Episode 013 • April 16, 2026

Listen on: Spotify | Apple Podcasts

Call intelligence is more than recording calls. Jay's team extracts every transcript, summary, and participant list from Fathom into a Postgres database with a vector layer for semantic search. The goal isn't just storage—it's security (some calls are private), speed, and the ability for every teammate to query across all company conversations, not just their own. A large language model is only as good as the context you give it.

The "context harness" is the real unlock. Handing everyone a ChatGPT or Claude subscription isn't an AI strategy. Without a centralized harness—a shared, queryable knowledge base built around your specific data, processes, and people—every individual is starting from zero. Jay argues this is the single most important infrastructure investment CS leaders and operators need to be making right now.

Balboa OS turns company knowledge into an agentic operating system. Inspired by a product manager who documented her entire team's workflows in Markdown, Jay built Balboa OS: a centralized folder of plain-text files covering everything from role definitions and scoping approaches to the employee handbook and team roster. It auto-distributes to every teammate's machine and connects directly to Claude via Claude Cowork. When someone needs to turn a discovery call into a deliverable, Claude finds the right skill and the right transcript on its own.

Autonomous agents will evolve your systems without you. The next layer Jay is building: agents that continuously scan new call transcripts and update the company's skills and documentation with best practices that surface naturally in client conversations—even ones leadership wasn't on. An agent is currently building a full company wiki from thousands of calls. "It's a continuous learning and evolution opportunity for our companies."

CSM ratios are at a breaking point. Companies that previously assigned a CSM to any account are now drawing lines at $100K or $250K in ARR. The CSM is becoming a high-value consultant, not a generalist relationship manager. The answer, Jay argues, is treating customer success as an operating principle—not a single role—and investing in digital, one-to-many engagement models that scale the expertise CSMs carry.

Most teams are still stuck at AI as a personal sidekick. The forward-looking teams Jay has seen are moving toward autonomous agent loops that sense signals (product telemetry, emails, call recordings), decide what to do, take action, and then feed results back into the system to improve future decisions. Tools like Pendo Predict are early examples. The human review step still matters in B2B, but the loop itself is becoming continuous and self-correcting.

The product-to-CRM mapping gap is the real NRR problem. At Pendo's annual conference, nearly every SaaS leader Jay spoke with said the same thing: our data is a mess. The foundation that unlocks almost everything—CSM intelligence, product-led growth, retention analytics—is a simple mapping of product tenant IDs to CRM account records and user IDs to contact records. Jay estimates that 90% of SaaS companies would be in a fundamentally different state if they had just that.

Onboarding is still the highest-leverage moment in retention. Jay's team helped a $300M company cut its year-one attrition by fixing a broken, disorganized onboarding program. The result was a 10+ point improvement in retention. In an era of consumption-based pricing, where revenue no longer turns on automatically at contract close, getting customers activated quickly is even more critical—and the leading indicators (onboarding completion rates by cohort) have to stand in for the lagging signal of retention until the results show up.

Fathom – AI meeting recorder and call intelligence tool; used by Balboa for all Zoom recordings, transcript storage, and call summaries

Postgres – Open-source relational database; how Balboa stores structured call data in their central intelligence hub

Claude (Anthropic) – AI platform Balboa uses as their core system; referenced throughout for skills, agentic workflows, and Claude Cowork

Claude Cowork – Anthropic's desktop tool for connecting local files and folders to Claude conversations; how Balboa distributes Balboa OS to the team

MCP (Model Context Protocol) – Protocol used to expose Balboa's call intelligence database to Claude so teammates can query it directly

Pendo – Product analytics and in-app engagement platform; referenced as a key signal source in agentic customer journey loops

Pendo Predict – Pendo's AI feature for predicting customer health and recommending actions based on behavioral signals

Braze – Customer engagement platform referenced as a channel for in-app and email engagement alongside Pendo

Salesforce / HubSpot – CRM systems referenced as the account-of-record side of the product-to-CRM mapping gap

Looker / Tableau – BI and dashboarding tools referenced for reporting against a unified customer data layer

Akash Gupta's Growth Podcast – Podcast that inspired Jay's Balboa OS concept; featured an episode with a PM who built a team operating system in Markdown

Pulse (Gainsight) – Gainsight's annual customer success conference; Jay and team attending May 27 in Las Vegas