Six AI capabilities run inside EmployeeSight every minute — anomalies, attrition, auto-classification, predictive close, document AI, and natural-language reports. Every decision is explainable. Nothing trains on your data.
Anomaly
OT spike · EMP-014
Attrition
Risk 0.71 · EMP-008
Payroll
Closes 30 May · OK
Auto-class
Expense → travel
Six capabilities · always on
Each capability runs continuously, surfaces findings into your dashboard, and explains itself. None require setup. All are toggle-able per workspace. The product works fully without any of them — AI is added autonomy, not a dependency.
01 · Anomaly detection
Continuously scans payroll, attendance, and productivity for patterns that don't match the workspace baseline. Flags two per week for a 50-person team — not 200.
Anomalies · last 30 days · 2 flagged
02 · Attrition prediction
Per-employee attrition risk score with the top three signals that contributed. Surfaces 4-6 weeks before resignation, with enough lead time for an HR conversation.
Attrition risk · per employee · explained
“Manager 1:1 cadence dropped” — top signal for EMP-008
03 · Auto-classification
OLA receipt? Travel. Zoom invoice? SaaS. Stand-up calendar block? Engineering. The AI applies the same labels your finance team would, before HR even opens the file.
Auto-classified · 3 expenses · 0 manual
04 · Predictive payroll close
Models historical close patterns against current-cycle progress. Predicts your close date and surfaces what's blocking it — missing attendance approvals, pending tax sheets, unverified bank accounts — before the 30th.
Payroll close prediction · May 2026
On track · 0 blockers · confidence 0.94
05 · Document AI
Upload an Indian KYC document, get structured fields back. PAN, Aadhaar, bank, Form 16, salary slips, offer letters — all the document types Indian HR actually handles, none of the generic OCR.
PAN.pdf
Doc AI · 1.2s · 0 manual entry
06 · Natural-language reports
Type 'overtime by team for May' or 'attrition risk by tenure'. The AI generates the SQL, runs it against your workspace data, and renders the chart. Saved reports become a one-click dashboard.
Ask in English · get a chart
Honest about scope
Salary calculation
Payroll math is statutory law. The Income-tax Act says what it says. We use deterministic rules with citations — not a model that might hallucinate ₹40 on someone's PF.
Compliance decisions
ESI applicability, PT slabs, gratuity eligibility — all hard rules in our codebase, version-controlled per statute. The AI flags edge cases but never decides them.
Hiring or firing recommendations
We don’t score candidates and we don’t recommend terminations. Attrition prediction surfaces a risk score for an HR conversation — not a verdict on someone’s job.
Common questions about the AI
No. We don't fine-tune models on customer-identifiable data. Anomaly detection and attrition prediction run on per-tenant baselines computed in your workspace; document AI uses pre-trained extraction models and never sends fields to third-party LLMs. Our DPA at /legal/dpa makes this a contractual obligation, not just a policy.
Document extraction uses our own fine-tuned vision models for PAN, Aadhaar, bank, and Form 16 layouts. Anomaly detection uses lightweight statistical models (z-score with seasonal decomposition) and a gradient-boosted classifier per workspace. Attrition prediction uses a transformer trained on de-identified aggregate signals — never on PII. Natural-language reports use an LLM via API with no data retention.
Yes. Every AI decision has an explanation field — the top contributing signals, the threshold crossed, and a per-decision audit-log entry. HR admins can export the full decision trail for any flagged item, which is what makes the system defensible under DPDP's right-to-explanation provisions.
Every flag is a recommendation, never an action. HR makes the call. Wrong flags can be marked 'false positive' in one click, which adds a counter-signal to the per-workspace model. Over time, your AI becomes more accurate for your team specifically.
Yes. Per-tenant data residency in Mumbai (AWS ap-south-1). No cross-tenant training. Right-to-explanation surfaced in-product. Data retention configurable per workspace. Our DPA at /legal/dpa documents every obligation; security posture at /security covers the controls.
Yes, per capability. AI is a feature flag at the workspace level, not the whole system. Many teams start with anomaly detection on, attrition prediction off, and gradually expand. The product works fully without any AI — it's added autonomy, not a dependency.
14-day beta access · No card required · Workspace ready in 1 business day