Numbers we will not quote
We do not publish a return on investment, revenue recovered, hours saved, or a conversion uplift. Not because they would be unflattering — because RAJEH does not yet measure them to a standard that would let a customer check them. When it does, the figures will arrive with the method attached.
Autonomy, stated exactly
RAJEH does not run a company by itself, and we do not sell it as though it does. Approval is the default: AI employees prepare work and a person decides. Where an owner has explicitly authorised an employee to act on its own, that authority is narrow, capped, recorded, and revocable by them at any time.
An AI employee never gains authority because time passed. It gains it because an owner granted it.
One line inside that authority is worth stating on its own: RAJEH answers, it does not initiate. Even where an employee may reply without a person pressing send, it may only reply to a conversation the customer opened. Nobody has authorised it to start one, and starting one is a different question with a different answer about consent.
Excluded from this version
- Advertising source attribution — the identifier it would need cannot be read from a connected workspace.
- Marketing campaign execution, course content and lifecycle messaging.
- Importing a product catalogue from the connected workspace.
- Writing conclusions into custom contact fields — tags are used instead, and they are read reliably.
Known limits we would rather say out loud
Lead rotation is proven at every step we control, but has not yet been exercised against a live staff account in a customer workspace. Alerting covers scheduled work and provider limits, not every failure path. We would rather a customer read that here than discover it later.