Skip to content

AI

AI in Elevator

More time for the human part.

AI-native means intelligence is part of the work: understanding a résumé, preparing an interview and finding an answer. Elevator by Superset brings that support into your team’s daily workflow, with evidence to review and people in control.

Request a demo See the lineup

Three modes

AI drafts. Rules route. People decide.

Clear responsibilities make automation useful. See what AI prepares, what your policies route and which decisions belong to your team.

  • AI

    AI prepares evidence for review.

    AI helps interpret candidate information and prepare scores, summaries and interview questions. Recruiters can review the supporting evidence before deciding on the next step.

  • Rule-based

    Rules keep the process consistent.

    Your policies define stage transitions, approval routes and reminders. Each rule has a clear trigger and action, with a history your team can review.

  • Human decides

    People own consequential decisions.

    Hiring decisions, offer approvals and access permissions remain accountable to named people. The record shows who acted and when.

The lineup

Intelligence throughout the hiring journey.

AI capabilities and workflow automation support your team where the work happens, from candidate intake to the next hiring review.

  • AVA screening

    An asynchronous conversational interview as a pipeline stage. Returns a summary card with per-competency signal and a recommended action a recruiter confirms.

    Screening and assessments

  • TalentLens scoring

    Ranks every candidate against the requisition on six weighted axes, with a written explanation quoting the evidence behind each score.

    Screening and assessments

  • Identity and fraud checks

    Photo-match, document verification and duplicate detection aimed at fabricated experience and proxy interviews. Runs quietly and continuously.

    Offers and verified hire

  • Résumé intake with confidence

    Any format parsed into structured, searchable fields, each with a confidence level, so a recruiter fixes a doubtful field in seconds instead of re-reading the file.

    Sourcing and talent CRM

  • Scheduling automation

    Places interviews on real panel capacity and load, and names every conflict it cannot resolve for a person to settle.

    Interviews and scheduling

  • Pipeline automation rules

    Trigger and action pairs written by recruiting operations: nudge, escalate, tag, advance, request approval, auto-reject with a recruiter’s name on it.

    Applicant tracking

  • Interview co-pilot

    Question sets generated from the résumé and the role, live in the room, so the panel probes the gaps rather than the summary.

    Interviews and scheduling

  • Careers-site assistant

    Finds roles, answers process questions and takes the application in one conversation, using the same fit logic recruiters see.

    Sourcing and talent CRM

  • Ask-anything search

    Search, navigate and ask in one ranked list. Answers are computed from the records and cite the ones they used.

    Analytics and reporting

TalentLens

Understand the experience behind the score.

TalentLens compares each candidate with the role across six weighted criteria, including domain expertise, seniority and availability. Written explanations show the evidence behind each score, helping recruiters build a shortlist they can discuss with the hiring manager.

What’s included
  • Every score carries a written explanation quoting the résumé lines behind it
  • Reopen a requisition and rank every past candidate against it in one action
  • What happens at a score of 40 is a written, versioned rule, never the model’s decision
Explore screening and assessments

AVA

Give candidates room to tell their story.

AVA lets candidates complete a conversational AI screen in their own time. Recruiters receive a competency summary, communication score and integrity checks, together with a recommended next step for their review.

What’s included
  • Runs as a pipeline stage, so it can be placed, moved or left out per template
  • Five competencies for operations roles: scenario judgement, customer empathy, process adherence, communication, role knowledge
  • A recommended action, not a decision: the recruiter’s confirmation is what moves the candidate
See the high-volume pipeline

Identity and fraud

Review identity concerns early.

Photo matching, document verification and duplicate checks help identify possible proxy interviews or fabricated experience. Concerns appear on the candidate record so an authorised person can review the evidence and decide what happens next.

What’s included
  • Photo-match across the application, the interview and the identity document
  • Document verification and duplicate-identity detection with fuzzy matching
  • A discrepancy is a state on the offer, reviewed by a person, never an automatic rejection
Explore offers and verified hire

Boundaries

Your policies. Your decisions.

Approval requirements, decision thresholds and access permissions remain under your organisation’s control.

  • Approval routing

    “Who must approve a senior requisition is written policy, not judgement. A model here would be a liability.”

    The approval chain on a requisition or an offer is a rule: level, role, approver, SLA. It is never inferred.

  • Thresholds

    “The model produces the score; a written, versioned rule decides what happens at 40.”

    TalentLens and the assessments produce numbers. What a number means for a candidate is a rule recruiting operations wrote and can replay.

  • Rejections

    “A rejection is a consequence for a real person. It carries a recruiter’s name.”

    AVA recommends advance, review or reject. A recruiter confirms. An automation rule that auto-rejects below a floor does so in a named recruiter’s name.

  • Access control

    “Access control is the one place where a probabilistic answer is simply a breach.”

    Who can see what is a tri-state permission grid across seven roles and 43 capabilities, enforced at query level. No model is consulted.

Trust

A draft with an explanation, and a person’s name on the decision

The AI in Elevator is accountable the same way the rest of the product is: through the audit log, the permission grid and the retention schedule. The security page covers how the data behind it is protected.

Security and trust
  • Candidate data is never used to train a model
  • Every AI output is a draft, and every draft carries a written explanation
  • Every consequential action carries a named person, recorded in the append-only audit log
  • Three of TalentLens’s six axes are arithmetic, not model output
  • No campus signals in TalentLens: no grades, no institution tier, no internships
  • AI steps are pipeline stages and rules an administrator enables per template

Explore the rest of the platform

AI in Elevator, answered

What is AI recruiting software?
AI recruiting software uses machine learning and language models to do work in a hiring process that previously needed a person to read, rank, schedule or summarise. In Elevator that means TalentLens résumé scoring, AVA conversational screening, fraud and identity checks, résumé parsing, scheduling and pipeline automation, an interview co-pilot, a careers-site assistant and ask-anything search. The governing rule is that AI drafts, rules route and people decide.
Does Elevator’s AI reject candidates?
No. A model in Elevator can score, rank, summarise and recommend, but a rejection is committed by a person or by a written rule that carries a recruiter’s name. AVA returns a recommended action of advance, review or reject that a recruiter confirms, and an automation rule that auto-rejects below an assessment floor is a deterministic rule recruiting operations wrote, versioned and can replay from the audit log.
Is candidate data used to train models?
No. Candidate data in Elevator is never used to train a model. It is processed to score, screen and verify the candidate for the roles they applied to, on the customer’s instruction, and it is subject to the retention schedule and erasure queue described on the security page.
How does TalentLens score a candidate?
TalentLens scores a candidate against the requisition on six weighted axes: domain depth, scope and seniority, role continuity, compensation fit, availability and evidence provenance, with the weights summing to 100. The scoring is done against the requisition’s own band, grade and target date, and the rubric is chosen by title rather than grade. Three of the six axes are arithmetic rather than model output, and every score carries a written explanation quoting the evidence.
Is the AI explainable?
Yes. Every AI output in Elevator is a draft with an explanation attached: TalentLens quotes the résumé lines behind each axis, AVA returns per-competency signal rather than a verdict, parsed résumé fields carry a confidence level, scheduling names each conflict it could not resolve, and ask-anything answers cite the records they were computed from. Rules, being deterministic, are explainable by definition.
Can we turn AI features off?
Yes, at the level each feature lives. Each AI step is a pipeline stage or an automation rule that an administrator enables per pipeline template, so a template can run without an AVA stage or without TalentLens ranking and another can include them. Scoring, screening and automation are configured, not imposed.
Which roles does AVA screen?
AVA runs as an asynchronous screening stage after a candidate clears the skill assessment, and is built into the high-volume support and operations pipeline template. For operations roles it assesses five competencies: scenario judgement, customer empathy, process adherence, communication and role knowledge. The candidate sits it on their own time and the recruiter receives a summary card, not a transcript.

Your next team starts here

Your hiring.
All together.

A walkthrough of Elevator, shaped around your roles, your people and your process.

Request a demo Explore the platform