How an AI outbound run works
An AI outbound agent is only as good as its loop. On Moatt the loop has four fixed steps: signal, draft, gate, memory. A monitoring sweep watches your signals, subreddits and mentions and files what matters. When a company shows a buying signal, the agent drafts a first email to the right person at that company. At this point nothing has been sent, and nothing will be until you say so.
The gate is the step most tools skip. Every run pauses there with its credit price shown before anything fires. A batch of five outreach drafts costs one credit, and you see that number before you approve. Approve, and it fires. Ignore it, and nothing moves. There is no setting that lets the agent send on its own.
The last step is the one that pays off later. Replies and outcomes land in shared memory as either lessons, which are backed by a measured result, or hypotheses, which are still untested. The next batch starts from those lessons instead of starting from zero, which is the whole point of running outbound through an agent rather than a sequencer.
- Signal — a sweep files what matters from your signals, subreddits and mentions
- Draft — a first email to the right person, written on the signal
- Gate — the run waits with its credit price shown; you approve or nothing moves
- Memory — the outcome is recorded and the next batch starts smarter
What an AI SDR replaces — and what it doesn't
Run outbound through one agent and a few line items drop out of the stack: the separate monitoring tool that watches mentions and signals, the hours spent drafting first emails, the competitor-research subscription, and the scheduler you used to chain it all together. On Moatt, one sentence in chat becomes a scheduled run, so a daily sweep fires with your laptop closed.
Be honest about what stays. Your sending accounts stay yours: the agent asks for the account when a run needs it, you approve an OAuth popup, and the agent never sees a password. Your CRM stays. And Moatt does not ship a contact database — dedicated AI SDR platforms with large built-in databases do that job well, so if raw contact data is your bottleneck, keep a data provider next to the agent.
What outbound automation costs, run by run
Most AI SDR tools price by seat, by tier, or by quote — several publish no pricing at all. Moatt prices by run, and the rate card is public. A monitoring sweep is 1 credit. A batch of 5 outreach drafts is 1 credit. A full competitor profile, useful before a new outbound push, is 3 credits. Conversation with the agent is free; only runs are billed.
That makes the arithmetic doable before you buy. A sweep plus an outreach batch every weekday is two credits a day, around 44 a month — inside the Starter plan at $99 for 60 credits. Growth is $299 for 250 credits, Scale is $799 for 800, and annual billing takes two months off. Unused credits roll over up to twice your plan, failed runs refund automatically, and overage bills at 1.25x with a cap at 3x and alerts at 50, 80 and 100%.
Lessons, hypotheses, and what the agent actually learns
Most outreach tools test copy variants and show you a dashboard. Moatt's memory splits what it knows into two piles: lessons, backed by a measured outcome, and hypotheses, not yet tested. An angle that earned replies becomes a lesson. A hunch about tone stays a hypothesis until a batch proves or kills it. You can see which pile a claim sits in, so you know when the agent is working from evidence and when it is guessing.
That memory is shared across disciplines, and outbound is where the sharing shows up first. A competitor profile run for research feeds the next outreach batch. What community threads taught the agent about objections shows up in cold email drafts. One agent, one memory — you never have to brief the outbound tool on what another tool already learned.
When not to use an AI outbound agent
Skip the agent when the list is twenty people you already know. A founder writing to former colleagues will beat any AI cold email agent on warmth, and there is no signal stream to watch. The same goes for a single high-stakes deal: one buyer, one thread, one human. An agent earns its credits on volume it would be tedious to watch and draft by hand.
Skip it too when the message isn't settled. If you can't yet say who you sell to and why they switch, the agent will draft consistent emails around an unsettled story, and consistency is the wrong goal at that stage. Get a handful of manual replies first, learn what lands, then hand the working pattern to the agent to repeat.
Finally, if what you want is full autopilot — sending with no human in the loop — Moatt is the wrong shape on purpose. The approval gate is the product. Tools built for hands-off sending exist and some run it well; they just leave you owning whatever goes out under your name without having read it.