People Analytics — Platform Guide

• Connectivity test and schema discovery on every source
• Per-field sensitivity classification
• Manual or scheduled refresh — fixed interval or cron expression with configurable timezone
• Freshness monitoring and a certified historical coverage window

• Canonical model — 30 entities: core workforce (person, employment, organization, position) + optional domains (absences, compensation, performance, training, recruiting, engagement, planning, benefits, time, skills, talent/succession, employee relations).
• Governed domain extensions — organization-specific attributes (x_*) are declared in an administered registry with data type and allowed values; they become queryable only after declaration and stay traceable.
• Visual mapping + governed SQL mode: each source field gets a sensitivity classification; transformations are declarative and safe — no free-form code.
• Fan-out — one source row can produce multiple records (skill lists, repeated columns like salary_2023), with controlled limits.
• Quality: per-field rules, sample preview, versioned mappings with publish/rollback.
• Quarantine — records violating critical quality rules are diverted to a separate zone instead of blocking the whole job, with a documented reason and reprocessing or manual resolution by authorized operators.
• Full lineage — every Gold record traces back to the source, job and mapping version that produced it.

Governed metrics

• Core: headcount, FTE, hires/separations, turnover, average daily headcount — plus custom metrics built in the UI on the semantic model.
• Event-driven: metrics on the event ledger — external hires, re-hires, organizational transfers, exits, internal mobility, net headcount change, early retention/attrition, median tenure, contractor share, point-in-time and average FTE. Events (hire / rehire / transfer / termination) are derived from employment period history, not source labels — an internal move never shows as a hire/exit.
• Module metrics — activated with each license: absences (days, rate, mean duration), recruiting (applications, offer acceptance), training (completion rate, hours), performance (average rating), engagement (mean score), compensation (mean variation), planning (plan vs. actual variance).
• Organizational health: hierarchy depth, average span of control, promotion rate.
• Historical coverage: each source has a certified coverage window — metrics outside it show “unavailable”, never a misleading zero.
• Small-cohort protection: values computed on groups below the configured privacy threshold are automatically suppressed (including complementary suppression on adjacent groups) with a visual indicator — contributor counts are never shown. Sensitive domains (pay, absence, surveys, performance) have dedicated thresholds.

Dashboards, reports & executive briefing

• Interactive dashboards with drag-and-drop editor, deployment-wide validated themes for accessibility, controlled sharing.
• Executive Briefing — a board-ready page with governed KPI tiles (headcount, flow, retention, mobility, FTE, org health), month-over-month / year-over-year comparisons, visible states for suppressed or unavailable values, and one-click PDF/PPTX export — honoring the same scopes, cohort thresholds and licenses as the rest of the platform.
• Scheduled reports delivered over email (corporate SMTP) or webhook, PDF/PPTX export.
• Data endpoints — controlled exposure of governed data over API with dedicated tokens.

Alerts & notifications

Rules on metrics and data-quality events; in-app notifications; delivery over configured channels.

AI & ML

• Conversational assistant for questions on governed data, with usage budgets and automatic PII redaction before anything reaches the language model (emails, phone numbers, national identifiers).
• AI Agents — defined with a restricted set of approved tools (mapping suggestions, metric explanations, dashboard drafts).
• ML Studio — curated models deployed locally (regressions, clustering) over governed features; zero external dependencies. Model registry with a draft → approved → retired lifecycle, model cards (intended use, limitations, evaluation metrics) and dual approval — the author cannot approve their own model.

Data catalog & quality

Data dictionary with classifications, source profiling, canonical quality rules (with quarantine for diverted artefacts) and cryptographic integrity checks on the audit chain.

Operational monitoring

• System Health console — configuration, secrets, database, jobs and audit checks (pass/warn/fail).
• Operational metrics endpoint in standard format (Prometheus): API traffic, job queue & duration, lake size, triggered alerts.
• Job correlation tracing for fast diagnosis.

• Scopes on organizations and, optionally, on persons (direct/recursive reports).
• Authentication via corporate IdP (OIDC); group → role mapping.
• Break-glass access — time-boxed role elevation with mandatory reason and separate approval; grants expire automatically, are revocable and fully audited.
• Every query is re-authorized server-side for the current viewer — UI visibility never replaces authorization.

• Data never leaves your infrastructure; no telemetry, no “phone home”.
• Mandatory sensitivity classification on every source field; special-category fields cannot be mapped into analytics.
• Automatic small-cohort suppression — aggregates below the configured threshold are never shown; the threshold applies consistently across metrics, dashboards and visualizations.
• Data subject rights (DSR): controlled subject lookup (raw identifiers are never stored — only pseudonymous forms), right-to-erasure with an approval step and propagation through every data layer including quarantine, with an auditable proof-of-deletion (tombstone).
• Governed retention: configurable retention policies per data layer (including indefinite retention), scheduled sweeps honoring legal hold — an active hold blocks deletions on its perimeter.
• Identities across different sources stay separate by default — linking records belonging to the same person happens only via an audited governed decision (approve/reject/reassign), never automatically.
• The AI assistant honors the same scopes and masks as normal queries; personal data typed into conversations is redacted before reaching the model.

1. 1
   Source onboarding

   Admin → add source → connection test → schema discovery → field classification → activate refresh.

2. 2
   Mapping

   Developer → create mapping → preview → quality rules → publish (audited).

3. 3
   Analytics

   Governed metrics → dashboard → sharing → scheduled report or alert.

4. 4
   Operations

   Job monitoring (with on-demand cancellation of running executions), freshness, quality; audit journal; periodic health check.

• Upgrade: new pinned bundle, verified, with documented migration steps; documented rollback.
• Modules: enable/disable via license; disabling never deletes data.
• Support: locally-generated diagnostic report for support sessions, without exposing secrets.

• It is not an HRIS — it analyzes data from your source systems rather than replacing them.
• It executes no arbitrary user code (free SQL, scripts, macros) — every transformation is declarative and validated.
• It has no mandatory cloud dependencies and sends no data outside your perimeter.
• It does not guarantee GDPR / AI Act compliance out of the box — it provides the controls; policies (retention, purpose, cohort thresholds) remain your decision.