Saved sample run · Meera Iyer, fictional

Flexport

This is a saved example. Answering a flagged line needs the demo.Try the demo

Tell us about an AI product you shipped and how you measured whether it worked.

The role owns a critical part of global logistics and ships features that drive measurable business outcomes.

I shipped an AI triage feature for the exceptions desk that suggests a category and next step for every exception.

I set up the evaluation using 400 past exceptions labelled by two senior dispatchers as a test set.

The first version got 71 percent of categories right, which was not good enough to ship because a wrong category sends the ticket to the wrong team.

We changed it to suggest only when confident, reaching 92 percent accuracy on 64 percent of tickets.

A six-week pilot with three customers cut median time to first action from 19 minutes to 11 minutes.

A weekly human review of 30 suggestions caught a systematic mislabel of weather delays as carrier faults.

134 words of 200 · every line backed or yours

Where each line came from

You'll define the vision and roadmap for your product area, ship features that drive measurable business outcomes, and collaborate with engineering, design, commercial, and operations teams to solve hard problems at scale.Job posting
“The biggest thing I shipped is an AI triage feature for that exceptions desk.”Said aloud
“Every exception that came in got a suggested category and a suggested next step.”Said aloud
“I wrote the spec, I set up the evaluation, and I ran the pilot.”Said aloud
“We took 400 past exceptions that dispatchers had already closed, and two senior dispatchers labelled the right category for each one.”Said aloud
“The first version got 71 percent of categories right on that set.”Said aloud
“That was not good enough to ship, because a wrong category sends the ticket to the wrong team.”Said aloud
“We changed it so that it only suggested a category when it was confident, and otherwise it said "needs a human".”Said aloud
“On the cases where it did suggest, it was right 92 percent of the time, and it suggested on about 64 percent of tickets.”Said aloud
“We piloted it with three customers for six weeks.”Said aloud
“Median time to the first action on an exception went from about 19 minutes to about 11 minutes across those three customers.”Said aloud
“The model started tagging every weather delay as a "carrier fault", because one customer typed weather notes in the carrier field.”Said aloud
“We caught it because I had asked for a weekly sample review of 30 suggestions.”Said aloud