How many times an account churned within a month of an observation date.
Recall (catch rate): of accounts about to churn, how many we alert in time.
Precision: when we alert, how often a churn actually happens within the month.
Lift: how much better than chance (base rate 1.9%).
Threshold vs metrics (drag to explore)
Alert threshold: (default is the F1-optimal point, dashed).
precisionrecallF1 default
Top churn drivers
Meetings add-on abandoned (no scheduling in 90 days)
Sales pipeline slowing: fewer new leads, calls, and wins
Silent disengagement despite long tenure
Rising meeting cancellations and lead unsubscribes
Search and configuration activity dropping off
Expansionintent model
Expansion account-months to detect
820
How many times an account upgraded within a month of an observation date.
Recall (catch rate): of accounts about to upgrade, how many we alert in time.
Precision: when we alert, how often an upgrade actually happens within the month.
Lift: how much better than chance (base rate 2.3%).
Threshold vs metrics (drag to explore)
Alert threshold: (default is the F1-optimal point, dashed).
precisionrecallF1 default
Top expansion drivers
User seats exhausted (zero headroom)
Surging won-deal volume and revenue
Growing saved-search and search usage
Scaling lead generation and outreach
Accelerating CRM contact management
Method notes. Every account is scored at its monthly observation date on fresh features; the label is whether the event actually occurs within one month. Accounts younger than two months (tenure ≤ 63d) are excluded from training and evaluation. Precision is alert-centric, recall event-centric, and lift is precision over the base rate. Drag the slider to see the tradeoff; the default is the F1-optimal threshold. Counts are account-months (an account paired with an observation date), not distinct accounts: the same account can churn, return and churn again, or upgrade more than once, so it can appear several times.