AI processing notice
Where Watchly uses AI, and what it is and is not allowed to see.
This notice covers the AI features Watchly has today. It will be extended, not replaced, as more are added. Accepting it at signup records that you agreed to this document at the version shown on that form, which we keep on file.
What is true today
Watchly has one AI-assisted feature: an owner-triggered second opinion on a payroll run, before it is approved. It does nothing on its own — a button press starts it, and it never edits a figure or approves a run. It is off for the whole product until Watchly sets a single platform credential that switches it on — this is not a setting an individual firm turns on for itself. With no credential set, pressing the button does nothing and no figure leaves Watchly, for any firm.
What is sent, and what never is
Before anything is sent, every officer on the run is replaced with an anonymous label (A, B, C…). Only the label, and figures already computed by Watchly — hours, gross pay, overtime, statutory totals — are sent. A name, PSIRA number, ID number or contact detail is never sent: Watchly maps the label back to the real officer locally, after the check returns, to show you the result. The provider is Anthropic (the Claude API); see our privacy notice for how we handle personal information generally. What Anthropic itself does with a request once we send it is governed by our arrangement with them, which is being finalised with legal counsel.