How we use AI — and the guardrails around it
AI powers our assessments, scoring, and interviews. Here is where it fits, what keeps it consistent, and what we do and do not claim about it.
For the detailed position on fairness, adverse impact and data handling, see fairness and defensibility.
Where AI fits in
AI plays three roles in Kaairo — all designed to help your team evaluate talent faster and more consistently.
Assessment creation
AI generates role-specific test content — case studies, scenarios, and questions. Your team reviews and edits everything before it goes live.
Scoring
Candidate responses are scored by AI against structured rubrics. Scoring runs at a zero-randomness setting, and the rubric, the questions and their order are fixed before anyone applies.
Voice interviews
Kaaira, our AI interviewer, conducts real-time voice conversations. Questions adapt to the candidate’s responses. The transcript is scored after the interview.
What makes it fair
Structured rubrics, not gut feelings
Each test type has its own structured rubric. Case studies evaluate problem-solving depth, voice interviews are scored against the competencies you configure for each role, and SJTs and MCQs use deterministic rule-based scoring. All scores normalise to a 0–100 scale.
Blind to who’s answering
The scoring model receives only the candidate’s response text. No name, gender, age, or background information is included.
Consistent across every candidate
Scoring runs with randomness switched off, so results are highly reproducible rather than variable. Every candidate for a role faces the same questions in the same order, and automated safeguards flag low-effort responses on the same basis for everyone.
Humans make the decisions
AI produces scores and the evidence behind them; shortlisting and hiring are your team’s calls. If you would rather automate a cut-off, rule-based automatic rejection is available as an option you switch on and set the threshold for — the rule is still your decision.
What candidates see
Candidates are informed at every step of the process.
- Candidates are asked for consent before an AI-scored assessment begins
- Voice interviews include a clear notice that they’re speaking with an AI interviewer and that their responses will be transcribed and scored
- Where the organisation shares results, candidates see their overall result, score breakdown, and per-competency analysis
- If a section could not be scored, candidates are told it was not counted against them rather than shown a zero
How we handle data
Isolated by organisation
Candidate and assessment data is isolated per organisation at the database layer, so one organisation’s users cannot read another’s.
No continuous video
AI camera analysis runs in the candidate’s browser. During camera-proctored assessments we store periodic still snapshots and behavioural flags (e.g., “looked away for 3 seconds”) — never continuous video.
Not used for training
We do not use candidate data to train or improve AI models, and our AI processing runs on business terms that exclude customer data from provider model training.
Access controls
Candidate results are visible only to people the hiring organisation has given access to. Owners and managers control who that is, and can grant read-only access to others on the team.
Frequently asked questions
Does AI make hiring decisions?
No — your team does. AI produces scores and the evidence behind them, and your team reviews, shortlists and hires. Organisations that want a cut-off applied automatically can switch on rule-based automatic rejection and set the threshold themselves; that rule is your decision too, and it is off unless you turn it on.
Can the same answer get different scores?
Objectively scored sections — multiple choice and situational judgement — always give the same answer the same score. AI-graded sections run with randomness switched off and are highly reproducible, but we don’t claim bit-identical output, because we can’t guarantee it. What is fixed for everyone is the rubric, the questions, their order, and the requirement that every score carries the reasoning it was based on.
Is candidate data used to train AI models?
No. We do not use candidate data to train or improve AI models, and our AI processing runs on business terms that exclude customer data from provider model training.
What data is collected from candidates?
Written responses, transcripts, timestamps, and — if camera proctoring is enabled — periodic still snapshots and behavioural flags. No continuous video recording.
Can data be deleted on request?
Yes. Contact the organisation that invited you, or reach out to us directly.
Questions? Get in touch
Want to learn more about how Kaairo uses AI responsibly? We're happy to walk you through it.
Contact Us