Agent Assist, Reimagined
Balto CEO Marc Bernstein leads a conversation on the state of agent assist. We'll cover the UX revamp, intent-based guidance, and the platform shifts behind it all.
RTG3: Agentic Intelligence for the Frontline
Agent Assist is now the #1 tech investment priority for contact center leaders. But most AI rollouts backfire due to a top-down push, reluctant workforce, or "us-vs-them" silos. Balto's bet: use AI to power people, not replace them.
The Idea
Insights don't belong in a dashboard. They belong on the call, live. That's where data turns into wins. Balto’s customer base sees 10–50% higher conversions.
What the agent gets, live on every call:
- Find what lifts conversions
- Hold every call to your quality standards
- Coaching that moves the business
- Catch compliance issues before they happen
- AI notes that erase after-call work
RTG3
A new intelligent workspace for frontline agents. Built to feel like an iPhone app, personalized like Netflix, and act as the agent's home base.
Under the Hood
Four stages run every ~800ms, all call long:
- Listen: speech-to-text
- Understand: 4B embedding model fingerprints the meaning
- Focus: re-ranker narrows thousands of options to a few
- Decide: optimized LLM updates the guidance live
Why it's hard: naive frontier-model approach ≈ $10/call → millions/day at Balto's volume (hundreds of billions of tokens daily). Balto's edge: 100s of millions of real conversations distilled into compact custom models — a 4B embedder + 26B MoE adjudicator, small enough for real time, sharp enough to rival far bigger models. Demo takes a weekend. This takes years.
Getting Started
- Available now: included in the Balto RTG Platform
- Easy setup: reach out to your CSM to enable
- Full support: upload docs, connect SharePoint, Balto optimizes for LLMs
