Don’t Let AI Insights Go to Waste ft. Mark Vovsi (Proofpoint)

Imagine walking into your new CS operations role with no named accounts, no digital motion, and no success plans. Some might fold under the pressure, but Mike Vovsi saw an opportunity to build exactly what Proofpoint needed.

On this week’s episode of the [Un]Churned podcast, “205. Headless AI vs. CSP: Where CS Insights Should Live,” Mark Vovsi, Senior Director of CS Operations (GTM Automation & AI) at Proofpoint, sits down with host Josh Schachter to walk through what actually got built, in what order, and why. They also cover Proofpoint's four-stage AI maturity build, why generic AI tools fall short for CS work without a data connection, and the debate every CS leader is about to have about where AI-driven insights should actually live.

---

Listen on YouTube, Spotify, Apple Podcasts, and Gainsight.com.

Subscribe now and get the [Un]Churned take straight to your inbox.

---

🎯 The [Un]Churned Take: Why Your AI-Powered Health Scores Are Gathering Dust

The customer health score has been the golden child of Customer Success for over a decade. We’ve made countless investments as an industry to achieve more accurate scoring through product telemetry, support data, and eventually by incorporating conversation signals. Scores were getting better, but churn still slipped through the cracks.

Health scores may be useful for an aggregate view of your entire customer base, but what about the CSM that has to take action? A CSM who logs in and sees seventeen accounts flagged yellow has information, but they likely lack clarity on which of those seventeen actually needs attention today, why, and what specifically to do about it. The score became part of the noise. By itself, it’s just a measure. What matters is what happens next.

When AI Enters the Equation

Mark Vovsi, Senior Director of CS Operations (GTM Automation & AI) at Proofpoint, focused his first two years on building the data foundation that his early churn detection AI tool would eventually read. This included assigned accounts, a functioning CSP, clean data, and digital motion.

“You can’t do anything in digital, you can’t launch a new CSP, you can’t automate anything—you’ve got to organize [data].”

What he’s building now is an early warning system that pulls together structured telemetry, unstructured conversational data, human-reported signals, and ML-based prediction through a partnership with QuadSci. The output is a risk summary for the account, including the key issues driving it and recommended next steps, generated and organized before the CSM opens the account. They receive a narrative brief rather than a dashboard to interpret.

Most CS orgs have invested heavily in improving the prediction. But the real shift that Mark realized came from designing an output with the CSM’s workday in mind.

Surfacing Risk and Acting On It Are Two Different Problems

Surfacing risk is a data and modeling problem. Acting on it is a product and workflow problem. Mark’s build at Proofpoint went through foundation first, then role-based AI assistants with customer data access, then auto-loaded success plans, and now the early warning system. Each layer required the one before it to be stable.

Customer Success has always sought to detect churn risks early. With AI, we now have unprecedented access to data for predicting those risks. But more data isn’t the answer on its own.

The real value comes from building systems that translate signals into clear, actionable steps, empowering CSMs to respond effectively.

---

🎧 Listen for These Moments

For any CS leader thinking through what an AI transformation actually looks like in sequence, this one is worth the full listen. Here are three moments to listen for in this week’s episode:

• The success plan architecture: Mark describes how Proofpoint went from CSMs who didn’t know how to create a success plan to auto-loading product-specific plans for every customer, every quarter. Who owns the templates, how the guardrails get built, and why it’s not just a time saver.

• Why Amazon Q beat ChatGPT for post-sales: Learn why Mark thinks generic AI tools fall short for CS work, and what “connected to the data” actually has to mean before a role-based assistant becomes useful.

• The 12-month vision: Hear what a successful early warning system looks like from the CSM’s perspective a year from now, and why Mark thinks CS has finally caught up to a promise it’s been making for fifteen years.

---

🔎 Where to Find the Speakers

• Mark Vovsi — LinkedIn: https://www.linkedin.com/in/vovsi/

• Josh Schachter — LinkedIn: https://www.linkedin.com/in/jschachter/

📎 Referenced in This Episode

• Proofpoint — enterprise cybersecurity platform

• Amazon Q — AWS’s AI assistant for business, used by Proofpoint’s post-sales team as their connected, role-based AI tool

• QuadSci — predictive analytics platform powering the ML layer of Proofpoint’s early warning system

• Gainsight MCP — connects the entire Gainsight platform to any MCP-enabled LLM

---

Wrapping Up

Mark started with nothing and built a system that can see churn coming from a year away. What future are you looking to build?

See you next week 🧠