Campaign Autopilot, honestly: what it does and what it refuses to do
Every outbound tool promises the same thing, so here is the part most of them leave out: the mechanics. Every number on this page is either a setting you can read in the product or arithmetic you can do yourself, and the interactive parts run the same rules the engine runs, on your own text.
That is the whole point of this post. Not what outbound could be. What this one actually does, and where it stops.
The problem is not the list
Nobody is short of ways to build a list. What breaks is everything after it: the message that needed writing for this specific person, the follow-up that was due on Tuesday, the reply that came in while you were on a call and got answered four days later, and the fact that all of it lives in a different tool from the CRM where the deal is supposed to end up.
So the work splits into two kinds. One kind needs judgement: who is worth talking to, what to say to someone who just told you their setup is fine, when to let a thread go. The other kind is just diligence: capture the profile, wait the right number of days, notice the reply, keep the history in one place. Diligence is what people quietly stop doing in week three.
Campaign Autopilot takes the diligence. It does not take the judgement, and it is built so that it cannot.
How a campaign actually runs
You write a couple of sentences about who you want to reach. The AI turns that into a plan: the search that finds those people, an ideal customer profile in words, and a sequence of messages with the waits between them, including an optional A and B version of the first message. You read it, change what is wrong, and approve. Nothing exists until you do.
From there the browser extension does the repetitive part from your own LinkedIn account, at a human pace, while your browser is open. Captured leads park in a review state until you look at who was actually found, because a bad search filter should not become a hundred sent invitations. When someone replies, their sequence stops on the spot and the reply is drafted for you to review.
What is automatic, and what is not
This table is the honest centre of the post. The pattern is deliberate: the system is allowed to be tireless, and it is not allowed to be the one who decides.
| Step | Who does it | Guard |
|---|---|---|
| Campaign plan and messages | AI writes | You approve before anything exists |
| Finding and saving leads | Automatic | Parked for your review before any are contacted |
| Invites and sequence steps | Automatic, from your own account | 25 a day, spaced out, only while your browser is open |
| Someone replies | Automatic | Their sequence stops immediately |
| Answering that reply | AI drafts | You send it. Auto send exists, is off by default, and is capped at 15 a day |
| Follow-up when a thread goes quiet | AI drafts a nudge | Never sent on its own, ever |
| Collateral for the campaign | AI writes, on request | Six formats, in your own colours, gaps marked rather than invented |
The numbers that are real
These are defaults you can read and change in the product, not results.
A lead marked not interested is never nudged, because that default is off. A lead who has never replied and is still moving through a campaign is never nudged either: the campaign is already their follow-up, and two systems writing to the same stranger is how outbound starts feeling like spam.
The part I have not seen elsewhere
Four rounds of real feedback on a real thread taught me something I did not expect, and it is now the design principle of the whole drafting layer.
A rule written into the prompt can be ignored. A check that runs after the text is generated cannot.
The same defect survived three rewordings of the instruction. It stopped the first time it became arithmetic. So the drafter now has both: rules in the prompt, and a set of checks that run on every generated reply and reject it before a human ever sees it. These are the checks, exactly as they exist in the code:
- Uses filler that means nothing: "hope this finds you well", "circling back", "touching base", "synergies", "quick 15 minutes".
- Borrowed authority: "we often see", "teams like yours", "most teams". A claim about a pattern across other people's companies is unprovable to the reader and usually invented, so it is banned outright rather than left to judgement.
- Asks more than one question, which reads as an interrogation.
- Asks something they already answered in their last messages.
- Makes them choose between two options you invented, when they have shown no intent.
- Tells them their situation is a problem after they said it was fine.
- Answers a short message with a long one. If they wrote 12 words, three paragraphs back is pressure. The limit is roughly three times their length.
- Asks for an open ended assessment ("what do you think is missing") instead of one small specific thing.
Try it on your own draft
This runs the same checks, in your browser. Nothing is sent anywhere. Paste something you would actually send, and what they wrote to you, since two of the checks depend on it.
One check is missing here and present in the product: the similarity test that spots a question they have already answered in their own words. It needs the full conversation, which stays in your workspace.
What the pace actually gets you
Outbound arithmetic is not mysterious, and a vendor should not be the one supplying the reply rate. So the calculator below asks for yours. If you have never run outbound, leave it and treat the output as a shape rather than a forecast.
Your numbers, not mine
The daily cap is the engine's real default. Everything else is yours to set.
The hours line counts only the mechanical part: opening the profile, writing, sending, logging it. Reading replies and deciding what to say is still yours, and should be.
What it will not do
If any of this is a dealbreaker, better to know now than after you have moved your pipeline in.
- It does not run while your browser is closed. The engine works from your own session in your own Chrome. Close it and the campaign pauses until you are back. That is a deliberate trade: the alternative is handing over credentials to a server that logs in as you.
- LinkedIn restricts automation. Their user agreement is clear about it. Running from your own browser at a human pace with no stored password keeps the risk as low as this can be, but the account risk is yours, and I will not tell you otherwise on a sales call.
- The AI can be wrong. Every draft is a draft. The checks above catch the failures I have seen repeatedly. They will not catch a claim about your own product that only you know is untrue.
- Free LinkedIn search cannot filter "these twelve companies". That facet only exists in Sales Navigator. Where a play needs it, the product says so on the screen instead of quietly returning the wrong people.
- You are the sender. GDPR, ePrivacy, CAN-SPAM and their local equivalents apply to you, not to the tool. The product will not pretend otherwise.
Why this is built this way
The market is full of tools that will send more, faster. That is not a hard problem, and it is not the one worth solving. The hard part is a system that is tireless about the mechanical work and unwilling to speak for you, that stops the moment a person answers, and that tells you what it could not do instead of papering over it.
An outbound engine that quietly sends a bad message under your name has not saved you an hour. It has spent something you cannot buy back.
If you run outbound, tell me on the way in which of the limits above matters most in your setup. That is what the next release gets built from.
Join the Waitlist