HiringBranch

Company overview

HiringBranch is a Montreal, Canada based skills-assessment vendor that began as the soft-skills training company LearningBranch in 2014 and rebranded to HiringBranch in 2022 after building a proprietary Small Language Model on more than a decade of training data. Co-founded by CEO Stephane Rivard and President Patricia Macleod, it sells a 'Soft Skills AI' that scores communication, empathy, teamwork and language proficiency from open-ended job-simulation responses, claiming up to 98% scoring accuracy and roughly 1% mis-hire rates. It targets high-volume, customer-facing hiring (contact centers, retail, banking, telecom) for enterprise customers such as Bell Canada, integrates with 15+ HR/ATS systems (Workday, SAP, ADP, Oracle, UKG), and raised a CAD 5M Series A led by Credit Mutuel Equity.

Regulatory exposure

The product is squarely high-risk under EU AI Act Annex III (recruitment and candidate evaluation), an Automated Employment Decision Tool under NYC Local Law 144, and within scope of Illinois and Colorado AI-in-hiring rules; its voice/acoustic analysis of candidate responses also raises heightened data-governance and biometric-adjacent scrutiny. HiringBranch is comparatively strong on the bias-audit dimension: it publishes a public BABL AI audit summary with gender impact ratios and states it satisfies LL 144, and it deliberately excludes demographic inputs (gender, race, native language, orientation, location). Its main gaps are EU-facing: it references EU AI Act high-risk alignment and a NIST/Canada-AIA-based governance framework but provides no Article 27 FRIA template or explicit deployer-obligation guidance, and no ISO 42001, model card, or downloadable technical pack.

Path to a higher score

To raise its score, HiringBranch should make the full BABL AI report downloadable and commit to a visible annual (multi-year) audit cadence, and publish a model card / instructions-for-use and formal explainability statement alongside an ISO 42001 certification. It should add concrete EU AI Act deployer materials (a FRIA template and Article 26/27 obligation guidance) and a DPA, subprocessor list and data-retention schedule to complement its SOC 2 Type II. Finally, standing up a public status/uptime page, a model-update changelog, and a security.txt / responsible-disclosure channel would turn its claimed 'continuous adverse-impact monitoring' into verifiable post-market monitoring evidence.