A screen that asks a follow-up
Every candidate gets the same rubric and a real conversation. The interview adapts as it goes — a strong answer earns a deeper probe, a thin one earns a follow-up — and every score cites the words the candidate actually said.
This is one of four parallel screening agents in VectorHire, our own recruiting product.
7
stages from rubric to a scored candidate in your ATS
22
languages our voice product already handles
0
scores without the candidate's own words behind them
Five reasons the first screen is the bottleneck
Screening is the one stage where the cost scales with applicants and the quality scales with how tired the recruiter is.
The screening backlog sets your time-to-hire
Phone screens take a recruiter fifteen to twenty-five minutes each, and a single role can bring dozens of applicants who all need one.
Every recruiter screens differently
Each improvises questions, applies a different rubric and writes free-form notes, so two candidates are never actually compared.
Language coverage bottlenecks the pipeline
Roles that draw candidates across several languages depend on whichever recruiters happen to speak them.
One-way video screens lose people
Pre-recorded assessments ask a candidate to talk into a void, and the strongest candidates are the ones with the option to skip it.
High-volume roles never fit the budget
Seasonal and high-turnover hiring needs thousands of screens a month, which is not a headcount problem anyone can solve by hiring.
From rubric to scored candidate
The rubric comes first, because a screening system with no agreed standard just automates whatever the last recruiter happened to value.
The rubric is defined
The role and a weighted scoring rubric with the evidence expected per criterion — the part that decides whether any of this is fair.
Scheduling
A scheduling agent reads candidate availability from the calendar and books the slot without a recruiter in the loop.
The call is placed
Outbound dial with quality monitored per call; a degraded line triggers a callback rather than a bad recording.
The conversation
Speech-to-speech runs the interview at conversational pace, with fallbacks configured for the cases where it cannot.
Adaptive follow-up
Each response routes through an evaluator: strong answers branch deeper, weak signals trigger probing, and off-topic answers are steered back.
Scoring against the rubric
Scored live against the rubric, with each score tied to a cited utterance rather than assembled from a transcript afterwards.
Report and write-back
A report with evidence citations writes to the ATS, and the recruiter is notified when a candidate clears the threshold.
Six capabilities behind one phone call
Unified speech-to-speech
No transcription hop in the critical path, so turn-taking is fast enough that the conversation feels like one rather than a form being read aloud.
Carrier-grade telephony
Outbound and inbound calls with per-call quality telemetry, and automatic fallback when the line degrades rather than a screen that silently fails.
Adaptive dialogue policy
The interview is a state machine, not a fixed script: senior candidates get deeper probes, and a thin answer triggers a follow-up rather than a shrug.
Evidence-cited scoring
Every scored criterion cites the exact thing the candidate said. The report is defensible because it shows its work, not because it sounds confident.
Many languages natively
Speech-to-speech in the candidate's own language, with the rubric calibrated per language rather than translated and hoped for.
ATS write-back
Scorecards, transcripts and reports push into the applicant tracking system your team already lives in, so nothing needs re-keying.
A score you can defend, and audits you actually run
Every score cites the candidate's own words. A number with no evidence behind it is not a screen, it is a filter nobody can explain.
Screening decisions affect people's livelihoods and sit under real regulation, so the system is built to be examined: the rubric is yours, the evidence is attached to every criterion, the recordings and transcripts are retained under your policy, and adverse-impact audits run on a schedule rather than when somebody asks. The system also escalates what it cannot judge instead of guessing — a candidate who does not fit the rubric's assumptions is a person for a recruiter, not a low score.
Four pipelines, four different rubrics
High-volume recruiting
Retail, hospitality and support pipelines screening thousands of applicants per month per region.
Multilingual screens, availability and reliability scoring, automated reschedule flows for no-shows.
Senior technical screening
Competency-based screens for senior engineering roles before they reach the hiring-manager loop.
System-design probes, depth checks on the languages claimed, and behavioural evidence on systems actually shipped.
Professional services
Associate pipelines where structured behavioural scoring drives the decision.
Case-style follow-ups, communication scoring, and structured evidence for partner review.
Early-career pipelines
University recruiting and apprenticeship programmes with thousands of applicants per cohort.
A lower-anxiety format: candidates talk to something that responds, instead of recording a one-way video.
Shadow-scored before it decides anything
The pilot runs beside your recruiters rather than instead of them, so the first thing you see is where the two disagree.
Week 1
Discovery and rubric design
Your scoring criteria, evidence patterns, calibration examples and ATS configuration captured — with your recruiters, because they own the standard.
Weeks 1–3
Pilot: one role, one language
Real candidate screens run alongside recruiter shadow scoring, and the calibration report shows where the two disagree.
Weeks 4–8
Production rollout
Multiple ATS integrations, a multilingual rubric pass with native-speaker calibration, and production monitoring.
Continuous
Ongoing tuning
Regular adverse-impact audits, rubric-drift detection, call-quality regression monitoring and model upgrades.
Bring a rubric and ten past candidates
This is our AI engineering practice
It is real work and it is where our four products came from. But what Cognilium leads with is narrower: optimization apps that run in tandem with Microsoft Dynamics 365, computing the decisions the ERP records but does not derive — the optimal price, the optimal pick path, the optimal stock level. See the optimization apps · How we build inside the ERP.