Worldwide Express says Amazon Bedrock agents cut its freight billing exception backlog by 75%

What Worldwide Express built

Worldwide Express, part of ShipStation Global, says it handles roughly $5 billion in systemwide revenue with more than 75 less-than-truckload (LTL) carriers and over 130,000 customers. In a case study on the AWS Machine Learning blog published on 30 October 2026, the company and AWS describe how about $15 million in free cash flow each month was held up by billing exceptions. These are invoices that fail the three-way match between purchase order, goods receipt and carrier invoice, which the post says happens to roughly 14 percent of LTL invoices.

Each exception used to go through a five-step manual process run by an outsourced team: classify it, research the cause, email the carrier or warehouse, wait, then record the fix in the SpeedShip transportation management system. Worldwide Express worked with AWS and the partner LTM to replace most of that with three specialized agents built on Amazon Bedrock. An Invoice Matching Agent finds the shipment record and flags mismatches, a Pricing Variance Agent checks charges against contracted rates, and an Accessorial Validation Agent checks whether fees such as liftgate or residential delivery match actual delivery events.

How the system is designed

The agents run on Amazon EC2 with Auto Scaling and call Bedrock for model inference. The case study says the team built its own framework on EC2 to keep full control over how each agent reasons, where its decision boundaries sit and how exceptions are routed. Each agent has an instruction prompt, a set of actions that call SpeedShip through the same APIs analysts use, and access to a knowledge base of each carrier's tariff rules, surcharge definitions, dispute steps and variance thresholds. The same three agents serve every carrier without carrier-specific code.

A human reviewer approved each recommendation in the SpeedShip interface during the first rollout. According to the post, bulk approval became the default once accuracy exceeded 99 percent. Every resolution comes with a plain-language note, and dashboards track automation rate by carrier, error rate by exception type, resolution time and how often reviewers override the agents. Rejected recommendations are logged and used to refine prompts and knowledge bases.

The numbers, and who claims them

All results come from Worldwide Express and AWS and have not been independently verified. The case study says the agents now cover more than 40 LTL carriers and close about 10,000 exceptions a week, roughly half of the weekly total, with an error rate under 1 percent. It reports the active exception backlog falling from $15 million in June 2025 to $4 million six months later, $10 million in freed cash flow, more than $1 million in annual savings, a 60x first-year return on investment and time-to-cash four weeks faster. A quote from Nathanael Pearson, VP of FinOps at Worldwide Express, puts automation at more than 40 percent of weekly billing exceptions, slightly below the roughly 50 percent figure in the body of the post.

The case is a clear example of where agents fit better than rule-based automation: the post argues that judging a carrier's accessorial policy against shipment data needs context that fixed RPA rules handle poorly. It also shows the controls that made finance teams accept it, from per-carrier knowledge bases to a human approval gate that was only relaxed after accuracy was measured. A similar logistics deployment is covered in our C.H. Robinson customs agents story, and more case studies are collected under AI Automation.

What to watch next

The case study does not name the foundation models the agents use, the cost of running them, or the size of the team that maintains the prompts and knowledge bases. Worldwide Express says it is extending the approach to other parts of the order-to-cash cycle, where an earlier AWS assessment identified 19 automation opportunities.

Questions people ask

What AI does Worldwide Express use for freight exceptions?

Three custom AI agents running on Amazon EC2 that call Amazon Bedrock for model inference, according to the AWS case study. The specific foundation models are not stated.

How much did Worldwide Express save?

The company reports more than $1 million in annual cost savings, $10 million in freed cash flow and a 60x first-year return on investment. These are company figures, not independently verified.

Do humans still review the agents' decisions?

Reviewers approved each recommendation at first. The case study says bulk approval became the default after accuracy exceeded 99 percent, and reviewer rejections are still logged.

Sources

1. How Worldwide Express automated freight exceptions with Amazon Bedrock (AWS, 30 Oct 2026)
2. About Worldwide Express (Worldwide Express, 12 Apr 2025)