Covey

Company overview

Covey (Covey, Inc.) is a San Francisco-based AI recruiting company founded in 2018 that raised roughly $10M (Series A, 2022) and operates as a lean ~20-person team. Its flagship product, Covey Scout, lets recruiters describe in plain language how they would evaluate a candidate, then trains custom generative-AI/LLM bots that score and rank millions of LinkedIn/GitHub/Dribbble profiles for outbound sourcing (Scout Outbound) and screen the top slice of inbound applicants (Scout Inbound), integrating with ATSs such as Greenhouse, Lever, Ashby, and Workday. The product is positioned as transparent decision-support — Covey markets 'full visibility of the selection process' with 'no AI algorithm black box' and lets recruiters validate and fine-tune each bot — placing it firmly in the sourcing category with a strong secondary screening function.

Regulatory exposure

Because Covey Scout substantially assists or replaces human screening of candidates, it is an Automated Employment Decision Tool under NYC Local Law 144 and a likely high-risk system under the EU AI Act (Annex III employment use). Covey has taken the LL 144 obligation seriously: a public crowd-sourced ACLU tracker documents two consecutive independent bias audits — ConductorAI dated 8/10/2023 (later removed from the site) and Idiro Analytics dated 7/29/2024 (posted at getcovey.com/nyc-local-law-144) — which is more than most peers publish. On data/security it has a credible GDPR posture (DPA with SCCs, documented Privacy Impact Assessments, AES-256 encryption, role-based training, a cross-functional Security Council). Where it is thin is the EU-facing and forward-looking obligations: there is no published Article 11 technical/model documentation or ISO 42001, no FRIA or deployer-obligation guidance, and no public post-market monitoring surface (status page, changelog, or security-disclosure channel).

Path to a higher score

Covey's clearest win is converting its two-year audit history into a durable, easy-to-find bias-transparency hub: keep each year's audit publicly downloadable (don't remove prior years, as happened with the 2023 ConductorAI report), and publish the full selection-rate/impact-ratio tables with sample sizes and intersectional categories. Next, publish a system/model card and instructions-for-use describing how Scout bots are built, what data is and is not used to train models, and explainability of its scoring, and pursue ISO 42001 to lift the Article 11 score. Add explicit EU AI Act deployer guidance (an FRIA template and human-oversight instructions) and per-jurisdiction configuration notes. Finally, stand up a visible post-market surface — a status/security page with a vulnerability-disclosure contact and a model-update changelog — to move post-market monitoring off the floor.