Agents in action
Worldwide Express agents close 10,000 freight billing exceptions a week and cut the backlog 75%
Freight shipping firm Worldwide Express, part of ShipStation Global, built three AI agents with AWS and LTM on Amazon Bedrock to resolve billing exceptions, invoices whose purchase order, receipt and carrier bill do not match, inside its own transport management system.
What the agent did
- Split the work across three specialized agents, each with its own prompt, tools and carrier knowledge base
- Started with a person approving every recommendation in the existing app; once accuracy passed 99%, bulk approval became the default
- Reached over 40 carriers in six months and now close about 10,000 exceptions a week, roughly half of them, with under 1% errors
- Went from assessment to production in less than 6 months
What it means for you
- Begin with human approval on each agent recommendation and set an accuracy bar, here 99%, before moving to bulk approval.
- Give each exception type its own agent and knowledge base instead of one general agent with every tool.
- Measure the money tied up, not just tickets closed: the cash locked in open exceptions is what made this business case.
Why it matters
As reported, open exceptions fell from $15M to $4M in six months, freeing $10M of cash once, with over $1M in yearly savings and time-to-cash four weeks shorter. About 14 percent of freight invoices fail the standard match, so this is a large, repetitive queue where agents with an approval step fit well.
Figures as reported in an AWS blog post co-written with the customer; the page gives about 50 percent in one place and more than 40 percent in a customer quote for the share automated.
Source: AWS Machine Learning Blog
Every number on this page comes from the source. Our charts redraw those numbers; videos are embedded from the publisher.
Comments
- Tried something like this? Share how it went for you.
More agents in action
Agents gone wrongA Cursor agent wiped PocketOS's production database in 9 seconds with one API call
Agents gone wrongOne PR title was enough: 4,714 GitHub workflows with AI agents could be hijacked to leak secrets
Agents gone wrongA Claude triage bot open to any GitHub user led to an unauthorized Cline CLI release on npm