Why usage data sees churn last
Most churn models start with product usage because it is the cleanest data a company owns. It is also the slowest. By the time logins fall, the account has usually already decided; the decline is the exit, not the warning.
The warning is in the words. Support tickets and customer calls carry frustration, comparison and resignation weeks before behaviour changes, because people say what they are about to do before they do it. A churn read built on support and call signals is earlier by design, and it explains itself: the evidence is a quote, not a coefficient.
- The order of signals. Words first (tickets, calls), relationships second (champion left, sponsor silent), behaviour third (usage, seats), commercial last (renewal notice).
- What this article covers. A method for reading the first two from the systems you already run, scoring accounts with a rule you can defend, and acting before the renewal call is booked.
The churn signals inside support tickets
Tickets are a churn instrument if you read them as a series per account rather than a queue. Five patterns matter, and none of them is “more tickets”.
- Repeat on one theme. The same organisation raising the same job three times in a quarter. It means the workaround failed, not that they are engaged.
- Tone shift. A move from questions to statements: “how do I” becomes “this does not”. Read the words, not the sentiment score alone.
- Escalation language. Mentions of a manager, a contract, a deadline or “we need this resolved by”.
- Comparison. A competitor named in a ticket. The account is already shopping; this is late, but it is unambiguous.
- Silence after a spike. A burst of tickets followed by nothing is worse than a steady trickle. The customer stopped asking.
Method: export tickets with organisation, plan, ARR and date. Tag each ticket for the five patterns above. Roll up per organisation per month.
The churn signals inside customer calls
Call transcripts, from a recorder such as Gong or from CS notes, carry what tickets cannot: who was in the room and how they talked about the future.
- Absent champion. The person who bought is no longer on the call. Note attendance, not only content.
- Future tense goes missing. Healthy accounts talk about next quarter. At-risk accounts talk about this quarter’s problems.
- Competitor named unprompted. The account raised the comparison, not your rep. Weight this above a mention that answered a question.
- Budget and procurement words. “Renewal”, “budget review”, “procurement is asking”, “consolidating vendors”.
- Unresolved commitments. Things your side promised on a previous call that are still open. Count them.
Method: for each account, list calls in the window, tag the five patterns, and record who attended. Keep the quotes.
Combining signals per account
Neither source is enough alone. Tickets tell you what is broken; calls tell you what the account intends. Put them on one timeline per account, with ARR and renewal date, and the picture appears.
- One row per account. Columns: ARR, renewal date, ticket patterns this month, call patterns this month, competitor mentions, open commitments, champion present.
- Look for co-occurrence. Repeat-theme tickets plus an unprompted competitor mention in the same month is the single strongest pair in our experience of reading these series. Either alone is noise; together they are a plan.
- Add the market clock. If several accounts mention the same competitor in one fortnight, look for the move at that competitor. The reaction has a cause, and the cause tells you what to counter.
Scoring with a rule you can defend
Resist the temptation to build a model before you have a rule. A rule is explainable in a renewal meeting; a model is not, and the point of this read is to make a CSM act.
- Score the pairs, not the counts. Repeat theme and competitor mention: high. Escalation language and absent champion: high. Any single pattern: watch.
- Attach a confidence. How many independent records support the score? Three tickets and a call from two different people is strong. One ticket from one person is weak. Say the number.
- Price it. Multiply nothing. Just show the ARR and the renewal date beside the score. “$214K renewing in 60 days, high, four sources” is a sentence a leader acts on.
- Set a floor for action. Below the floor the read is advisory. Above it, someone owns a play this week.
Acting before the renewal call
A score without a play is a dashboard. Three plays cover most cases, and each has an owner and a deadline the moment the score crosses the floor.
- Product play. The repeat theme is a product gap: bring the quotes to product, get a date or an honest no, and tell the account which it is.
- Executive play. The champion is gone or the sponsor is silent: an executive sponsor call this week with the seat maths and the roadmap.
- Competitive play. A competitor is named: a battlecard for that competitor’s move, and a call that addresses it directly rather than waiting for the account to raise it.
Guardrail: anything touching pricing or an account inside its renewal window is a human decision, whatever the score says.
Measuring the programme
- Lead time. Days between the first high score and the renewal date. It should grow as the read matures.
- Time to owned action. Days from a signal landing to a play with an owner. This is the number to cut. In HyperOrbit’s worked example it falls from 8 days by hand to 36 hours with the agents; see the methodology page for what that measures.
- Plays run versus scores raised. If scores are raised and plays are not run, the floor is wrong or the owners are not real.
- Saved ARR, honestly. Count renewals that were high-risk and renewed, and say plainly that you cannot prove the counterfactual.
Where agents fit
Everything above can be done by hand for 30 accounts. It cannot be done by hand for 300, weekly, and that is the case for an agent.
In HyperOrbit, Chorus reads every ticket and call as it lands and clusters them into themes and sub-themes with the quotes and the ARR behind each. Recon logs competitor moves and mentions on the same timeline, so a spike in comparisons is tied to the move that caused it. Atlas, in beta, fuses both with usage and CRM into one health score per account with the factors explained, watches every renewal window, and drafts the play with an owner and a confidence number, at the autonomy you set. The method is the same. The reading is done for you.