Your entire sales team can run for $691 a month. You just won't build it.

How the data gets sourced without bleeding credits

Sourcing is a waterfall, ordered cheapest-first, because the providers that find verified email addresses are the ones that charge per lookup.

Apollo pulls names, LinkedIn URLs, and firmographics for $49 a month. It deliberately does not pull emails, because that burns Apollo's credit budget on the most expensive field. Clay then orchestrates the fan-out: it backfills any missing LinkedIn data, runs a web search for a trigger signal, and only then goes looking for the email. The email finder is LeadMagic first, at roughly 3 cents per lookup; Findymail is a fallback at 4 cents. Hard stop at two providers. A free validation pass confirms the address before anything moves forward.

The trick that keeps this cheap is bringing your own API keys for the expensive lookups instead of paying Clay's marked-up credits. That alone cuts enrichment cost by about 45 percent. The result is a verified email for roughly 70 percent of contacts and a real trigger signal for about 60 percent.

That trigger signal matters more than it looks. It's the difference between an email that references something true about the prospect's last 90 days and an email that opens with "I came across your profile."

The qualification gate that says no for you

Every enriched contact goes to a cheap, fast model whose only job is to be strict. It scores the fit on a scale of 1 to 5, selects a messaging angle, and flags a disqualification reason if one exists. The instruction it runs under is blunt: the cost of contacting a bad-fit lead is higher than the cost of missing a good one. When in doubt, reject.

The gate is a hard rule, not a suggestion. Score four or higher with a real trigger, and it advances. Anything less, it's marked disqualified, and the process stops. No human reviews the rejections. That's the point. This is the discernment you currently apply by hand, at the rate of maybe twenty contacts before you get pulled into a client call, running instead across hundreds without fatigue.

It costs about sixty cents a month to run.

The personalization engine, and why it doesn't hallucinate

This is where most automated outbound dies, and it's the part your prospects would smell instantly if it were wrong, so it's worth the detail.

Before the AI writes a single word, there's a hard data gate. No verified trigger signal, no AI. The contact falls back to a static template instead. The model is never once asked to be interesting about a person it knows nothing about. That single rule kills the most common failure mode in cold email.

When the model does write, it's the more capable one, and it's boxed in hard. It sees only a structured block of verified facts. It's banned from the tells that make cold email read like cold email — the "I noticed," the "hope this finds you well," the "congrats on." Subject lines are capped at a few lowercase words. Openers are one-sentence statements that make an observation and connect it to a likely problem. No flattery, no invention.

Then the guardrail. A second, cheaper model reads what the first one wrote and asks one question: does this reference any fact, number, date, or name that wasn't in the source data? If yes, the line is regenerated at a lower temperature, and if it fails twice, the contact is dropped entirely. One model writes, another audits, before a word reaches a prospect.

Everything is stored in a real database with an append-only audit log. Three weeks later, you can ask exactly what the model saw when it wrote a given opener. That is not how slop is built. It's how infrastructure is built.

Deliverability, the unglamorous part that decides everything

You can write the best cold email in the world, and it means nothing if it lands in spam. This is the work a founder doing outbound from their main domain at midnight never does, and it's why their open rates quietly collapse.

The setup is five separate sending domains, none of which is your primary brand, with ten mailboxes spread across them. Every domain carries full email authentication from day one. Each mailbox goes through a 21-day warmup, sending gentle volume to friendly addresses, before it's allowed near a real prospect. Even at full tilt, each inbox is capped at a low daily send to stay under the radar of spam filters.

It is tedious and non-negotiable. It's also exactly the kind of thing a system does perfectly, and a busy founder never does.

The replies handle themselves

When a reply comes in, a fast model reads it and sorts it into one of six buckets. Out of office, wrong person, unsubscribe, and not-interested are all handled silently. The sequence pauses, or the lead is suppressed, or an alternate contact is queued, with no human involved.

Only two categories ever reach you: interested and a genuine question. Those trigger a notification. Out of roughly fifteen replies a month, about six need your attention. The rest of the machine clears on its own.

This is what turns the whole thing into a 30-minute-a-week job instead of another inbox to drown in.

The unit economics

Here's the part that should sting.

The total stack cost is roughly $691 per month. Clay at $149, Instantly for sending at $78, the mailboxes at $78, Apollo at $49, and AI compute capped at $50 that never gets close. The cost per qualified contact is around $2.76. Cost per booked meeting is around $138.

An outsourced SDR costs you $200 to $500 per meeting and needs to be managed, motivated, and replaced. So the machine is 2x more capital-efficient than hiring, the thing you keep deferring until revenue allows. The revenue was never the blocker.

Building it is five weeks of one engineer's time. That's the whole ask.

What does this cost you to ignore

While you remain in the pipeline, this system will send 500 considered, verified, personalized emails on your behalf each month. Every month. Whether you're in a client meeting, asleep, or on holiday for the first time in three years.

You are not faster than the machine. You are more expensive and less consistent, and you are the single point of failure; your eventual buyer will discount you 30 to 50 percent.

Most of you won't build this. Not because it's hard. You won't build it because being the pipeline feels like control, and handing it to a system feels like losing your grip on the one thing you're certain you're good at. That's the real barrier. It was never the money.

The founders who break past $15M aren't the ones who work harder at being the pipeline. They're the ones who decided the pipeline shouldn't depend on them being awake.

The blueprint exists. The maths works. The only variable left is whether you'd rather stay indispensable or get free.

If you want to see the full architecture, I'm happy to walk you through it.