Gong's CCO: AI Agents Won't Fix a Broken Post-Sales Operating Model

“AI can make a broken process faster. It cannot make that process work.” That’s Gong Chief Customer Officer Simon Frey’s blunt verdict on why deploying more AI agents won’t fix an underperforming post-sales organization if the operating model underneath it stays broken. His evidence for the underlying breakdown: NPS and CSAT email survey response rates have fallen from 20-25 percent in 2019 to just 10-15 percent by 2025, meaning post-sales teams have measurably less visibility into customer sentiment than they did six years ago — right as renewal and expansion motions depend more on catching problems early. Companies are compensating in an expensive, manual way: Forward-Deployed Engineer job postings rose more than 800 percent in 2025, evidence of technical staff being embedded directly with customers to paper over fragmented handoffs rather than fixing the handoffs themselves. Frey points to Gong’s own “Revenue Graph” — which he says captures roughly 100 times more customer interaction data than a typical CRM record — as the kind of unified signal source post-sales teams need but usually lack.

The piece’s core argument is about sequencing, not tooling: the post-sales organizations that win won’t be the ones running the most AI agents, but the ones that redesign around one continuous revenue motion — spanning sales handoff through renewal — instead of layering AI on top of the fragmented, siloed handoffs that define most current models.

There’s a live example of what redesigning first actually looks like. A SaaStr recap of customer-success leader Haya Kamola’s work at Backstory describes rebuilding an account-tiering framework for 141 accounts — a project that previously took her team plus four others a full quarter — in three to four days, by first defining what a “golden customer” looked like through team judgment, then building genuinely new signals (including an AI maturity score per account) rather than just automating the old scoring model. They found one existing signal had been scored backwards the whole time. That’s Frey’s thesis in practice: the speed came after the model was rebuilt, not from bolting AI onto the one that was already broken.