Care
Two pillars
Estimates, not certainties
The model computes a probability that an account churns soon, based on its behavior in the data you imported. That probability informs where your attention goes first. It does not replace your judgment. A quiet account might be on vacation. A high-scorer might have a conversation in progress that changes everything. Churnora ranks attention. You decide what it deserves.
Your customers' data
The data you import is yours. It is encrypted in transit and at rest, scoped to your workspace in the database, and never used to train a shared model. When a language provider writes an explanation for one account, it receives only that account's behavioral fields, under an agreement that prohibits training. You can export everything at any time, and delete means delete: a delete in Settings removes data from the database immediately, and backups roll off within 30 days.
On the model
The churn score comes from a behavioral model running in Churnora's own code: a calibrated logistic regression over eight engineered features. It only reads what you imported. It does not guess. It does not have a sense of the market or your company's trajectory.
A language model writes the plain-language explanation and drafts the save play. It does not predict. That means the words on the factors panel are generated text, and they describe a number that already exists in that account's row. The score never depended on the language model. When no language model is configured, explanations fall back to code-written phrasing and the interface says so.
What the model sees
Sign-in recency, sign-in volume, usage trend, feature breadth, seat utilization, support ticket count, payment failures, and tenure. Eight features, engineered from the usage data you import, each one traceable to a number in that account's row.
What the model does not see
Conversations, account notes, sales context, market conditions, or anything outside the behavioral data you upload. If you know something the data does not show, your judgment takes priority.
How to disagree with a score
Open the account, read the factors, and decide. If the top factor is last sign-in 31 days ago and you know the account's champion just went on leave, that context belongs with you. Trust yourself.
Where to go next
The security page describes the technical controls: encryption, session handling, how payment data is kept separate, and the honest early-stage note about SOC 2. The your-data page lists exactly what Churnora stores and how to export or delete any part of it. The how the model works page is generated from the code and shows the features, the coefficients, and how each factor is calculated.
If something here is not clear, or if you have a question that is not answered by any of those pages, the contact page goes to a person.