While much of the enterprise AI market has focused on deflecting calls and cutting headcount, a New York startup just raised $30 million on the opposite bet: that AI agents should be generating revenue from customer conversations, including recovering balances.
Encore AI, founded in 2022 as Insait IO and recently rebranded, announced a Series A round led by Team8, Planven and The Garage, with participation from Lukatz and several large commercial banks and insurers. Notably, those financial institutions were customers first and invested after seeing results in their own operations.
The company’s approach should sound familiar to anyone who has spent time thinking about agent performance in a collection environment. Its patented “Interaction Mining” technology ingests call recordings, chats, emails and CRM data, breaks customer interactions into stages, and identifies which specific behaviors from an organization’s top performers actually move outcomes. It then trains AI agents to replicate those behaviors, deploying them autonomously or as real-time assistants alongside live agents across voice, chat, IVR and text.
Chief Executive Dr. Dvir Ginzburg, a former Microsoft recommendation-systems researcher, told TechCrunch the agents sometimes even repeat the jokes and anecdotes top performers use, because the platform runs the playbooks it observes working. The company says its agents can convert leads, close applications, recover balances, and drive upsell and cross-sell while meeting the compliance requirements of heavily regulated industries. Financial services firms make up the majority of its more than 40 enterprise customers, and Ginzburg says annual recurring revenue has grown more than 5x since the seed round less than 18 months ago.
Deployment reportedly takes weeks rather than months, and one of the company’s largest lending clients claims a 10x return on investment within months of implementation.
The competitive question is whether Encore’s head start holds. Large CRM providers like Salesforce and HubSpot sit on similar data, but Ginzburg argues they would need to rebuild their entire implementation stacks to make historical conversation data the foundation of their agents.
