Agents verify insurance benefits by telephone against rules that are unforgiving, and the mistakes are expensive to unwind. Training happened on real calls, and the feedback arrived after the damage was done. We built this speculatively to show an outsourcing company what could be done about it.
A course, a simulator and a live assistant that work together. The course covers the codes, the coordination rules and the errors that cost the most, taught with flashcards, a decision tree and a spot-the-error exercise. The simulator puts a trainee on a call with an AI caller, and the characters and situations are written as plain text files so adding another costs nothing. The assistant listens to a real call, fills the verification checklist as the answers arrive, suggests the next line to say, and answers questions in the middle of the call.
The checklist fills itself from the conversation while the call is still running, so the agent is reading a completed record rather than typing one. The simulator and the live assistant both work and the assistant is deployed. Nothing here was ever sold. The company it was built for never became a client, the introduction was drafted and never sent, and one lesson of a twelve-module outline was built rather than the whole programme. It is on this page because the work is real and the page is about work.
The checklist, the characters and the scenarios are configuration in the built system. The characters and situations are plain text files composed at the moment a call starts, which is what makes a new vertical cheap rather than a rebuild. This entry previously described the same work as a delivered client engagement. It was not one, and the correction is recorded rather than quietly made.
This page once carried claims that described the work inaccurately. They came down in an audit of our own copy, and each one went back up only when its numbers could be traced to something measured. A case study you cannot check is worth nothing to you, and one that turns out to be wrong costs us more than it ever earned.