Methodology
How Kynoa builds its workforce data.
Where our data comes from, how often it refreshes, how we protect personal data, and how we measure the accuracy of our attrition model.
Data sources
Kynoa combines several source types and checks them against each other. No single source is trusted on its own.
- EPFO records: formal-employment additions and provident fund contributions, used to validate employment and pay structures.
- Job postings: roles, skills and salary ranges from more than 12 job platforms, normalised for role, level and city.
- Consented professional data: anonymised profile and offer information shared by professionals who use our resume tools.
- Employee-reported pay: public salary sites, used only after cross-checking with other sources.
- Public filings: MCA and stock-exchange disclosures for executive pay and ESOP pools.
- Economic indicators: RBI policy, Ministry of Statistics releases and venture funding data.
Refresh cadence
- Job postings: daily
- Compensation benchmarks: weekly
- EPFO data: monthly, as published
- RBI policy: as announced
- City intelligence scores: weekly
Privacy and anonymisation
Personal identifiers are replaced with salted one-way hashes, and records are grouped so that no published result describes fewer than 50 people. Personal data is stored in India on AWS Mumbai (ap-south-1). See trust and security for the full pipeline.
How we measure the attrition model
Attrition is imbalanced: in most months, most people stay. Raw "accuracy" would look high even for a model that predicts nobody leaves, so we report metrics that reflect real predictive power:
- AUC-ROC 0.947: how well the model ranks leavers above stayers, measured on held-out data.
- Precision at the top 10% 0.891: of the 10% flagged as highest risk, the share who actually left.
- Recall at the top 10% 0.823: of those who left, the share the top 10% caught.
- False positive rate 5.3%.
Models are validated on data they were not trained on, retrained regularly, and checked monthly for disparate impact across gender, age, tenure and department. Read more about employee attrition prediction.
Example data on this website
Charts and tables marked "Illustrative data" (currently dated Sep 2026) show what Kynoa's outputs look like. They are examples and must not be used for decisions. Customers see live data in the platform.
Framework status
- DPDP Act 2023: Designed for. Consent, purpose limitation, erasure and grievance redressal built in.
- SOC 2 Type II: Audit-ready controls. Controls are in place; the independent audit report has not been issued yet.
- ISO 27001: In progress. ISMS documentation underway.
- GDPR: Principles applied. Applied where EU data subjects are involved.