Gong and Chorus have solved the recording and transcription problem. Every customer call is captured. The transcript is searchable. The system can surface moments where a competitor was mentioned or a pricing objection was raised. For sales teams, this is operationally useful and well-understood.
What is less understood is how to use the same transcript corpus for customer success analysis, specifically to understand which topics cluster around accounts that ended up churning versus accounts that renewed. The recording is not the analysis. The transcription is not the insight. The analysis that connects call content to customer outcomes is mostly not being done, and it requires a different approach than the real-time call coaching that Gong's core use case is optimized for.
The structural difference between sales call analysis and CS call analysis
Sales call analysis in Gong is fundamentally a single-call problem. Each recorded call is analyzed for specific patterns: did the rep ask discovery questions? Did the customer express urgency? How much did each party speak? These metrics are meaningful at the level of the individual call and can be aggregated over a rep's performance over time.
CS call analysis is an account-level problem over a longer time horizon. The question is not "what happened in this call?" but "how has the conversation between this account and our team evolved over six months, and what does that evolution predict about the renewal?" That question requires correlating content across multiple calls on the same account, identifying which themes appeared early versus late in the relationship, and then comparing those patterns against renewal outcomes.
This is a retrospective analysis problem, not a real-time coaching problem. It requires the full call history for an account, not the last call. And it requires comparing across accounts, not just within one account's history, to establish which patterns are predictive versus which are simply present in all customer relationships.
What the data actually contains and how to approach it
A typical CS team with Gong will have call transcripts in two forms: the raw ASR (automatic speech recognition) transcript, which is a verbatim word-for-word record, and the AI-generated summary, which is a structured condensation of the call into topics and action items.
For account-level analysis over time, the summaries are a more useful starting point than the raw transcripts, precisely because they have already extracted the topical structure. A six-month account history might have 15 to 20 calls. Reading the raw transcripts of all 20 is not practical. Reading the summaries and identifying which topics appeared when is tractable.
The topic extraction problem is where the real NLP work lives. Gong's AI-generated summaries use consistent phrasing for common sales and CS topics, which makes them somewhat amenable to keyword-based extraction. But the topics that predict renewal risk are not always named explicitly. An account where the champion starts raising "process questions" about how other customers use the product, or where call summaries start including "timeline clarification" items, may be in evaluation mode. Those patterns require semantic understanding, not keyword matching.
The specific call topics that cluster around churn risk
Based on the structure of what CS calls typically contain, certain topic clusters are worth looking for in the call history of accounts that end up churning.
Escalation callbacks are one. When call notes reference "following up on the recent escalation" or contain language about a service failure that prompted the call, the tone and resolution of those calls matters. If the escalation-callback call resolves with the customer expressing explicit satisfaction, that is a positive signal. If the summary shows the customer was "acknowledged but noted ongoing concerns," that phrasing is worth flagging.
ROI justification requests are another. Early-relationship calls are typically about implementation and adoption. When a CS call summary starts including items like "customer asked about reporting metrics to show value to their leadership team," that shift in topic is meaningful. It suggests the customer champion is being challenged internally on the value of the product, which precedes many non-renewal decisions even when the product itself is working well.
Scope reduction inquiries deserve attention too. Questions about whether certain seats or modules could be removed, whether pricing is tiered in a way that would allow for reduction, or what the process would be for an account size change: these are distinct from standard billing questions and should be tracked separately. A customer who is shrinking rather than churning is still a retention event, and scope reduction requests in call notes are an early indicator.
The retrospective analysis that most teams have not run
The analysis that is most directly useful but rarely done is a retrospective cohort comparison: pull the call history for accounts that churned at their last renewal, pull the call history for comparable accounts that renewed, and look at what topics were discussed in the 90 days before the renewal date for each cohort.
This analysis tends to surface a few consistent patterns. The churned-account cohort typically shows more escalation callbacks, more ROI justification discussions, and more scope-related conversations in that 90-day window. The renewed-account cohort tends to show more expansion-oriented topics: questions about new features, interest in adjacent products or integrations, discussions of how to grow usage in the account.
Running this analysis does not require a data science team. It requires pulling the Gong call summaries via API, labeling them by account and date, grouping them by renewal outcome, and doing a basic topic frequency comparison across the two groups. The labeling step is where human judgment is required to map the free-text summaries to topic categories. But the analysis itself is straightforward once the labeling is done.
What call note analysis cannot do on its own
Call notes are a high-quality signal for the CS relationship level of the account: what the champion says, what the AE hears, what gets discussed in scheduled touchpoints. What they do not capture is the ground-level user experience, which is where support tickets live.
An account can have positive call note sentiment right up until a renewal is declined, if the champion is managing the conversation with the vendor while their users are accumulating support ticket frustrations that the champion is not surfacing. The call record and the ticket record are different windows on the same account, and they can diverge significantly.
The full picture requires both. Call notes for the relationship-level signal. Ticket data for the user-experience signal. Email threads for the async communication between those two layers. Sturdy ingests all three and looks at them together at the account level. The reason we built it that way is that any single channel gives you part of the picture, and the most dangerous renewal risks are often the ones where the channels diverge in ways that would not be visible from either one alone.
Call note mining is one of the more underutilized analyses in CS operations today. The data is there in Gong. The tooling to access it via API exists. What has been missing is a clear methodology for what to look for and how to connect it to renewal outcomes. That connection is what makes the analysis useful rather than just interesting.
