Most companies are grading their customer-facing AI on deflection — how many people it stopped from ever reaching a human. Dvir Ginzburg thinks that's like grading a website by how fast people leave it.

Dvir is the founder & CEO of Encore AI, holds a PhD in geometric deep learning, and spent years as a recommendation systems researcher at Microsoft, with two granted patents to his name. Encore AI builds revenue-generating conversational AI for banks and lenders — running millions of customer calls a month across voice, chat, and IVR.

In this conversation with LeanScale co-founder Anthony Enrico, Dvir dismantles the deflection metric, explains why 'sounds human' is already solved while 'acts human' is the real moat, and tells the story of a customer who lied to an AI agent rather than tell it no. They get into interaction mining, cloning your top performers, the 80-20 rule that separates demos from production, and why the next wave of AI is about effectiveness, not efficiency.

WHAT YOU'LL LEARNWhy deflection is the wrong metric for customer-facing AI — and what to measure insteadHow AI deflection can quietly tank revenue (the travel-insurance PPD story)Why 'sounds human' is already solved and 'acts human' is the real battlegroundWhy your moat is your own conversation data, not the foundation modelHow interaction mining finds exactly where revenue is leaking inside a businessHow AI agents learn to clone the tactics of your top performersThe customer who lied to an AI agent — and why that's a signal of trustWhere the real ceiling on AI agents is (hint: it's integration, not technology)The 80-20 rule that separates an impressive demo from productionWhy the next wave of AI is about effectiveness, not efficiency

CHAPTERS00:00 — Cold open + intro01:52 — The deflection trap: grading AI by how fast people leave03:45 — From 'show my board I have AI' to 'what's the ROI?'05:20 — Travel insurance: deflection up, revenue down07:01 — The $100-for-$90 problem and the subsidized AI party08:00 — Deployment strategists and interaction mining11:43 — Sounds human vs. acts human — your moat is client data13:30 — Cloning your top performers15:44 — The customer who lied to the AI about his wife19:25 — Do agents actually beat humans?22:42 — The real ceiling is integration, not technology26:01 — The next barrier: 'LLMs see quantity as quality'28:10 — The omni-channel shift: 'text Jim'31:23 — Build vs. buy and the 80-20 rule of production36:44 — The last mile + wrap: effectiveness over efficiency

ABOUT THE GUESTDvir Ginzburg is the founder & CEO of Encore AI, which builds revenue-generating conversational AI for banks, lenders, and financial-services enterprises. He holds a PhD in geometric deep learning and spent years as a recommendation systems researcher at Microsoft, with two granted patents to his name. Encore AI's technology combines agentic landing pages with dynamic customer journeys and 'interaction mining' — a layer that classifies every customer-facing conversation (email, chat, text, and voice) to find where revenue is leaking and reverse-engineer how a company's top performers win. Encore AI runs millions of customer calls a month across voice, chat, and IVR, deployed as fully autonomous agents or as a live 'wingman' assistant to human teams.

ABOUT THE LEANSCALE PODCASTThe LeanScale Podcast is the show for GTM operators building the next generation of revenue infrastructure. Hosted by Anthony Enrico, co-founder of LeanScale, we go deep with the operators, founders, and executives shaping how modern go-to-market teams are designed, scaled, and transformed.

▶ Learn more: https://www.leanscale.team▶ Follow Anthony on LinkedIn▶ Follow Dvir on LinkedIn

#AI #ConversationalAI #VoiceAI #AIAgents #AgenticAI #RevOps #GTM #CX #Fintech #Banking #AITransformation #EnterpriseAI #EncoreAI #LeanScale #LeanScalePodcast

Why Deflection Is the Wrong Way to Measure AI | Dvir Ginzburg, Founder & CEO of Encore AI

The LeanScale Podcast with Anthony Enrico

Why Deflection Is the Wrong Way to Measure AI | Dvir Ginzburg, Founder & CEO of Encore AI