Worldwide Express used three artificial intelligence agents built on Amazon Bedrock to work billing disputes, according to an AWS case study published Oct. 30. The company works with more than 130,000 customers and more than 75 less-than-truckload carriers, and about 14% of carrier invoices didn’t match the purchase order and delivery record. By June 2025, unresolved cases held $15 million in frozen cash flow. Within six months, the backlog of active exceptions fell from $15 million to $4 million, freeing about $10 million in cash flow, cutting time to cash by four weeks and saving more than $1 million a year. Worldwide Express, part of ShipStation Global with about $5 billion in systemwide revenue, earned a 60-fold first-year return on the project, the case study said. Each exception goes to an invoice-matching agent, a pricing-variance agent or an accessorial-validation agent, and the same three agents cover every carrier without carrier-specific code. Human reviewers first approved every recommendation, with bulk approval becoming default once accuracy passed 99%. Nathanael Pearson, Worldwide Express’ vice president of FinOps, said the company automated more than 40% of weekly billing exceptions; the results section puts the figure at about 50%, or roughly 10,000 exceptions a week, with an error rate under 1%. The project went from an assessment in June 2025 to production in under six months.