When a customer churns, the retrospective almost always concludes the same way: the signs were there. They usually were. The problem is rarely that the signals did not exist. The problem is that they were scattered across a support tool, a product analytics dashboard, an inbox, and a CSM's memory, and nobody connected them until the renewal was already lost.
Churn in B2B SaaS is a lagging event with a long list of leading indicators. If you know where to look, you can usually see it coming a full quarter out. Here is the paper trail, grouped by where it lives.
Signals in your support queue
Support is the most honest channel you have, because customers are talking to you when something is already wrong. The trick is reading patterns, not individual tickets.
- A shift in tone, not volume. A customer who used to open tickets with "quick question" and now opens them with "this is still broken" has changed how they feel, regardless of how many tickets they file. Sentiment trend beats ticket count.
- Repeated escalations on the same issue. One escalation is a bad day. Three escalations on the same root cause is a customer concluding the product cannot do the job they bought it for.
- Questions that sound like exit planning. "How do I export all my data?" and "What happens to our integrations if we leave?" are rarely idle curiosity. They are due diligence on a decision that has, internally, already been made.
- Silence after a bad experience. A customer who escalated hard and then went quiet has often stopped fighting because they have stopped caring. Disengagement after conflict is more dangerous than the conflict itself.
Signals in your product telemetry
Usage data is where most teams look first, and it is genuinely useful, as long as you watch the right shape of decline rather than the raw number.
- Breadth contraction. Total usage can hold steady while the number of active users quietly drops from twelve to three. That is a champion-dependency risk forming in plain sight, and it is invisible if you only track aggregate activity.
- Abandonment of the north-star workflow. Every product has the core action that means a customer is getting value. When an account stops doing that specific thing, even while logging in for peripheral features, value realization is collapsing.
- Onboarding that stalled and never recovered. Accounts that never reached first value during onboarding rarely renew. A flat adoption curve in the first 60 days is one of the most reliable churn predictors there is.
- Declining stickiness. A dropping ratio of daily-to-monthly active usage means the product is moving from habit to occasional tool. Habits renew. Occasional tools get cut in the next budget review.
Signals in your inbox and on your calls
This is the richest signal source and the one no dashboard captures, because it lives in language. It is also where intent shows up earliest.
- Your champion stops replying. Email response latency is an underrated metric. When the person who used to answer in an hour now takes a week, the relationship has cooled even if nothing has been said.
- The economic buyer disappears from the thread. When the budget owner stops attending QBRs and delegates to someone junior, the account has been deprioritized internally. That decision precedes the cancellation by months.
- An org change at the customer. A reorg, a layoff, or your champion taking a new job is a structural risk that can erase a healthy account overnight. These show up in email signatures and LinkedIn long before they show up in usage.
The earliest churn signal is almost never a number. It is a sentence, said on a call or typed in a thread, that a dashboard will never see.
Why these signals usually get missed
Any one of these is catchable by an attentive CSM. The failure is structural: no single person sees all of them for all of their accounts. The support trend lives in Zendesk, the usage contraction lives in PostHog, the cooling champion lives in Gmail, and the reorg lives on LinkedIn. Correlating them by hand across a book of eighty accounts is not a discipline problem. It is a volume problem that humans cannot solve at scale.
This is exactly the gap Merrily closes. It reads every one of these channels continuously, correlates the signals per account, and surfaces the few accounts where multiple indicators are firing at once, with the underlying evidence attached. Instead of a CSM hoping to notice a cooling email thread, the risk is on the screen the day the pattern forms. For the deeper case on why automated reading beats periodic surveys, see signals over surveys.
The accounts you lose are rarely the ones that blindside you. They are the ones whose signals you had, but never assembled in time.