How Orme OS drafts outreach a buyer can check
AI email that gets details about the buyer wrong costs replies and hurts the sending domain. How Orme OS builds a first email only from claims it can cite, refuses when the evidence is thin, and holds every first touch for a named person's approval.
On this page7
Short answer
Orme OS drafts a first email only from claims it can cite. The model proposes an opening, a set of claims each tied to a piece of evidence, and an ask. The software keeps only claims whose names and numbers appear in that evidence, refuses to draft when there is too little, and holds the message until a named person approves it.
What goes wrong with AI-written outreach?
#It says things that are almost true. Asked to personalise an email, a model reaches for a plausible detail: a funding round from the year before, a launch that belongs to a competitor, a title the person no longer holds. The buyer notices and does not reply.
The cost does not stop at one email. Gmail requires senders to keep the spam rate reported in Google Postmaster Tools below 0.3%, and every recipient who marks a careless message as spam pushes the whole domain toward that line.
How does Orme OS keep a draft grounded in evidence?
#The model does not write the email freely. It returns three parts: an opening, a list of claims, each carrying the id of the evidence it rests on, and an ask. Code then assembles the email from the claims that pass. Every number and every proper noun in a claim has to appear in the evidence it cites, or the claim is dropped.
The source material is also scanned for instructions hidden inside it, and no part of the model's output can name a recipient, call a tool or grant an approval. Those decisions stay outside the model.
When does Orme refuse to draft an email?
#When the evidence is too thin to say anything true. The drafting engine refuses when fewer than two signals have fired on an account, and when the contact belongs to a persona you have excluded. A second check then lists any claim the evidence does not support. The draft gets one retry, and if it still fails, it is shown as rejected with the failing claims, so the seller can see exactly why.
Who approves the email, and what do they see?
#The seller who owns the account, one message at a time. The approver sees the email exactly as it will be sent, with the full envelope. If the message or the envelope changes after approval, the approval no longer holds. Approving the first touch approves the follow-up steps in that sequence.
What happens after the email is sent?
#Replies are classified as interested, objection, referral, not now or unsubscribe. The inbox agent drafts a response for approval and proposes meeting times from the seller's calendar, and an unsubscribe suppresses the contact automatically. When a meeting is booked, it is recorded against the play that produced it.
How does Orme OS protect the sending domain?
#With limits that apply before anything goes out. By default, a mailbox sends no more than 30 messages a day, no person receives more than 2 messages in a week, and sending pauses on its own if bounces pass 2% or spam complaints pass 0.1%. The domain's own authentication still has to be in order, which is what the free deliverability check is for.