The AI SDR math: what $500 a month of automation actually replaces

Every AI SDR pitch runs the same comparison: the agent costs $250 to $500 a month, a human SDR costs $60,000 a year, therefore the agent is a 90 percent discount on a salesperson. An AI SDR is software that runs some or all of the outbound loop on its own: finding accounts, researching them, writing the email, sending it, and in the boldest versions handling the reply. The $60,000 comparison is the wrong benchmark, and it is wrong in a way that flatters the category.

Nobody running small-team outbound is choosing between an agent and a hire. They are choosing between an agent and more of what already works: more domains, more mailboxes, a bigger verified list, a higher sending tier. Cold outbound at this scale is a volume game played at 1 to 3 percent reply rates, and the honest question is whether the agent's marginal reply beats the marginal reply from another $350 of infrastructure. So that is the comparison this piece prices, both paths at roughly $500 a month, at the published rates of tools in the SaaSTracker database. The arithmetic is not close.

Half the category will not tell you the price

Start with a number that should worry anyone evaluating these tools: of the 16 AI SDR products we track, only 8 publish a flat dollar entry price at all. The median among those that do is $44.50 a month, which sounds approachable until you notice what sits at that price point. The cheap end of the category is the research-and-copy half: Regie.ai's Pro plan at $49, Floqer at $49 for 2,000 credits, tools that draft and enrich but leave sending to you.

The full agents, the products that actually promise to replace the loop, cluster far higher and bill in ways that resist comparison. AiSDR starts at $250 a month for its Solo tier, and that price is metered: 200 contacts. Its Explore and Scale tiers, at $900 and $2,500 a month, are billed quarterly in advance, so the smallest realistic evaluation above Solo runs about $2,700, and there is no free trial on any tier. Salesforge sells Agent Frank, its autonomous SDR, from $499 a month, also billed quarterly, so the smallest realistic commitment is roughly $1,500, with mailboxes a separate purchase at $33 to $69 a month unless you bring your own. Regie.ai's own history is instructive: before it published the $49 self-serve tier, its seat-based pricing ran roughly $180 to $499 per user per month, and that is still the anchor its Enterprise quotes negotiate from.

Quarterly prepayment, demo-gated tiers, contact metering at $1.25 per name: none of this is how confident software prices itself. It is how software prices itself when the customer might do the math.

Same $500, two ways

So do the math. The infrastructure path below starts from the $147.50 complete outbound build we priced in our under-$150 playbook: three secondary domains, ten mailboxes, warming, list building, verification, and sending. The remaining $350 or so buys the scaled version of the same machine. The agent path is a named tool at its published rate.

Same $500	Line item	Monthly cost
Infrastructure path	Complete base stack (3 domains, 10 mailboxes, warming, 1,000 contacts, verification, sending)	$147.50
	12 more secondary domains, Mailforge at $14 a year each	$14.00
	40 more mailbox slots, Mailforge at $3 per slot, annual billing	$120.00
	Instantly upgrade, Growth to Hypergrowth	$60.00
	GetProspect upgrade to Growth 5K, 5,000 verified contacts a month	$35.00
	MailStrike upgrade to Growth, warming plus placement testing across the fleet	$40.00
	Second MillionVerifier credit block	$20.00
	Apollo.io Basic, one seat, supplementary data and signals	$49.00
	Total: 15 domains, 50 mailboxes, ~5,000 new contacts a month	$485.50
Agent path	AiSDR Solo, 200 contacts a month, domains and warmup included	$250.00
	or Salesforge Agent Frank, billed quarterly, mailboxes at $33 to $69 extra unless you bring your own	$499.00

One credit where it is due: AiSDR bundles domains, mailboxes, and warmup into the plan price, which is genuinely unusual and worth roughly $50 to $200 a month of tooling on the other path. The comparison above already grants that, because the infrastructure column carries those costs explicitly and the agent column does not.

But look at what each column produces. Fifty warmed mailboxes at 20 to 30 sends a day support 1,000 to 1,500 daily sends, and the list side feeds them about 5,000 new verified contacts a month. The Solo agent touches 200 contacts a month, full stop. Twice the money buys Agent Frank, and Salesforge's fine print notes that new sending domains need roughly two weeks of warmup before live campaigns, so the early weeks of the quarterly commitment are paid ramp either way. AiSDR's own billing notes say the same thing more bluntly: new domains take thirty to sixty days to reach full capacity, so the first month of any contract is largely warmup you are paying full price for.

The breakpoint: what reply rate does the agent need?

Here is the arithmetic, simple enough to redo with your own numbers.

Cost per contact on the infrastructure path: $485.50 divided by 5,000 contacts is about $0.10. Cost per contact on AiSDR Solo: $250 divided by 200 is $1.25. The agent's contact costs 13 times as much, so the agent wins the math only if its reply rate is more than 13 times yours.

Put reply rates on it. At an unremarkable 1 percent, the infrastructure path returns about 50 replies a month, $9.71 each. For the agent to match that cost per reply, it needs 26 replies from its 200 contacts, a 13 percent reply rate. If your manual writing already earns 2 percent, the agent needs 26 percent. Nobody, human or agent, holds a 13 percent cold reply rate across a quarter, and the pitch decks that imply otherwise are measuring warm segments or counting out-of-office autoresponders.

The multiple is the whole argument, so state it as a rule: an agent beats infrastructure only when its reply-rate multiple exceeds its cost-per-contact multiple. Every vendor in this category should have to answer that inequality, and the pricing opacity documented above is easier to understand once you realize most of them cannot.

Two honest caveats, because the infrastructure column has its own soft spots. Volume scales linearly only while list quality holds; a team that exhausts its addressable market at 2,000 good-fit contacts a month should not buy capacity for 5,000. And the infrastructure path spends your hours, a few per week on list review and copy, where the agent spends none. Both caveats narrow the gap. Neither is a factor of 13.

The part worth automating today

None of this means the AI is useless. It means the AI is pointed at the wrong end of the loop. The stage where an agent already earns its keep is the front: account qualification, signal gathering, first-pass research, the work of deciding who deserves an email before anyone writes one. Errors there are cheap and visible. A bad research summary wastes a credit; you read it, discard it, and no prospect ever knows.

The tools built this way look sane on price and sane on architecture. Aomni sells the research half on its own, $300 a month flat for three team members with a free tier to test against your own accounts, and it deliberately does not send at volume, which we read as understanding exactly where the value sits. Floqer lets you assemble the research workflow yourself at $49 a month for 2,000 credits, with unlimited seats, every action showing its credit cost before it runs, and credits rolling over for twelve months. Regie.ai at $49 gives you drafting, enrichment, and agents while sending through your own Gmail or Outlook, which keeps deliverability where it belongs: visible, and yours.

Full-loop autonomy is a different bet, and the failure mode is what kills it. When a junior SDR writes a bad email, you see it in the sent folder that afternoon. When an agent's sending pattern trips a filter, nothing announces it. Reply flow just thins, week over week, while the agent keeps reporting activity, and by the time the silence is unmistakable the damage has compounded across every domain it touched. A bad agent burns in a fortnight the same domains a good operator spent a month warming, and the monitoring layer that would catch it early is precisely the part the autonomous pitch says you no longer need to think about.

Where this category is in 18 months

Our call, plainly: most of it dies. The research half does not die so much as dissolve; account research and signal gathering are features, not companies, and the platforms that own the surrounding workflow are already selling them. Clay sells AI research by the credit inside an enrichment product. Apollo sells data, sequencing, and research credits on one $49 bill. When a sequencer or a data platform offers the same first-pass research as a checkbox, a standalone research agent needs to be dramatically better to justify a second invoice, and "dramatically better at reading websites" is not a durable moat.

The autonomous-sending half has a harder problem than competition: its unit economics require the reply-rate multiple this article just showed to be implausible, and its delivery channel is controlled by mailbox providers who owe it nothing. Google and Yahoo's bulk-sender rules already enforce authentication and spam-rate thresholds as of 2026, and every tightening of those screws hits high-volume templated mail first, which is what most agent output is under the personalization varnish. Quarterly prepaid billing is what you charge when you expect the third month to be a cancellation conversation.

What would have to be true for the bet to flip? The strongest version: mailbox providers choke volume outbound so hard that the 1 percent game stops working entirely, and the only viable motion becomes low-volume, deeply researched, genuinely individual mail. In that world the economics invert, because volume infrastructure buys nothing and research quality buys everything. It is not a crazy scenario; it is the direction the screws are turning. But notice it is an argument for the research tools, which survive that world as features, not for autonomy, which still has to send through the same filters as everyone else.

And there are two structural niches where an agent genuinely fits now. The first is the buyer with no time at all rather than little time: the solo technical founder who will simply never do outbound otherwise. The infrastructure path at $485.50 assumes hours nobody in that situation has, and 200 agent-sourced contacts a month beats zero by an infinite multiple; AiSDR's all-inclusive Solo tier is honestly shaped for exactly this buyer. The second is the team whose market was never a volume market: a few hundred addressable accounts, where 5,000 contacts a month is meaningless and the only lever that exists is research depth.

The short version

Benchmark an AI SDR against $500 of sending infrastructure, not against a $60,000 hire. At published rates, $485.50 of infrastructure runs 15 domains, 50 warmed mailboxes, and about 5,000 verified contacts a month at roughly $0.10 per contact; the same money buys an agent 200 to 400 contacts at $1.25 each, so the agent must reply-convert about 13 times better to break even, and it will not. Buy the research half, which is cheap, inspectable, and already priced like a feature. Skip the autonomous half unless you are the founder who would otherwise send nothing, and expect most of this category to be absorbed or gone within 18 months.