How We Broke Our Metric:
What 0.93 Candidate Relevancy Means for AI Sourcing
It’s one thing to make a metric move. It’s another to make it truly matter.
When we built Relevancy (we spoke about it in our previous material Candidate Relevancy:
The Metric AI Sourcing Was Missing), our goal wasn’t just to have a number that changes.
We wanted a metric whose growth actually correlates with user satisfaction — with better results, happier clients, and real measurable value.
That turned out to be one of the hardest parts:
how to align our internal instruments with real-world usefulness — how to make our candidate relevancy benchmark mean something tangible for users.
We spent dozens, if not hundreds, of hours talking to our clients — recruiters, founders, product teams.
We listened to how they search, what they notice, what makes them pause or reject a candidate — and we tried to decode the intuition they never verbalized and often weren’t even conscious of.
We gathered those insights piece by piece, and then turned them into algorithms, heuristics, prompts, and ranking models.
All so that the metric we chase inside the company would translate directly into real-world candidate sourcing quality — the kind that can be felt, not just measured.
When we built Relevancy (we spoke about it in our previous material Candidate Relevancy:
The Metric AI Sourcing Was Missing), our goal wasn’t just to have a number that changes.
We wanted a metric whose growth actually correlates with user satisfaction — with better results, happier clients, and real measurable value.
That turned out to be one of the hardest parts:
how to align our internal instruments with real-world usefulness — how to make our candidate relevancy benchmark mean something tangible for users.
We spent dozens, if not hundreds, of hours talking to our clients — recruiters, founders, product teams.
We listened to how they search, what they notice, what makes them pause or reject a candidate — and we tried to decode the intuition they never verbalized and often weren’t even conscious of.
We gathered those insights piece by piece, and then turned them into algorithms, heuristics, prompts, and ranking models.
All so that the metric we chase inside the company would translate directly into real-world candidate sourcing quality — the kind that can be felt, not just measured.
How the numbers evolved
After a year of pursuing our core metric relentlessly, we improved it by more than 2x — from 0.43 to 0.93 – In this article, you could find another way of interpreting Relevancy while conducting a plain and simplistic research for paper.
That’s not an incremental step. That’s a leap.
That’s not an incremental step. That’s a leap.
Candidate Relevancy score: 0.93 — measured on top results.
That’s 93% of candidates in the top results judged as strong matches by independent recruiters — real recruiter rated relevance.
At this point, we’re outperforming most human junior sourcers — consistently, across thousands of searches.
It didn’t happen overnight.
At this point, we’re outperforming most human junior sourcers — consistently, across thousands of searches.
It didn’t happen overnight.
At this point, we’re outperforming most human junior sourcers — consistently, at scale, across hundreds of millions of candidates.
It didn’t happen overnight.
It didn’t happen overnight.
Each step came from hundreds of micro-decisions — refining ranking logic, adding context models, retraining evaluation sets, improving ai sourcing precision and ai sourcing accuracy through real-world feedback loops.
And all this, while running search on hundreds of millions of profiles in tens of seconds.
This isn’t just about building an algorithm.
It’s about building infrastructure — one that allows hundreds and thousands of clients to find the best candidates on the market among hundreds of millions of profiles, in just seconds.
Just think about that scale — and the candidate ranking quality it takes to make it work.
And all this, while running search on hundreds of millions of profiles in tens of seconds.
This isn’t just about building an algorithm.
It’s about building infrastructure — one that allows hundreds and thousands of clients to find the best candidates on the market among hundreds of millions of profiles, in just seconds.
Just think about that scale — and the candidate ranking quality it takes to make it work.
What it means for us
These numbers aren’t just internal milestones.
They mean we’re setting goals that are hard, measurable, and deeply tied to user value.
We can see it clearly — in client feedback, in demos, in sales momentum.
When our Relevancy goes up, so does satisfaction, conversion, and retention.
And that’s what recruiting search metrics are for — they should predict outcomes that matter.
They mean we’re setting goals that are hard, measurable, and deeply tied to user value.
We can see it clearly — in client feedback, in demos, in sales momentum.
When our Relevancy goes up, so does satisfaction, conversion, and retention.
And that’s what recruiting search metrics are for — they should predict outcomes that matter.
What it means for the industry
When we published the AI Sourcing Benchmark 2025 this spring, it became the first real attempt to measure ai recruiting quality metric and ai sourcing evaluation across platforms.
But benchmarks don’t last forever.
The one we released months ago is already on the edge of becoming outdated.
We’ve crossed the line it defined.
Now it’s time to create new boundaries — higher, tougher, more real.
Benchmarks aren’t there to celebrate the past.
They’re there to define the next frontier.
We’re building that frontier right now — with new datasets, harder queries, broader evaluation, and a next-generation ai recruiting benchmark that goes beyond anything we’ve done before.
We’re creating our own next challenge — and setting the bar the entire industry will chase next.
You can check what this solution consists of right now by exploring our API docs
But benchmarks don’t last forever.
The one we released months ago is already on the edge of becoming outdated.
We’ve crossed the line it defined.
Now it’s time to create new boundaries — higher, tougher, more real.
Benchmarks aren’t there to celebrate the past.
They’re there to define the next frontier.
We’re building that frontier right now — with new datasets, harder queries, broader evaluation, and a next-generation ai recruiting benchmark that goes beyond anything we’ve done before.
We’re creating our own next challenge — and setting the bar the entire industry will chase next.
You can check what this solution consists of right now by exploring our API docs
We hit the ceiling. Now we’re raising the roof.
Continue Exploring