Catch churn anomalies 3–4 weeks earlier.
Forecast learns the normal churn pattern and alerts customer success whenever actual churn deviates more than 1.5 standard deviations from baseline. The team intervenes with at-risk accounts three to four weeks earlier than before.
Run Forecast- 1
Describe the task
Forecast can learn what normal churn looks like for your product, week by week and cohort by cohort, and then watch for the weeks that don’t fit. Instead of a monthly retrospective that tells you churn spiked three weeks ago, the anomaly node fires the moment actual churn drifts past 1.5 standard deviations from its learned baseline, so customer success intervenes while the accounts are still savable.
The tighter you scope the baseline, the fewer false alarms. Tell Forecast which series to watch, how sensitive the threshold should be, and who gets the alert, and it holds that behaviour every night.
Watch our churn and MRR series in weekly-churn-mrr.csv for anomalies. - Learn the normal churn baseline per segment from the trailing 12 months - Fire an alert whenever actual churn exceeds 1.5σ above baseline for a segment - Include the segment, the deviation size, and the accounts driving it - Suppress alerts on known seasonal dips so we don’t page on noise Route each alert to the owning CSM. End the run with a summary: segments watched, alerts fired, and total MRR sitting behind the flagged accounts.
- 2
Give Forecast context
In a Forecast workspace, pipe your weekly churn and MRR export into a source node and wire it into an anomaly-detection node. Forecast fits a per-segment baseline first, so a high-churn SMB tier and a low-churn enterprise tier are each judged against their own normal instead of a single blended average.
Give the baseline enough history to be stable, ideally a full year, so genuine seasonality (year-end non-renewals, summer slowdowns) is learned rather than mistaken for an anomaly.
Required contextA weekly churn and MRR series with at minimum week, segment, churned accounts, and MRR lostA segment or cohort key, so the baseline is learned per tier rather than blendedOptional contextAn account-level detail feed, so each alert can name the specific accounts driving the spikeA Slack or email connector, so alerts land in the CSM channel the moment they fireA list of known seasonal windows, so expected dips don’t trigger a false alarmweekly-churn-mrr.csvsegment-map.csvSlack (optional) - 3
What Forecast creates
Forecast fits a baseline per segment, then evaluates each new week against it. You get a running anomaly log, an alert whenever a segment crosses the threshold, and a summary of how much recurring revenue sits behind the accounts that triggered it.
From Forecast: Watched 6 segments in weekly-churn-mrr.csv against a 1.5σ deviation threshold.
1.5σAlert threshold3Segments over baseline£214KMRR behind flagged accountsAlerts fired this week3 segmentsSegment Deviation Accounts driving it Owner SMB, monthly +2.3σ Meridian Labs, Coilworks, Tapstack Priya Shah Mid-market +1.7σ Delcourt Group, Vane & Fisher Marcus Lin Enterprise, EU +1.6σ Ostend Freight Dana Ortiz Within baseline3 segmentsSegment Deviation Note SMB, annual +0.4σ Normal, no action Mid-market, EU −0.2σ Slightly below baseline this week "The SMB monthly spike is 2.3σ over baseline and three accounts explain most of it. Want me to draft save-play talking points per account, or hold the alert until it persists a second week?"
- 4
Follow-up prompts
Draft save plays for the flagged accounts
Forecast can pull each flagged account’s recent usage trend and pair it with a suggested intervention, so the CSM opens the alert with a next step, not just a red flag.
For the three accounts in the SMB monthly alert, pull their last 8 weeks of usage and draft a one-line save play for each, referencing what actually dropped.
Tune the sensitivity
If the team feels paged too often, raise the threshold; if spikes are slipping through, lower it. Forecast re-evaluates the same history so you can see how many alerts each setting would have produced.
Re-run the last quarter at a 2.0σ threshold instead of 1.5σ and show me how many alerts I would have gotten and which real spikes I would have missed.
Route alerts into Slack automatically
With the Slack connector, Forecast can post each alert to the CSM channel the night it fires, tagged to the owning rep, so nobody waits for the Monday review.
Post every fired alert to #cs-early-warning with the segment, deviation size, driving accounts, and owner tagged. Send a "no anomalies" note on clean nights so the channel stays trusted.
- 5
Tips and troubleshooting
Segment before you threshold
A blended baseline hides tier-specific spikes, a bad enterprise week gets swallowed by a big SMB denominator. Give Forecast a segment key so each tier is judged against its own normal.
Suppress the seasonality you already know
Year-end non-renewals will trip a naive detector every December. Feed Forecast your known seasonal windows so it treats those dips as expected rather than paging the whole team.
Confirm-then-alert avoids whiplash
A single noisy week can cross 1.5σ and correct on its own. If your team prefers fewer, higher-confidence pages, ask Forecast to require the deviation to persist two weeks before alerting.
Ready to try it yourself?
Pipe your weekly churn into Forecast, let it learn the baseline, and get an early-warning alert weeks before a quarterly review would have caught the drift.
Run Forecast