What is Tessary's frustration classifier?

It’s a built-in classifier that finds the conversations where users are unhappy with your agent, cheaply and in real time.

Those conversations are hard to find any other way. Most unhappy users never file a ticket, negative ratings catch a few more, and reading every conversation by hand doesn’t scale. What the rest have in common is what error rates miss: an agent that keeps misunderstanding the same request, loops without progress, or makes the user repeat themselves.

The classifier pulls out the context around the user’s reaction, their message and the few before it, and TypeSafe’s Jev decision model scores how unhappy the user is with the assistant, at about $0.04 per 1,000 messages. Unhappiness about something outside the chat, a late delivery or a bad day, doesn’t count.

The limit: it scores what users write. A user who leaves without saying anything gives it nothing to flag.