Your call center is the training corpus

Encore AI is betting that the next generation of enterprise voice agents will not be trained on the open internet — but on your own call recordings. Its platform ingests a company's call recordings, emails, text messages, and CRM history, then stages those interactions to figure out which conversational moves actually moved a deal — and which ones stalled. The winning moves get packaged into a voice agent that mirrors the company's strongest reps, including the jokes and the anecdotes, because the playbook is literally built from the calls that worked.

That changes what a customer-interaction platform is. It is no longer software a bank licenses; it is a model trained on that bank's own conversations. The strongest falsifier is also visible: a system built this way will encode the loudest patterns, not necessarily the best ones. If the senior rep who closes the most mortgages is also the one who overrides compliance, the agent will learn that too. "Interaction mining" inherits the habits of the org it watches.

The market is sorting into two camps. One camp sells deflection — keep the customer away from the human as fast as possible. The other, where Encore AI just staked $30 million of its own capital behind it, sells revenue: turn every conversation into an upsell or a save. Team8, Planven, and The Garage put $30M behind the second camp. The round is not the story. The training corpus is.