How Figo's AI works

Last updated: May 26, 2026

Figo is an ATS with two AI surfaces: Cowork, an assistant that helps recruiters work faster, and Enrichment, a conversational profile builder that the candidate reviews and approves before any recruiter sees it. Figo does not score candidates. Figo does not rank candidates. Figo does not make hiring decisions. Figo does not record audio. Your recruiters stay in control; your candidates stay in control; you stay out of high-risk AI classification.

1. The five rules we built around

Each rule states what Figo's AI is prohibited from doing and what it does instead.

1.1 No AI judgment of candidates

Figo's AI does not produce scores, ranks, recommendations, "top N" lists, or quality grades. It does not infer personality, fit, competence, communication style, or protected-class attributes. Instead, the AI surfaces evidence — verbatim transcript quotes, objective counts (e.g. "4 of 5 React concepts mentioned in conversation"), and deterministic match details computed from candidate-stated facts (e.g. "5 of 6 required skills met"; "590 km from the job location"). The recruiter forms the opinion from the raw material.

1.2 No solely automated decisions

Figo does not auto-reject, auto-disqualify, auto-shortlist, or auto-advance candidates. Pipeline stages move only when a recruiter moves them. Where filtering is needed, Figo applies recruiter-defined rules on candidate-stated facts — work authorisation, required skills, location compatibility, etc. The rule is deterministic math, not AI judgment.

1.3 The candidate controls their own profile

After the enrichment conversation, the AI assembles a structured draft from what the candidate said. The candidate reviews each extracted item, edits anything inaccurate, and either approves the draft for the recruiter or rejects it (purging the draft). Drafts that aren't approved within seven days auto-expire and are deleted. The approval action itself is logged as consent evidence.

1.4 No voice recording

The enrichment conversation runs as real-time speech-to-text. The AI listens, transcribes, and immediately discards the audio frame by frame. No audio file is ever stored. Only the transcript text is persisted. This is the same regulatory category as voice dictation tools — not voice authentication or biometric systems.

1.5 Honest marketing

We never describe Figo as "AI screening", "AI hiring", "AI that decides", or "AI that ranks". We do not claim compliance certifications we have not earned. The language we use in our marketing, sales calls, and contracts is the same language we use inside the product.

2. What Figo's AI actually does

Cowork — chat, voice, and hand-mode AI assistants embedded in the recruiter UI. They draft, summarise, and take actions on the recruiter's behalf, on the recruiter's instruction.
Resume parsing — converts a CV into structured fields the recruiter can read at a glance.
Enrichment conversation — a structured AI-led conversation with the candidate, following a recruiter-configured agenda, that surfaces context a CV can't show. The candidate's stated facts become a draft profile the candidate reviews.
STAR-structured story extraction — when a candidate describes an experience using Situation-Task-Action- Result, Figo extracts the structure for easier review. It does not grade the story.
Deterministic match details — count-based comparisons of candidate-stated facts against role requirements. Numbers, not opinions.
Audit trail — every AI extraction, every recruiter action, every candidate approval, captured in an append-only log.

3. What Figo's AI explicitly does not do

Score, rank, grade, or evaluate candidates
Recommend "top candidates", "best fit", or "interview these"
Auto-reject, auto-disqualify, or auto-advance candidates
Make hiring decisions of any kind
Record audio of any conversation
Infer protected-class attributes (age, gender, ethnicity, health, religion, sexual orientation, etc.)
Use candidate data to train external AI models

4. The enrichment conversation, end-to-end

Recruiter invites the candidate. Figo prepares a per-candidate enrichment package (agenda items, focus areas, questions to clarify resume gaps) and sends a tokenised invitation email.
Candidate joins in the browser. No app to install. A device-setup screen confirms mic and speaker work.
The AI agent runs the conversation. Real-time speech-to-text, language-model response, text-to-speech. Audio frames are dropped as soon as the transcript is produced. The agent follows the recruiter's agenda and uses tools to mark items complete.
Figo assembles a draft. Deterministic extraction analyzers pull out structured facts (skills, years, location, work authorisation, salary expectations) and STAR-structured stories from the transcript.
The candidate reviews and approves. Per-item edit, approve, or remove. Approve sends the draft to the recruiter; reject discards it entirely. Unapproved drafts auto-expire after seven days.
The recruiter reviews the approved profile. Alongside the recruiter's own notes, deterministic match details, and transcript quotes. The recruiter decides what happens next.

5. Privacy & data residency

Primary storage is in the European Union.
AI inference (the language model, the speech-to-text, the text-to-speech) runs in the European Union.
Voice transport runs in the European Union. No audio recording is enabled on any session.
A small number of operational services (transactional email, error monitoring) run outside the EU under Standard Contractual Clauses. The full vendor list is available with our DPA — see sub-processors.
Every interaction is logged in the customer's append-only historian record. DSAR exports, retention auto-purge, and candidate erasure are first-class platform features.

6. What this means for your AI Act exposure

Figo is engineered so that, when used as documented, deployers do not fall into the EU AI Act's Annex III high-risk classification for recruitment. Because Figo's AI does not score, rank, recommend, or make decisions about candidates, it is not an AI system used for the recruitment or selection of natural persons in the sense intended by Annex III(4). Similarly, because Figo does not record audio and does not extract biometric identifiers, it is outside the scope of BIPA-style voice statutes. Because Figo does not produce automated employment decision tools, deployers in NYC are outside the scope of Local Law 144 (AEDT). Because there is always a human decision in the loop, GDPR Art. 22 (solely-automated decisions) is not triggered.

We do not claim "AI Act compliant" or "bias audited" — those are attestations that require third-party assessment. We do design our product so that the conversation about those frameworks is short.

7. Want to dig deeper?

8. Questions

Email hello@usefigo.com — we'll route security and compliance questions to the right person.