Revenue

Sequences

Multi-step outbound where every touch is written, not interpolated. A sequence names an AI employee and a brief; the employee drafts each step for each contact from that person's real context — the threads on their timeline, the open deal, the signal that enrolled them. Find it under Revenue → Sequences.

Why it is not a mail merge

A template engine can put a first name in a subject line. It cannot write step 3 as "reference whatever they said in reply to step 2". That is the difference here: a Sequence stores instructions, never message bodies. At send time the named employee reads the contact's timeline and writes the actual email. The cost of that is real — every touch is a model call, and a bad brief produces bad mail at volume — so the default is that a human still presses Send.

Building one

  1. Revenue → Sequences → New sequence. Give it a name and a description.
  2. Pick the mailbox every touch sends from — one of the accounts connected under Email — and the AI employee that writes them.
  3. Write the Brief. Who this is for, what you sell, what a good reply looks like, and what never to say. This is the single highest-leverage field in the whole feature; it is markdown, and it is handed to the employee on every step.
  4. Add steps. Each one has a name, a delay, and an instruction: what this specific touch should accomplish.
  5. Set the guardrails — send window, daily cap, stop-on-reply — then move the status from Draft to Active.

A step's delay is measured from the previous touch's send, not from enrolment, so pausing a sequence for a week does not bunch every pending touch into the moment you resume it. Reply in the same thread is on by default: a follow-up with no quoted history reads as spam. Turn it off for a step that is a genuinely new angle rather than a bump.

Enrolling contacts

Select contacts in Revenue → Contacts and press Enrol, pick the sequence from a deal, or let a Signal enrol somebody the moment they hit a condition in your product.

Bulk enrolment refuses per contact rather than failing the batch — one blocked person in eighty must not lose the other seventy-nine — and reports exactly who was skipped and why:

  • No email — you only hold a phone number for them.
  • Do not contact — the person is flagged as a hard opt-out.
  • Suppressed — the address is on the do-not-email list. See Deliverability.
  • Already enrolled — somebody is in a sequence exactly once. Re-enrolling a person who finished resets their existing row rather than adding a second, because receiving the same opening line twice is the most visible way an outbound tool embarrasses its owner.
  • Archived — the contact or the sequence is archived.

The review queue

With autoSend off — the default — a drafted touch lands in Email → Drafts, the same review queue built for exactly this job. Every row shows the AI employee and the routine or sequence that produced it next to the recipient, subject and preview, so a night's drafting is one scannable list. Filter or group by employee, open a draft to see the message, its attachments and the thread it replies to, then send the ones you like — individually, by ticking a selection, or with Send all.

Because a send cannot be undone, a batch stops at a confirmation showing who is about to receive mail — real addresses, a breakdown per source, and the mailbox it goes out from. Past 25 drafts the confirm button stays disabled until you acknowledge the size explicitly.

autoSend, and its two grants

autoSend is the one switch in Genosyn that spends your sending reputation with no human in the loop. It is off by default, and turning it on requires two independent grants on the same employee:

Revenue: send
Granted from Revenue → AI access. The employee is trusted to run outbound unattended.
Mailbox: send
Granted from Email → Settings → AI access for that specific mailbox. The employee is trusted to speak from that address.

Both are re-checked at send time, not just when you saved the sequence. Downgrade either grant and the next touch drafts instead of sending — nothing goes out on the strength of permission somebody has since revoked. Toggling autoSend is written to the audit log with the employee it applies to, because it is the field worth reading back a year later.

Send windows

A sequence sends only inside its send window: a set of weekdays, a start hour, an end hour, and a timezone. The default is conservative — weekdays, 08:00 to 17:00, UTC — because a sequence that mails at 3am reads as automated however good the copy is. Set the timezone to the one your buyers work in.

  • A window whose start and end are the same hour means never, not all day. "09:00 to 09:00" is far more likely to be a mistake than a request to mail around the clock.
  • An empty day list also means never — the supported way to freeze a sequence without pausing it.
  • A window that wraps past midnight is fine: set an end hour lower than the start hour.
  • Daylight saving is handled by the platform's timezone database, so a window pinned to America/New_York stays correct across the change.

A touch that comes due outside the window is not dropped. It waits for the next opening.

Daily caps

Daily cap is the maximum number of touches the whole sequence may produce in a day, defaulting to 50. Both sent and drafted touches count against it — the drafting is what costs model time and what you will have to review — and 0 removes the sequence-level cap entirely, which you should reserve for a sequence you have already watched run.

Underneath, the scheduler dispatches a bounded number of touches per sweep across the whole installation, and it makes the cheap refusals first: enrolments belonging to a paused sequence or one that has burned its allowance are filtered out before the budget is spent, so a large blocked backlog can never starve a live campaign.

Stop on reply, and the other ways an enrolment ends

Stop on reply is on by default, and turning it off is almost always wrong. Because mail sync already ingests inbound mail, a reply on the thread the sequence started stops the enrolment within a heartbeat — the prospect answers and the machine gets out of the way. An enrolment ends in exactly one of these states, and the distinction is operational:

completed
Every step was delivered. Nothing went wrong.
stopped_replied
They answered. The success case — hand it to a human, or to a deal owner.
stopped_unsubscribed
They opted out. A compliance record; the address is suppressed too.
stopped_bounced
The address is dead. Fix your data before mailing that domain again.
stopped_manual
Somebody pressed Stop.
failed
Something broke. Deliberately not retried — an employee producing a malformed draft would otherwise mail the same prospect every hour. Fix it, then resume the enrolment.

What was actually sent

Every attempt at every step is recorded on the enrolment: sent, drafted, skipped with the reason, or failed with the error, plus the subject line and a link to the message. A skip is not a failure — the most common one is a suppressed address, which is the system working correctly — but a silent non-send would be indistinguishable from a bug, so it is always written down. This is the page to open the first time somebody claims they were mailed after unsubscribing.

  • Deliverability — the suppression list, unsubscribe links, bounces, and how not to burn your domain. Read it before you turn autoSend on.
  • Signals — enrol people automatically when your product says the moment is right.
  • Revenue — contacts, deals, and the timeline the drafter reads from.