ISO 27001 | ISO 42001 AI Governance | Official Zoho Partner
ISO 27001 | ISO 42001 AI Governance | Official Zoho Partner
AI governance matters because it ensures that artificial intelligence systems are developed and used responsibly, transparently, and in compliance with legal and ethical standards.
Under the EU AI Act, organisations deploying AI must assess and manage risks associated with these systems, particularly when they can significantly affect individuals' rights, opportunities, or access to services.
From 2 August 2026, organisations using AI to screen job candidates must inform applicants that they are being pre-screened through an AI-assisted process.
Public sector organisations must evidence:
| Question | Smart Sifty Response |
|---|---|
| How is bias measured? | Smart Sifty excludes personally identifiable information (PII) from candidate scoring. Attributes such as name, age, gender, nationality, email address, mobile number and postcode have no influence on candidate rankings. Our models are independently assessed for bias every month, and customers can access bias assessment reports. |
| How are scores generated? | Candidates are assessed across eight weighted categories aligned to the job requirements. Categories are weighted according to their importance for the role—for example, work history carries a higher weighting than certifications, where appropriate. |
| How do you train your models? | Approximately 97% of Smart Sifty's decision-making is deterministic software logic, providing consistent and repeatable results. AI components operate within predefined rules rather than learning from customer recruitment data. Customer CVs and hiring outcomes are not used to retrain or improve the models. |
| Do you store resumes? | No. Candidate resumes remain within the customer's existing ATS or database. Smart Sifty processes the information without permanently storing candidate CVs. |
| Can recruiters override the AI? | Yes. Recruiters remain in control. Smart Sifty analyses the job description and structures it into editable categories—including mandatory requirements, preferred requirements, languages, responsibilities and benefits—which users can review and modify before candidates are assessed. |
| Is every recommendation logged? | Yes. Every candidate assessment is logged with the candidate's score, assessment details and timestamp, creating a complete audit trail for review and compliance purposes. This is stored in ATS candidate records. |
| Is there model versioning? | Yes. Customers can choose between two assessment models: Aurora and Meridian. Both models undergo monthly bias assessments, with reports available to customers. Feature updates are released quarterly and undergo penetration testing and quality assurance before deployment. |
| How are prompts stored? | Smart Sifty does not use prompt-based workflows or prompt engineering as part of its candidate assessment process. |
