An agent that runs the whole outbound motion
Not a sequencer with an AI button on it. Autopilot does the research, picks the targets, writes each email individually, sends from infrastructure it warms itself, answers the replies and books the meeting — then tells you what it did.
Lead research
Maren Ashby
Owner
Quillmoor Events
ICP fit score
Scored against the selected ICP filters
Hooks found
- Quillmoor Events runs 18 events a season
- Hiring for a lead designer
Fernhallow x Quillmoor Events
Hi Maren,
Saw Quillmoor Events runs 18 events a season. At that pace fresh stems get expensive and wilt under venue lights — our boxed silk stock is bought once and redeployed all season.
Want a sample box to compare against your last fresh order?
— Nell, Fernhallow Silk Co.
This is timely — we are re-stocking for the spring season. How do the arrangements hold up under warm venue lighting?
They hold colour under hot venue lighting — that is the main reason studios switch. Would Tuesday at 14:00 work for a short call?
Meeting booked · Tuesday 14:00 · 30 min
No human touched this threadSample run — a fictional wholesaler, fictional buyers and example domains, drawn on the real product schema.
How it works
Six steps, none of them yours
Each step writes real records you can open, audit and override. Nothing happens in a black box you cannot inspect afterwards.
- 1
Read your business
Project → MarketResearchYou give it a domain. It crawls your site the way a new rep would read it, then goes wider: who else sells this, how they position, what buyers push back on.
- What you sell, in the words your own site uses
- A competitor set with a note on each one
- Positioning, differentiators and proof points
- The objections that will come back, before they come back
- 2
Propose ICPs — you pick
Icp[] (suggested → selected)It drafts several ideal-customer profiles, each one a real filter set over the graph with a fit score, not a persona document.
- Every ICP carries a rationale you can argue with
- Reject one and it stops being targeted — permanently
- Selected ICPs become the campaign audience
- 3
Build the audience
Lead rowsFilters run against the company and people graph, then the list is cleaned before a single email exists.
- Deduplicated against your existing leads and customers
- Checked against your suppression list and every global unsubscribe
- Role, seniority and department matched — not just a job-title string
- Each lead scored for fit against the selected ICP
- 4
Research the person, then write
Lead.research → MessageEvery lead gets its own research pass — hooks, signals, a summary — and every email is generated from that. Sequence steps are prompts, not mail-merge templates, so no two emails come out the same.
- Hooks drawn from the company profile, not a spun first line
- Your value props matched to that buyer, not pasted in whole
- Length, tone and call to action follow the brief you wrote
- Preview any lead before the campaign starts
- 5
Send from warmed infrastructure
Message.status → sentSending is rotated across a pool of pre-warmed domains and mailboxes, inside your sending window, under caps that exist to protect reputation rather than to hit a number.
- Per-mailbox daily limits with a natural ramp
- Sending window and weekdays in the recipient-appropriate timezone
- SPF, DKIM, DMARC, MX and tracking CNAME verified before the first send
- Bounces and complaints feed straight back into suppression
- 6
Classify, answer, book
Thread.intent → MeetingEvery reply is classified and acted on. Straightforward answers go out immediately; anything that needs a human is flagged rather than guessed at.
- Nine reply intents, each with a defined action
- Questions answered from the research and your proof points
- Booking link offered and the meeting written to your calendar
- Anything ambiguous handed to you with the thread summarised
Your controls
Autonomous is not unsupervised
Five things stay yours. Everything else is the agent's problem.
The ICP
Accept, edit or reject every proposed profile. Filters are the same ones the search product exposes — countries, segments, headcount bands, technologies, seniority, department.
segmentscountriesemployeeMin/MaxsenioritiestechStackVoice and tone
Each sequence step is a written brief the agent follows: what to open on, what to prove, what to ask, how long. Change the brief and the next batch changes with it.
step promptsubject hintstop on replydelay daysVolume caps
A campaign cap, a per-mailbox cap and a sending window. The agent will scale inside those numbers and will never quietly exceed them to hit a target.
daily_capmailbox limitsending windowweekdaysExclusions
Upload domains and addresses you never want touched — customers, open pipeline, partners, competitors. Unsubscribes and hard bounces are added automatically and are permanent.
domainemailunsubscribebouncecomplaintApproval mode
Autopilot lets the agent start, pause and scale campaigns on its own. Turn it off and every batch and every reply waits in a review queue for you first.
autopilot onreview queueneeds_humanpause on bounceSending infrastructure
The part everyone underestimates
Deliverability is not a setting. We run the pool, warm the domains, hold the caps and keep the DNS correct — because a burned domain costs more than a slow month.
Pre-warmed pool
SampleStart sending the day you sign up on the shared pool, then move onto dedicated domains as they finish warming. You never wait six weeks with nothing going out.
Warm-up ramp
Volume per mailbox climbs on a schedule, not on demand. If engagement dips, the ramp pauses itself before the reputation does.
Authentication, handled
SPF, DKIM, DMARC, MX and the tracking CNAME are provisioned and verified for you. Nothing sends from a domain that is not fully aligned.
- SPFverified
- DKIMverified
- DMARCverified
- MXverified
- Tracking CNAMEverified
Per-mailbox caps, enforced
SampleHealth 98/100 · active
Health 95/100 · active
Health 92/100 · active
Caps are per mailbox and per domain, and the campaign cap sits on top of both. When a mailbox reaches its limit the send moves to the next one in the rotation instead of pushing through.
Reply handling
Nine intents, nine defined actions
A reply is classified before anything is written back. The action for each intent is fixed and visible — including the two that stop and ask you.
| Intent | What the agent does | Hand-off |
|---|---|---|
| interested | Answers the actual question, offers the booking link, writes the meeting to your calendar when they pick a slot. | handled |
| question | Answers from the market research and your proof points. Anything that needs a commitment we cannot make is flagged instead of invented. | if unsure |
| referral | Thanks them, then re-targets the named colleague as a new lead on the same account with the context carried over. | handled |
| not now | Confirms, stops the sequence and re-queues the lead for the date they named — not a generic 90 days. | handled |
| out of office | Reads the return date, pauses the sequence until then and consumes no step. | handled |
| auto reply | Recognised as machine-generated. Nothing is sent, nothing is counted, the sequence continues. | handled |
| not interested | Stops immediately, closes the thread and marks the lead so no future campaign picks them up. | handled |
| unsubscribe | Suppressed on the spot at address and domain level, across every campaign in the workspace, permanently. | handled |
| other | Handed to you with the thread summarised and the suggested reply drafted but unsent. | if unsure |
Every outbound reply the agent sends is stored on the thread, so you can read exactly what was said on your behalf. Turn autopilot off and the same drafts wait for your approval instead.
What it costs
You pay for emails that actually left
No seats, no platform fee, no minimum contract. The wallet is drawn down per send; research and enrichment run on credits.
$0.03
Per email sent
Charged on send, not on queue. Bounces detected before send cost nothing.
$30
Free to start
About 1,000 emails, granted once per workspace with no card.
10 cr
Deep research per lead
100 credits = $1. Person rows cost 2, email reveals 3.
$0
Replies and meetings
Classification, answers, booking and hand-off are part of the product, not a line item.
Sending infrastructure — domains, mailboxes, warm-up and DNS — is included in the per-email price.
Honestly
When this is not for you
Four situations where we would tell you not to buy. We would rather say it here than three weeks into an onboarding.
Markets where cold outreach needs prior consent
Some jurisdictions and some regulated sectors require opt-in before any commercial approach, and no amount of personalisation changes that. We will not build you a workaround, and we would rather lose the deal than help you break the rule.
Consumer audiences
The graph is business entities and business roles. There is no consumer data in it, and there will not be. If your buyer is a private individual at a home address, this is the wrong product.
Buying a list to send from somewhere else
Autopilot is not a contact export. Scraped lists blasted from an unwarmed domain are the fastest way to end up in spam permanently, and we will not be the supplier of that.
You need it live this week
A dedicated domain takes weeks to warm properly. You can start on the shared pool the same day, but if a big number has to be sent by Friday, the honest answer is that it should not be.
If you are not sure which side of the line you are on, ask us on the call — the compliance FAQ covers the common cases and we will tell you plainly if the answer is no.
Point it at your market and watch the first batch
Bring your domain. In one session it will read your business, propose ICPs and draft the first emails — before you have committed to anything. Our company and people catalog is being loaded. Until it is live, Autopilot writes to contacts you import into a campaign.
- $30 wallet, no card
- Sending infrastructure included
- Cancel by turning it off