Case Study
How HeyGen automated their hiring pipeline without adding additional sourcers
Company
HeyGen
use case
Recruiting Automation
company overview
HeyGen is a leading artificial intelligence video platform that lets users create realistic digital avatars, voice clones, and translated video content without cameras or studios
3-4 a day | 70-100 | 90 min | 3 |
candidates booked, fully automated | Manual reach-outs per day replaced | saved per fake applicant caught | Candidates in offer within 6 weeks |
Impact TL;DR
Three candidates in offer or executive review within the first six weeks, fully automated
Built a custom recruiting agent that carries the team's hiring judgment into every search
Screens fake and spam applicants out of the inbound before they cost interview time
Scaled sourcing, outreach, and fraud-screening without adding a single hire
Toward a more efficient way to hire
HeyGen creates AI avatars and AI videos, and they compete for engineers against companies many times their size. The company is deliberately lean. Every hire has to clear a standard where the whole team is enthusiastically behind the candidate. So they hire carefully, and they hire slowly.
The best-funded AI companies are always in the press, and that visibility does the top-of-funnel work for them. On the other hand, HeyGen builds their pipeline through candidate outreach and nurture rather than headlines. To fill a standard pipeline role, a new grad or an L4 or an industry engineer, the team had to send hundreds of reach-outs to get back a couple of replies.
The obvious fix for a top-of-funnel problem is to hire a sourcer, or two, and throw manpower at the volume. HeyGen went the other way. Instead of scaling headcount, they set out to make the recruiting team they already had more efficient, giving those recruiters infrastructure they could see into and control.
Before Crustdata
Before that, sourcing meant a person sitting on LinkedIn all day, clicking through a profile fifteen times over to send one reach-out. It was slow, and it did not scale with how many roles the team was trying to nurture at once.
The candidates they were looking for also started getting many times the messages they saw a couple of years ago, which meant every HeyGen message was buried amongst a ton of automated outreach from other companies.
HeyGen had tried the out-of-box SaaS tools. They used Juicebox for a while, which they found unreliable at the time. They also tried out Metaview and Fillmore.
What they set out to build
What Jake wanted was a workflow built around how HeyGen already hires, one where he could see every step and understand what was happening at each, not another tool that aimed at replacing a recruiter.
The team had shared instincts about what makes a candidate worth pursuing, the kind of rails every good recruiter carries in their head. For example, someone who has changed jobs four times in two years is not a good fit. The goal was to get those instincts out of people's heads and into a system that could operate based on this judgement at scale.
What changed using Crustdata
Jake did not come to Crustdata looking for a sourcing tool. He had an agent running on Browserbase, crawling GitHub for repositories relevant to a role he was trying to fill. The crawl surfaced good code, but it hit the same problem every time. There was no easy way to go from a repository's contributors to who those people actually were, where they worked, what their careers looked like. That gap is what led him to Crustdata.
Working with the Crustdata team, Jake turned that dead end into a custom recruiting agent. Crustdata supplied the data layer the GitHub crawl was missing, structured records of people and their full career histories. On top of it, the team wired HeyGen's own hiring instincts and guardrails straight into the agent, so the job-hopper pass and the other rules every recruiter there carries in their head apply on every search without being restated. Jake runs the whole thing in plain language, and never writes code to source a candidate.
What he really has is a digital twin of how HeyGen hires. It holds onto the feedback the team gives it, so every correction makes it more like a seasoned HeyGen recruiter, one who knows all the guardrails and has the right instincts. Over time it reads a profile, judges a candidate, and writes an outreach the way they would.
The same judgment is applied in outreach too. Most automated reach-outs are lazy and use a fill-in variable - like the past company name or their school of education dropped into a template.
Using the data from the full profile Crustdata returns, the agent writes outreach that reads like someone actually read the entire profile of a person. It might write about a videography club from school, a fair thing to mention at a company that builds avatars, or a piece of video-compression work that is directly relevant to what one of HeyGen's teams is building.
It only works because three things are present at once. Crustdata supplies live, structured data on people and their careers. Claude reasons over it in plain language. And HeyGen decides what a good candidate is, then encodes that decision into the agent. Take away the data and the agent is back to crawling and guessing. Take away the encoded judgment and it is back to a generic tool that surfaces the same candidates for everyone using it.
Beyond sourcing
What made Jake a champion is that the value did not stop at sourcing. He sees sourcing as laborious rather than hard, the kind of work nobody wants to do. The more interesting discovery was that the same people data filling the top of the funnel could solve a problem at the other end of it.
Roughly one in ten candidates who apply to HeyGen is a fake or a spam profile. A fraudulent applicant rarely gets far, but the cost of even one fake interview is high.
So, HeyGen uses Crustdata for their inbound recruitment too. Applicants are parsed into a database and matched against Crustdata, then sorted into verified, unverifiable, and highly suspicious. The suspicious profiles get deprioritized before a human opens them, which clears the obvious red flags off a reviewer's desk at the start of the day.
The same enrichment layer feeds sales lead-generation on top of that, from one source of truth on who people are.
The bottom line
Jake has been running Crustdata for about a month and a half, and most of the pipeline now runs on its own.
The candidates coming through hold up in review. HeyGen has an offer out to one candidate and two more in executive review, close enough to meet the CEO or CTO.
Without Crustdata, Jake estimates HeyGen would have to add sourcing headcount to maintain the current pipeline volume, a dedicated full-time sourcer and possibly a second, or watch their evergreen pipeline drop toward zero.
What HeyGen built is deeper than a sourcing tool. The same digital twin that sources candidates also writes the outreach, screens fraud out of the inbound, and feeds sales, one system carrying the team's judgment across the whole recruiting function.
For a company built on multiplying what each person can do instead of adding more of them, that data layer is the whole return. One small team now runs the sourcing, the outreach, and the fraud-screening at once, work that used to need several more people.
That is what Crustdata really does for HeyGen. It doesn't just give them a faster sourcer, it lets their recruiting team punch well above their headcount.