The problem
A team of 10–15 operations analysts handled the company’s internal service requests, around 200 tickets per analyst every quarter. The queue was always behind. In the worst cases, setting up a new partner entity took months. People were stretched thin, the team was tense, and good analysts were thinking about leaving.
Most of each ticket wasn’t judgment. It was legwork:
- Barcode rule changes. Find the logic for that barcode format, pull sample barcodes, edit them by hand for every test case, run the tests, write up the evidence.
- File requests. Someone asks for a settlement file. The analyst goes looking for it and sends it.
- New entity setup. Register the entity in the ERP, then open separate tasks for the connectors team, the database team and others. Three or four parallel steps, all by hand.
What the agent does
Each new request is picked up by an AI agent that does the legwork before a person opens the ticket:
- Barcode changes: it finds the logic, builds sample barcodes for each case, runs the tests in a test environment and attaches an evidence document to the ticket. Around 80% of the task is done when the analyst gets there.
- File requests: it finds the file and leaves it on the ticket.
- New entities: it sets the entity up in the test environment and drafts the tasks for the other teams.
People stay in charge
The agent never touches production. Everything it changes stays in the test environment until an analyst approves it and requests the move to production. The tasks it drafts for other teams wait for an analyst’s sign-off, and only people run the checks after go-live. No request is closed by the agent alone, and the approval each request type needs is set up front. It isn’t left to the AI.
Results
- Average handling time went from 15 to about 4 minutes per request. Some request types dropped even further.
- About half of all requests are resolved by the agent, with a person approving before anything goes live.
- Setting up a new entity went from months to days.
- The backlog was cleared. Analysts no longer carry overdue tickets.
Measured quarter over quarter, before and after: average time per ticket and tickets closed. It started as a pilot with one team of non-technical analysts and moved to production once the numbers came in.
The biggest change wasn’t in the numbers. The team stopped working underwater. They used the freed time to catch up first, then to organize work that had been pushed aside for months. One analyst who had been close to leaving said she had started to enjoy coming to work again.
What this looks like at a freight brokerage
Same shape, different queue. A tender arrives by email. The agent reads the PDF, drafts the load in your TMS and flags what’s missing. Your rep checks it and approves. Nothing reaches a shipper, a carrier or your TMS until a person signs off.
Start with a 5-day Diagnostic. We time 20–30 real cases of one process with your team, then hand you the numbers and a fixed-price pilot proposal.