A recruiter books intake calls with engineering leaders hiring for AI roles

You are a recruiter who places machine learning and applied AI engineers. Your buyers are the engineering leaders at software companies that are building AI into their product and cannot hire fast enough: the VP of Engineering, the CTO at a smaller company, the Head of AI or Machine Learning at a larger one. This use case builds that list in the four states where most of these teams sit, opens every email with the AI roles that company actually has open, and turns the replies into intake calls.

Step 1. Write the sentence first

Write the buyer in one sentence. Ours is this. "Engineering leaders at software companies building AI products, with 51 to 1,000 employees, in California, Washington, New York and Massachusetts." The size band matters here. Below 50 people the founders often hire through their own network. Above 1,000 there is an internal recruiting team and a vendor list.

Step 2. Set the filters

Open Search, then People, and set these. Read the count after each one.

Open this search in Argorant

Step 3. Read the count, then twenty rows

For this sentence, expect a count in the low to mid thousands. Here is how to read it.

• Above 15,000 means the keywords or the states did not apply. Check the chips above the results.
• 2,000 to 8,000 is the working range. Plan a first wave of about 2,000 people.
• Under 800 means the industries are too strict. Remove Data & Analytics first, then try without the industry filter and let the keywords carry the search.

Then scroll twenty rows. At companies with 51 to 200 people, you will mostly see CTOs. At 501 to 1,000, you will see directors and heads of AI. Both are right. What is wrong is a CTO at a company whose product has nothing to do with AI. If you see several, add LLM or generative AI as the only keywords and compare.

Step 4. Save the list, keep only valid emails

Click Select matching, choose how many people you want, then Save list. Name it so you know what it is in three months, for example US CA WA NY MA - AI software 51-1000 - eng leaders. A saved list keeps the filters, so next quarter it shows the market as it is then.

When you attach the list to a campaign, choose Valid under which emails to include and nothing else. Every address is checked before it can be sent to. Each person from search costs 1 credit, unless you already unlocked them before.

Step 5. Add the AI roles line

No filter can tell you which of these companies has an AI role open this month. Their careers pages can, and so can their job posts. That one fact is what makes an engineering leader read a recruiter's email, so it gets its own variable, {{custom_ai_roles}}, with a fallback for every company where you found nothing.

There are two ways to fill it.

• From a file. Export the list, add a column called custom_ai_roles with the open AI roles per company, and upload it to the campaign with Upload CSV. Columns you keep as custom fields become variables with the same name.
• From Claude or ChatGPT. Connected to Argorant, your assistant reads each company's careers page, writes the line, shows you five examples with the source first and saves the rest when you agree. If a company has no AI role posted, it keeps the fallback.

For my campaign "Demo: AI engineering leaders", set up a personal line called custom_ai_roles.
It says which AI or machine learning roles the company has open, from its careers page or job posts,
as a phrase that fits after "Saw that", 6 to 14 words,
e.g. "you have two senior ML engineer roles open in San Francisco".
Fallback: "you are growing the team that works on your AI features".
Show me five examples with the source first. Never guess a role or a location.

Companies with an AI role open right now are your best rows. If you fill the line for the whole list, start the campaign with the companies that have one, and send the fallback group in wave two.

Step 6. Write the three emails

Press New campaign on the Campaigns page, name it, attach the list and go to Emails. Keep the claims to what you can prove on the intake call. Engineering leaders check.

Their roles, what kind of engineer you find, how you charge, one small ask. Three anonymized profiles is easy to say yes to.

A new reason, here how you screen. Engineering leaders care more about how you filter than how many you have.

The polite exit that asks for the right person. At larger companies a talent lead often owns the agency decision.

Press View preview before you go on. With no leads yet, it fills every variable with your own details, which shows each fallback in place. "Saw that you are growing the team that works on your AI features" is fine for a company with nothing posted.

Step 7. Mailboxes for 2,000 people in 4 weeks

Cold email never goes out from your main domain. It goes out from separate mailboxes on domains that look like yours, for example a short variation of your brand name. If one of them gets tired, you retire it and your real domain never notices.

Here is the math for this campaign. 2,000 people over 4 weeks is about 100 new people per weekday. Each gets three emails, so once the follow-ups overlap you send roughly 300 emails a day. At 5 cold emails per mailbox per day, that is about 60 mailboxes.

There are two ways to get there. Under Outreach, Inboxes you connect mailboxes you already have, from Google, Microsoft or any other provider. Or you have new Microsoft 365 mailboxes set up on new domains. Those come in bundles from 100 mailboxes, you do not need a plan to buy them, and they connect to your campaigns on their own. 100 mailboxes carry 500 cold emails a day, which leaves room for a second campaign.

New mailboxes wait at least 24 hours before they send anything. Warm-up is included and starts switched off. Switch it on for new mailboxes and give them one to two weeks before full volume. Never raise the limit per mailbox to go faster. More mailboxes is the right lever.

Step 8. Schedule and launch

Your list covers two coasts, and one campaign has one timezone. On Pacific time, an email sent at 9 lands at noon in New York, which is fine. On Eastern time, an email sent at 9 lands at 6 in the morning in California, which is not. So either run one campaign on Pacific time, or split the list into a West Coast campaign on Pacific and an East Coast campaign on Eastern with the same emails. Splitting also tells you which coast answers better.

In the Senders step of the wizard, every connected mailbox is ticked. Untick the ones that belong to other campaigns or brands. Sends rotate across the ones you keep, and each one still respects its own daily limit.

In the Schedule step, set the timezone of the people you write to, here US Pacific (PT), not your own. Keep the sending window at 09:00 to 17:00, keep Skip weekends on and keep the gap between sends per mailbox at 60 to 120 minutes. Plain text stays on, so the email looks like it came from a person and carries no tracking pixel.

The last step is the review. Read it as if somebody else built the campaign. The audience with the grades you picked, the sequence with its waits, the senders, the start and the pace. Then press Launch campaign. Launch is not possible without senders and leads, and mailboxes still in their first day hold the launch until they are ready. Afterwards, open the Settings tab and check that Stop on reply and Bounce protection are on. Bounce protection pauses the campaign when more than 10% bounce once 100 people have been contacted.

Step 9. Replies and booked calls

Every reply lands in the Unibox, from every mailbox and campaign, and gets a label on its own. The standard labels are Interested, Meeting request, Out of office, Unsubscribe and Other. Switch on Referral and Wrong person in the settings as well. On our own largest campaign, one in four human replies pointed us to a colleague, more often than a straight yes.

A yes gets answered by the campaign itself, with a subsequence. It starts when a reply contains one of your keywords, takes the person out of the main sequence and answers in the same thread in the next sending window. A reply that says "not interested" or asks you to stop never triggers one.

Open the campaign's Subsequences tab, press Add subsequence and use these keywords. interested, send profiles, send them, tell me more, fee, terms, sure. Then write the first email.

Asking for the intake call before sending profiles is what serious recruiters do, and engineering leaders respect it.

Expect questions about your fee and your terms. Switch on the optional Question label too, so they are easy to find. They deserve a written answer from you the same day, not a subsequence.

Every morning, filter the Unibox by Meeting request and Interested. Those are the conversations one answer away from a call. Propose a specific time instead of just a link, and add your booking link as the second option. When the call is set, change the label to Meeting booked so your numbers count it. Answer referrals the same day. Thank them, write to the person they named, and mention who sent you.

If you would rather work from Claude or ChatGPT, your assistant can read the Unibox, sort replies by label and draft answers. Nothing is sent until you confirm. The use case Run it all from Claude or ChatGPT shows how.

What to expect

The ranges below come from our own sending, scaled to 2,000 people. Our healthy campaign, two emails to a valid-only list from mailboxes at 5 cold emails a day, got human replies from 1.3% of the people contacted. Campaigns from overworked mailboxes sat at 0.3%. About one in five human replies says yes, and about one in four points you to a colleague. These are reference points, not promises. Your offer, your segment and your copy move them a lot. The reply rate in Argorant also counts out-of-office notes, so it reads higher. Compare these numbers with the reply rate excl. OOO.

Engineering leaders get many recruiter emails, so the share of replies is often at the lower end of the range, and a fair number are "we only hire directly". The ones that do answer tend to have a real role and a real deadline. The companies with an open AI role in the personal line answer clearly better than the fallback group, which is why wave one should be those companies.

On the calendar it looks like this. The first replies arrive in week one, 65% of all replies come within an hour of the email landing and 84% answer the first email. After 4 weeks everyone has had the full sequence, and a week later nearly everything that will come is in. If human replies stay near zero after a few hundred people, check delivery first and copy second.

Do the same from Claude or ChatGPT

Every step above also works as a conversation, once Argorant is connected to your assistant. This is the request that builds the list. Your assistant counts, shows sample rows and asks before it spends a credit.

The full setup, from connecting the assistant to launching and handling replies, is in Run it all from Claude or ChatGPT.