Lead Generation for a Talent Intelligence AI Platform

A quick note on what we were delivering

Almost every Belkins engagement ends in a booked appointment. This one didn’t, which shaped every decision in this project.

Zeki Talent Market is a marketplace, so it needed two separate audiences in place before it could function. On one side, exceptional engineering and go-to-market candidates; on the other side, funded scale-ups willing to hire them and pay a placement fee. Belkins was brought in to reach out to the candidates first and the businesses a few weeks later.

So the unit of delivery was a registration: someone from Zeki’s vetted list who read a cold message, went to a landing page, and signed up for a product that hadn’t launched yet.

This made the outreach sophisticated in a specific way. In a standard campaign, an SDR books a meeting or offers a demo, and the client’s sales team picks up the conversation from there. In this project, there was nothing to offer but an invitation, and no one to hand a warm lead to. Candidates had to act on the cold message itself, leaving the campaign to the message and how many of the right people saw it.

It also had to happen at an unprecedented scale for our team. Zeki’s vetted candidate list held more than 200,000 names. Cold conversion on an offer like this was assumed at 2% or lower, so working backwards from a waitlist of several thousand meant contacting almost the entire list within several weeks.

The results: a waitlist built from a standing start

Zeki set the launch for the week of July 13, 2026. In under ten weeks, two SDRs contacted a quarter of a million people on Zeki’s behalf and sent three-quarters of a million emails. Here are the raw numbers:

• 760,524 emails delivered across the campaign
• 249,973 candidates reached across email, LinkedIn, and phone
• 979 registrations from Zeki’s vetted candidate list, before and after launch
• 10x improvement in email conversion between month one and month two
• First campaign live in 5 business days, against a standard 30-day domain warm-up

Outreach to talent began in mid-June, carried through the mid-July launch, and continued after it. Outreach to companies began in early July and was paused in August. Both sets of numbers are below.

Every channel at least doubled its conversion rate between month one and month two as messaging tightened. In month two, LinkedIn converted at more than eight times the rate of email while reaching about 25 times as many people.

Outreach to the hiring side of the marketplace ran for roughly six weeks against a vetted list of about 4,000 funded scale-ups.

• 35 businesses agreed to register or take a demo with Zeki Data’s sales team
• 5 businesses registered on the platform directly
• LinkedIn was the strongest channel for reaching founders, CTOs, and VPs of engineering

Those numbers fell well short of what the candidate side was producing, and the campaign was paused in August rather than continued. What that decision was based on is covered further down.

The context: The client had the data but not enough people

Zeki Data came with an unusual challenge for a company approaching a lead generation agency — they already had the leads.

Their data team tracks around 5 million technical profiles. From that pool, they had narrowed and hand-vetted a list of more than 200,000 candidates, each carrying more than 50 verified data points:

• Machine learning and LLM engineers
• PhDs
• former founders
• Go-to-market engineers (GTME) and leaders

Not the top 1%, whom Zeki works with separately, but the tier directly below it: the people scale-ups hire when the next hire can make or break the company.

Our client didn’t have a team that could reach them. Zeki Data’s marketing included five people, and their focus was building the brand. Outreach ran through email alone, out of HubSpot, producing around 10 signups a day. The launch needed several thousand people on the waitlist, and pushing that volume through the client’s own domain risked burning it.

The other gaps were just as concrete:

• No calling capacity. Zeki had no sales team and nobody to make the calls, which cut out the channel that tends to work best on senior technical audiences who ignore email.
• No LinkedIn outreach. Executive profiles existed, but nobody was running structured outbound from them.
• No phone numbers. The dataset was built for talent intelligence, not for cold calling. It carried emails and LinkedIn URLs but not the mobile numbers a calling campaign runs on.
• A strict deadline. The 300-candidate target had been committed during a funding round, and the platform build was locked to the week of July 13.

Why they chose Belkins

The client had spent weeks vetting agencies before the first call and arrived with a decision close to being made. Belkins won the work for three reasons:

1. We offered cold calling, a channel Zeki had no way to build in the time available.
2. We had run waitlist campaigns for platform launches before, so the approach was not theoretical.
3. Onboarding could compress to five business days instead of 4–6 weeks like in other agencies (while at Belkins, our standard onboarding process takes 14 days).

Partnership objectives

• Build a waitlist of several thousand registrations ahead of the mid-July launch, working from an assumed cold conversion rate of 2% or lower
• Support Zeki’s day-one goal of 300 active candidates on the platform
• Work only from the client’s vetted lists, with no prospecting or list building
• Hold the platform’s target ratio of roughly one company to ten candidates
• Do it without damaging the deliverability of clients’ own domain

The strategy: infrastructure first, then volume

Most Belkins engagements open with research and list building. This one started immediately with email infrastructure, because the list was already the strongest asset on the project.

Compressing onboarding into five days

A standard cold email campaign warms new domains for approximately 30 days before the first send. That would have consumed half the runway. Instead, the team sent from Zeki’s existing email infrastructure and layered new domains on top as the campaign scaled (more than 20 of them by the end), rotating mailboxes continuously and switching between sending tools to spread the load. First campaigns went live five business days after kickoff.

That decision is also what made the following weeks difficult, and the team knew it at the time. Sending 80,000 cold emails a week to personal inboxes is more than most sending setups will tolerate. It required daily domain health checks by a deliverability specialist watching for early warning signs and moving volume off a domain before it degraded.

Verifying data that was already good

Zeki’s data was the cleanest Belkins’ research team had worked with, but that didn’t remove the verification step. With our proprietary tool, our dedicated researcher re-checked domain lists, cross-referenced decision-makers on the company side, and added the one field Zeki’s dataset didn’t carry: phone numbers for the calling campaigns. Zeki exported leads weekly from their HubSpot instance into a shared master sheet so both teams could deduplicate against each other’s active lists.

Building messaging around what candidates valued

The offer had no price, no product to buy, and no calendar link to book a demo. So the messaging leaned on the four points Zeki’s own research said candidates cared about most.

We ran two sequence variants in parallel, one for engineering and technical candidates and one for go-to-market, with the same structure but different proof points. Both opened by naming the source: the profile came up through Zeki’s dataset, the same data cited in the Stanford AI Index. For a skeptical technical audience, that line did work no subject-line gimmick could.

Sequencing the three channels

Each candidate moved through a fixed order rather than three campaigns running independently:

• Email first. Four waves, spaced two to three days apart. Wave one made the invitation, wave two clarified that this wasn’t a job board, and waves three and four narrowed the window.
• LinkedIn second. Connection requests sent with no note, then a first message four hours after acceptance and three follow-up waves. Outreach ran from 3–4 Zeki executive profiles.
• Calling third. Reserved for contacts who had shown a signal, such as an open, a click, or an accepted connection, rather than dialed cold. One sales development representative (SDR) ran calling full time.

Non-responders were not retired. Email leads were recycled into fresh waves every few weeks with new angles and subject lines.

Phase two: running the same campaign to target the hiring side

Barely two weeks into the candidate campaign, and still weeks out from launch, Zeki’s executive team asked whether Belkins could run the other side of the marketplace in parallel. The early signal was strong enough that they wanted to move on it immediately.

A talent market needs companies as well as candidates, and Zeki had a vetted list of roughly 4,000 funded scale-ups, Seed to Series C with 50-plus headcount, most of them US and West Coast.

The channels and sequencing stayed the same. We only rebuilt the messaging around three arguments:

• The data. No other recruiter has this depth on candidates who are not looking for a job.
• The speed. A shortlist in 72 hours, not weeks.
• The cost structure. Zeki charges 15% of first-year compensation, and only after a successful hire. According to the client’s research, a traditional search firm charges 25–30% and bills it in three parts: on signing, on shortlist, and on hire. A company pays 2/3 of the fee before it sees a single candidate.

Belkins also built objection handling around Spotlight, Zeki’s feature that sends companies a few candidate profiles each week whether or not a role is open. It answers the most common reply on this side of the market: “We’re not hiring right now.”

Companies still converted more slowly than candidates. In mid-July, the ask changed from self-registration to a discovery call with Zeki’s team, which improved the conversations without lifting the volume.

By August 2026, the pattern was clear enough to act on: The two sides of this marketplace buy differently. Candidates respond to a personal invitation at scale, while companies work on a slower, more consultative cycle that cold outreach alone does not fit. Rather than keep spending on a channel the data didn’t support, we paused the company-focused campaigns and proceeded to reach out to candidates.

The six weeks were not wasted. We handed Zeki a set of 35 engaged companies for its sales team to work, and five converted into real registrations.

Challenges Belkins encountered and solved

Voluminous outreach to personal inboxes. Cold B2B outreach to corporate addresses is well-trodden ground for Belkins; 80,000 messages a week to personal Gmail and Outlook accounts is not. Deliverability had to be run as a live operation rather than a setup task, with 20+ domains, continuous mailbox rotation, and daily monitoring. The 0.09% email conversion in month one reflects how much of the first weeks went into finding a sending pattern that would hold.

An offer with no obvious hook. Every message had to sell exclusivity and privacy to people who receive recruiter outreach constantly and have learned to ignore it. The messaging that eventually worked didn’t describe a job at all. It described who else was on the list and what stayed private.

LinkedIn connection rates that outran conversion. Acceptance rates were very high, but conversion from accepted connection to registration ran around 20%. Some of that is structural, since people accept connections to grow a network rather than because they read the message. The answer was more waves rather than better opening messages: LinkedIn ran four waves against email’s two in the early weeks, and finished month two as the highest-converting channel on the account.

A deadline that never moved. The launch date was fixed in a funding commitment, so there was no version of this project where the team traded scope for time. Reporting ran daily, strategy shifted weekly, and messaging was rewritten in days.

“We’re aware we’re a client asking a lot in a short period with an enormous amount of reporting. I have real respect for how responsive the team has been to feedback.” — Lara, Marketing lead, Zeki Data, mid-campaign

What outbound at this volume actually requires

Sending 80,000 cold emails a week is not simply a larger version of sending 8,000. Past a certain rate, the campaign stops being about messaging and becomes about operations. To get tangible results fast, four conditions should be met before the volume is survivable, let alone productive:

1. Sending infrastructure built for rotation. More than 20 domains, mailboxes cycled continuously, and more than one sending tool in use, so no single asset ever carries enough volume to draw attention.
2. Deliverability as a live role. Domain health checked daily rather than reported weekly, with volume moved off a domain at the first sign of trouble rather than after a bounce rate confirms it.
3. People who have run this before. Two SDRs covered this account. At this rate, the constraint is judgment — knowing which signal means slow down and which means push, and headcount does not supply it.
4. Signal-based prioritization. Opens, clicks, and accepted connections determined who received a call and who received another email, so the most expensive channel only ever went to contacts who had already shown interest.

What’s next

Zeki Talent Market launched on schedule. The engagement has since settled into a steadier rhythm on the candidate side, with a standing weekly registration benchmark that keeps the platform’s candidate-to-company ratio in balance as more companies come on.

Company acquisition is the open question. Cold outreach answered it clearly enough, so the next phase is testing webinars and podcast placements for brand recognition. Zeki has also signalled interest in extending the partnership to its intelligence products, a separate line with an established client base.

The lesson we’ve learned: when a client arrives with data this good, the hard part is no longer finding the right people. It is reaching a quarter of a million of them in ten weeks without wrecking the sending domain. That is an infrastructure question before it is a messaging one, and it is worth answering in that order.