Why I made it
Here's what happens when a call centre agent can't find an answer.
They put the customer on hold. Ask a coworker. Flag down a floor supervisor. Post in the team chat and wait. If nothing works, they transfer the call.
One agent described exactly that in a thread I analysed. The same person said their knowledge base was good. The information was there. They just couldn't reach it fast enough.
That's three or four people pulled in for one question. Nobody measures it, but it counts against the agent's handle time (how long each call takes), which is what they're judged on.
So I looked into it properly: 45 agent accounts, two Cornell surveys of about 5,400 workers in the US and Canada, and an economics study of a real AI rollout.
The pattern: agents are the last stop for problems that started somewhere else. In 2017, 80% said customers often blamed them for things they couldn't control. By 2023 it was 51%, still one in two.
And the only type of AI workers rate positively is the kind that helps them find information. Not the kind that watches them.
Six years in call centres taught me the problem. A year of building with AI taught me why most tools for it fail.