical mix of old and new

What makes the approach different is how it combines two methods that rarely work together.

The first is Failure Mode and Effects Analysis, also called FMEA. It is a long used engineering method for identifying where a process is likely to fail and scoring the risk. It brings structure and discipline to the question, “Where are we most vulnerable?”

The second is the use of large language models. These models can take in messy, real world information and help interpret context that strict rules often miss. In this framework, FMEA provides the scoring system, while the language model helps make sense of the situation around the numbers.

Together, the two pieces complement each other. FMEA creates consistent, comparable scoring. The language model adds situational understanding, helping teams see not just the risk level, but what is driving it and what it could lead to next.

Built to work in real operations

Nikhil organizes the framework as a six layer architecture that moves from raw operational data, to risk scoring, to reasoning, and finally to clear guidance that a person can act on. The emphasis is on making the output useful in a real business setting, not just in theory.

According to Nikhil, the framework was tested on more than 180,000 supply chain records. It showed strong recall for disruptions, meaning it caught a large share of the problems that were actually developing.

The analysis also identified specific inventory and delay thresholds where the chance of disruption rises sharply. These thresholds are especially helpful because they give operations teams clear tripwires to monitor. Instead of relying on a vague feeling that something seems off, teams can use concrete signals that say, “Pay attention now.”

Why a few days can change everything

In supply chain work, time is often the most limited resource. A warning that arrives three days earlier can be the difference between quietly expediting a shipment and having to explain to a customer why an order is late.

Catching risk early expands the list of possible responses. It allows teams to choose fixes that are faster, cheaper, and less disruptive. Once a problem fully hits, the remaining options are usually more expensive and more damaging.

Nikhil does not claim the system is perfect. Supply chains are noisy, and no model will catch every issue. His goal is more focused and more realistic: surface meaningful warning signs early, explain them clearly, and give people more time to respond.