How to use data to hire AI talent better
AI searches are small, so every candidate counts. Here is the data worth tracking, and how to read it when the numbers are small.
The 5 metrics that matter
Track a few metrics well rather than many badly. The five that drive decisions are time-to-hire (job posted to offer accepted), cost per hire (total recruiting spend divided by hires), source quality (which channels produce hires, not just applicants), pass-through rate at each stage, and offer acceptance rate.
Each answers a different question. Time-to-hire shows how fast you move. Cost per hire shows which channels are efficient. Source quality shows which channels produce people you hire. Pass-through rates point to the stage that is not working. Offer acceptance shows whether your pay and process are competitive.
The 5 core recruiting metrics
- ⏱️ Time-to-hire: job posted to offer accepted
- 💰 Cost per hire: total spend divided by hires
- 📡 Source quality: hires by channel
- 📊 Pass-through rate: by stage
- ✅ Offer acceptance rate
Reading a small funnel
A senior ML search may produce a few dozen applicants and a handful of finalists, so percentages swing wildly from one candidate to the next. Look at counts as well as rates, and compare across several searches before drawing conclusions.
Still, patterns show up. If few applicants are qualified, the job description is probably vague about the work or the title does not match what the role does. If many pass the screen but almost nobody passes the technical round, the assessment is likely testing something the role does not need, such as research-level theory for a production engineering job.
Source attribution
Track where your hires come from, not just where applications come from. The channel that sends the most applicants is often not the one that sends the people you hire. A source that brings ten applicants and two hires is worth more than one that brings a hundred applicants and none.
On the Growth plan and above, TensorHunt records how each applicant first reached your job (search engines, social media, job alerts, other sites or direct visits) and shows the breakdown overall and per job. Combine it with outcomes from your own hiring process to see which channels lead to hires.
Questions source attribution should answer
- ✓ Which channels produce candidates who pass the technical round?
- ✓ Which produce accepted offers?
- ✓ Which have the lowest cost per hire?
- ✓ Which produce people who stay?
Diversity metrics
Track representation at each stage of your funnel, not only in final hires. Final numbers alone cannot tell you whether a gap comes from who applies or from what happens during the process. Stage-by-stage data shows where it changes.
In ML hiring, look closely at requirements that narrow the pool without predicting performance: a PhD for roles that do not need research training, or a preference for a few universities or labs. A drop at a particular stage usually points to that stage, such as how a take-home is scored or who sits on a panel, and fixing that stage works better than a general initiative.
Using TensorHunt analytics
Your TensorHunt analytics page shows views, applications and apply rate across your jobs, applications over the last week, how many applicants are at each stage of your pipeline, and your top jobs by applicants and views. Each applicant also has a match score comparing their profile with the job's required skills.
Use the stage breakdown to spot where candidates pile up. If many applicants sit unreviewed, or finalists wait days for the next round, that delay is costing you candidates, and fixing it is usually cheaper than buying more applicants.
What TensorHunt analytics shows you
- ✓ Views, applications and apply rate
- ✓ Applications over the last 7 days
- ✓ Applicants at each pipeline stage
- ✓ Top jobs by applicants and views
- ✓ Where applicants come from (Growth plan and above)