Overview
A recruitment platform was hiring against live Fortune 500 requisitions at Target, Airbnb, Uber, Goldman Sachs and Morgan Stanley, where the bar is set by the client and relevance is the entire product. Paid social was the only motion, and it was the wrong instrument: $10k-20k of job-targeted Meta ads produced signups that were irrelevant to the job in question. Tidalstead reassigned Meta to greenfield candidate acquisition, what it is actually good at, and built an outbound engine that took over targeted relevance. It served 120+ active roles at once, screening 10,000+ candidates a day to contact 500 to 700.
Paid spend that could not see relevance
An ad platform optimises for the click, not for the requisition. Every dollar bought volume at the top and none of it knew what the job actually needed, so the funnel filled with people no recruiter could send.
Enrichment billed before qualification
The standard waterfall enriches a profile and then decides whether it was worth enriching. At Fortune 500 volume that is a credit meter running against candidates who were never a match to begin with.
Relevance at 120+ live roles
Roles open, close and change their bar weekly. Any engine that needed a human to re-target it per requisition would stall at ten roles, let alone a hundred and twenty.
Tidalstead’s Approach
We always start from first principles thinking. The question was never how to send more, it was what has to be true before a single candidate is worth spending money on. So the engine was built as a sequence of gates, each one cheaper than the step behind it, and the expensive instruments were only allowed to fire after the cheap ones had already agreed.
Match before you spend
Candidates were surfaced from a proprietary 300M-profile index and matched to live jobs by Mercurial, an in-house vector-embedding model, before any spend at all. The model carried the relevance the ad platform could not.
Gate the credits behind the match
Lusha enrichment fired only after a match landed. That single inversion is what let 10,000+ daily screens resolve to 500 to 700 contacts: the budget followed qualification instead of hunting for it.
Remove the friction on the candidate’s side
Profiles were pre-created on Talent500, and emails carried magic sign-in links that landed a candidate directly on the skill assessment rather than on a signup form. Five-touch HubSpot sequences were personalised on work history, so the follow-up read as a conversation about their career rather than a broadcast.
Results
- 10,000+ candidates screened daily, with only 500 to 700 contacted
- 120+ active Fortune 500 roles served by one engine
- Roles filled at Target, Airbnb, Uber, Goldman Sachs and Morgan Stanley
- First version built in 5-8 weeks in the pre-AI era; the equivalent build today is about 2 weeks
Conclusion
The engine replaced a spend-first motion with a qualification-first one, and the economics inverted with it: more screened, far less sent, and every credit spent on someone who had already matched. The same architecture is what we build today in roughly a quarter of the time.