AI transparency
When you are talking to a machine, we say so.
Article 50 of the EU AI Act requires that a person be told when they are interacting with an AI system. We would do it anyway, because a customer who discovers it later is a customer we have lost, but it is also an obligation and we take it as one.
Where AI is used
- Answering the phone and messages. An AI agent may answer a call, a web chat or a WhatsApp message on behalf of a client. It identifies itself as an AI assistant at the start of the conversation, in the language of the conversation.
- Reading documents. Invoices, statements, leases and contracts are read by AI to extract terms and figures. Every extracted value carries the page it came from, and a value the system cannot source is left blank rather than guessed.
- Drafting. Emails, replies, filings and reports are drafted by AI.
- Bookkeeping and reconciliation. Transactions are categorised and matched by AI against a client’s own history.
What a human decides
Everything that leaves the building. Our services are draft-first: the AI proposes and a named person approves. Nothing is sent, filed, posted to a ledger or paid on the strength of an AI decision alone. Where a client asks us to automate a step end to end, that is a written decision by that client, recorded, and reversible. In the client platform, only the business owner can let a message type go out on its own, only review requests and visit reminders in our standard wording are eligible, and the owner can switch it back at any time.
The AI does not decide credit, does not opine on whether a clause is enforceable, does not certify anything to a third party, and does not give legal, accounting or investment advice. Those are decisions for a qualified person and we say so on the pages that sell the relevant service.
What we do when the AI does not know
It stops and says so. A system that produces a confident answer it cannot support is worse than one that produces nothing, because the first is expensive to discover and the second is obvious immediately. Our document work is measured on exactly this: how often it returns a cited value, how often it refuses and says why, and how often it goes quiet without explanation. Those counts are published.
Recording, and your data
The phone line we set up for a client's business does not make audio recordings of calls. Calls to Feerasta's own number are answered by an AI receptionist and are recorded and transcribed by our voice provider, ElevenLabs. Feerasta’s own line is different: Rachel introduces herself as the front desk agent and says plainly that she is an AI whenever anyone asks. The caller is told at the start of the call. See the privacy policy.
What we do with data, where it is stored, and who else touches it is set out in the privacy policy and the list of sub-processors. The short version: your data is not used to train anybody’s model, ours included.
Telling us it got something wrong
Write to hello@feerasta.ai with what the system said and what it should have said. If an automation produced a wrong output we will tell you what caused it and what we changed, in writing. We keep those in the test suite so the same mistake cannot return.
Last updated 19 August 2026. Questions about anything on this page: hello@feerasta.ai.