Reading time: 5 min
Table of Contents
Key Takeaways
- Interaction mining at scale: Encore AI analyzes thousands of conversations to build agent playbooks from proven strategies.
- Enterprise traction: Over 40 customers, mostly financial institutions, with 5x ARR growth in 18 months.
- Structural moat: The approach requires deep integration with existing CRM and call data — a barrier for big CRM vendors.
The Demo vs. Production Gap
Most AI voice agents look good in a controlled demo. Put them in front of real customers, with real hesitation and real edge cases, and they fall apart. Here’s what actually happens in production: the agent misses context, repeats itself, or fails to adapt when the customer pivots. That’s not automation — that’s a liability.
Encore AI, founded in 2022 as Insait IO by Dvir Ginzburg, takes a different path. Instead of training on sanitized datasets, they mine actual customer interactions — recordings, emails, texts, CRM records — and extract what worked. The result is an agent that doesn’t just follow a script but runs the playbooks that generated revenue for the best reps. The seed round funded the platform. This Series A proves it scales.
What $30M Buys in Production Reality
The round is led by Team8, with participation from Planven, Lukatz, Garage, and several banks and insurers. Some of those financial institutions were customers first. That’s the kind of validation that matters: they paid for the product, then doubled down with equity.
The capital goes to expanding U.S. sales operations and deploying with more large financial institutions. Let me be specific: this isn’t about vaporware. Encore now has over 40 enterprise customers and claims ARR grew 5x since the seed round. Ginzburg won’t share exact numbers, but the growth rate is real enough to attract serious backers.
How the Architecture Works
Most people get this wrong: they think training an AI agent means feeding it general data. Encore does something else entirely. Their platform collects call recordings, emails, text threads, and links them to CRM events. Then it divides each interaction into stages and scores which moves advanced the deal and which stalled it.
This is interaction mining, not machine learning. It’s a structural layer above raw AI models. The agent becomes a package of playbooks — including the jokes, the anecdotes, the specific phrasing — that humans used to close or resolve. Ginzburg says the agent sometimes tells the same jokes as the top sales rep, because that rep’s pattern won.
I’ve seen similar approaches fail because companies over-engineered the stack. Too many models, too many microservices, too much infrastructure debt. Encore keeps it focused on one thing: mining the conversation history and building agents from it. That’s production-grade simplicity.
The Real Cost of CRM Vendor Lock-In
The obvious question: what happens when Salesforce or HubSpot builds the same thing? The real cost is the migration path. Bad companies will sell you automation that ties you deeper into their ecosystem. The smarter ones build portable solutions.
Ginzburg argues that major CRM vendors don’t treat conversation history as core data. For them to rearchitect their stack to do what Encore does would require massive engineering upheaval. This isn’t theory — it’s why Salesforce’s Einstein AI still can’t replicate what a focused startup delivers on the first deployment. The demo worked. Production didn’t. Here’s why: structural inertia.
The Startup Reality Check
Not every company can rebuild their entire tech stack. For startups with limited resources, the path is incremental: start with one customer segment, one interaction type, one set of playbooks. Encore’s model — agent as assistant, not replacement — is the right incremental approach.
The agent can work side-by-side with human teams, recommending responses during live conversations. It’s not an oracle. It’s a force multiplier built on what’s already working in your organization. That’s automation that holds under load.
We built Rebirth Distribution around similar principles. OpenClaw and Hermes only deploy when the stack is proven at scale, not just in staging. Encore does the same thing in the voice agent space. It’s good to see.
What to Watch Next
Encore’s Series A is a vote of confidence for the enterprise AI agent playbook approach. But the real test is whether they can hold the production line as they scale. The financial institutions that invested were customers first — that’s the best kind of QA.
I’ll be watching their deployment velocity and churn rate. If they keep the architecture lean and resist overcomplicating the stack, they’ll give the incumbents a real headache.