Artificial Intelligence and Automated Decision-Making

Automated calls and messages identify themselves

Where an automated agent places a call or sends a message, it identifies itself as artificial intelligence. It does not present itself as a human being, and it will tell you it is not one if you ask.

Automated outreach contacts a number only where the operator holds a consent record for it — see Who is sending, and who obtained consent in the Text Messaging Terms, which describes the same record for both channels. It never cold calls. Calls are placed only between eight in the morning and nine at night in the recipient’s local time; outside that window they are queued rather than dropped. A request to stop is honored immediately across every channel at once, not only the channel it arrived on.

See also the Text Messaging Terms and the Call Recording Notice.

What is transmitted to model providers

We use third-party artificial-intelligence models to make specific inferences about records in an operator’s account.

Some personal information is transmitted to those providers. In particular, an enrichment run transmits party names and email addresses to the configured model provider. This happens even when a run is performed in dry-run mode: dry run withholds the write to our database, not the network call to the provider.

We do not use operator or consumer data to train artificial-intelligence models.

The categories of provider in use are listed on the list of who processes data for us.

Model output never becomes the record

When a model forms a judgment, it is stored beside the verified value rather than on top of it, together with the provider, the model, a confidence score, and the reasoning given. It does not overwrite a system-of-record value. This means a model’s inference can always be told apart from a fact, both by us and by anyone reviewing an account later.

Automated decision-making

Where you have a right to object to automated processing or profiling, or to ask that a decision involving it be reviewed by a person, you can exercise it through the privacy rights request form. Tell us which account or property the decision concerned, so we can find the record it was made against.

Because a model’s judgment is stored beside the verified value rather than on top of it, a decision can be traced back afterwards to what the model suggested and what the record actually said.

Questions

Questions about this disclosure can be raised through the privacy rights request form.