Security

Built so one customer never sees another’s data.

This page describes what is actually implemented in Churnity today, not a policy aspiration. It is an engineering summary, not a substitute for your own security review; email hello@churnity.com for more detail.

Last updated 17 November 2026

Hosting and data residency

Churnity runs on Microsoft Azure, Australia East region: the database, application, background jobs, logs, key vault and storage all live there. Backups are kept in the same region and are not geo-replicated.

We are evaluating an additional US hosting region ahead of a wider US launch. Until that lands, data for every customer, regardless of location, is processed and stored in Australia.

Tenant isolation

Every organization’s data is separated by PostgreSQL row-level security, enforced on every tenant table and forced even for the table owner, so isolation does not depend on application code remembering to filter by customer. The application itself connects with a restricted database role that cannot bypass this; the few operations that need broader access run under a separate, narrowly scoped role.

Within each request, the signed-in person’s organization and permissions are set once, inside a single database transaction, and checked by the database itself on every row. A reused connection carries nothing over between requests.

What we read from your systems, and how

Connectors (Salesforce, HubSpot, Microsoft Dynamics 365, Freshsales, Jira Service Management, Zendesk, Freshdesk, Snowflake, Mixpanel and Stripe today) are read-only by construction: every request goes through a single guarded transport that refuses write operations before they reach the network.

We read only the fields an accepted measure or an approved mapping actually uses. A triage step removes personal fields (names, emails, phone numbers, addresses) and free text before anything is classified or analyzed. Where a person still needs to be matched across your systems, we store a keyed cryptographic hash of their identifier, unique to your organization, rather than the identifier itself.

Encryption

Connector credentials and AI provider keys are encrypted with AES-256-GCM before they are stored. Logs carry counts and identifiers, never secrets, tokens or the contents of your records.

Your control over what reaches an AI model

AI features are optional and governed by settings your organization’s owners and admins control, not a blanket default:

  • A region switch: by default, Churnity will not send your data to an AI model that processes outside Australia.
  • An evidence ceiling: how much an AI assistant can see when it helps map or classify your data, from field metadata up to real sample values, set by your organization.
  • Sending a small number of real record values to an AI, to help resolve a setup decision Churnity cannot settle from metadata alone, always requires a person to explicitly authorize that specific send; nothing is sent by default and nothing is remembered for next time.
  • The AI analyst, which reads account-level results to help explain and prioritize them, never receives people: contact references are replaced with a generic term and email addresses are masked before anything is sent. It can be switched off entirely by an owner or admin.

Deletion and erasure

A specific person can be erased from an organization’s data on request: their contact record, records that reference them, and cached AI text. An organization owner can also delete every record Churnity has extracted for that organization.

Compliance and subprocessors

Churnity is pre-launch and has not yet started a SOC 2 process; we plan to begin one as we take on larger customers. A formal data processing agreement and subprocessor list are in progress. Today’s subprocessors are Microsoft Azure (hosting), Vercel (this marketing site and its cookieless analytics), and, only if your organization enables AI features, the AI provider your organization’s settings allow. Ask hello@churnity.com for current detail for your own review.