Analytics

The Analytics module is your one-stop view of how the whole company is doing. Unlike module-specific dashboards (e.g. the Performance dashboard, the Engagement dashboard), Analytics is the cross-module composition layer — it pulls signals from People, Recruiting, Performance, Compensation, Time & Leave, Engagement, and Career into a single narrative.

Pages

The tab strip carries seven tabs: Overview, People, Performance, Recruiting, Surveys, Governance, Builder. The deeper dives keep their own pages and open from inside those tabs: DEI, Retention, Succession, and Pay Equity from the "More people analytics" cards on the People tab, and Insights Hub from the "View all insights" link on the Overview.

Analytics Home (/analytics)

Your daily landing page. Three rows:

• Your saved reports — cards for reports you've built or pinned, each showing the headline value, directional delta, and a 12-point sparkline.
• Featured insights this week — top 3 insights ranked by severity. Click any card for evidence + recommendations, or "View all insights" to open the Insights Hub.
• Modules at a glance — 7 module cards (People, Recruiting, Performance, Compensation, Time & Leave, Engagement, Career), each with 3 KPIs and a deep-link to the module's analytics page.

Known limitation: the "Set as Home" action in Report Builder doesn't do anything yet — it's not wired up. There's currently no way to change your Analytics Home view.

Insights Hub (/analytics/insights)

Reached from the "View all insights" link on Analytics Home. Triage view for cross-module alerts. Severity tabs (All / Critical / High / Medium / Low) with count badges, search across title/description/module, and per-card actions: Details (opens the evidence + recommendations modal) and Acknowledge (marks the insight as resolved).

Insights have two flavors:

• AI (✦ badge with confidence score) — generated by Sparko's models with confidence 70-96%
• Rule (no badge) — deterministic threshold queries (e.g. "PTO accrual >25 days")

Performance Analytics (/analytics/performance)

Cross-module performance signals: OKR Health funnel (on-track / at-risk / off-track / completed), calibration band distribution, department × performance-band matrix, and the at-risk employees rail.

People Analytics (/analytics/people)

Headcount metrics: department breakdown, headcount-by-leadership (CEO excluded), tenure buckets, locations, 12-month headcount trend, and a movement summary (hires / promotes / transfers / exits).

Recruiting Analytics (/analytics/recruiting)

Hiring funnel (applicants → phone screen → interview → offer → hired), hiring-plan progress per department, open requisitions table with days-open and stage, source-mix bars.

DEI Analytics (/analytics/dei)

Reached from the "More people analytics" cards on People Analytics. Diversity, equity, and inclusion. Gender donut, ethnicity donut (gated by analytics.dei.view_pii), leadership representation by level, department × gender stacked bars, hire/promote/exit funnel by gender, pay-gap summary table. Cohort-row counts below 5 employees are suppressed under a k-anonymity floor.

Pay Equity (/analytics/pay-equity)

Reached from the "More people analytics" cards on People Analytics, or from the Compensation Hub. Controlled and uncontrolled pay-gap analysis across cohorts (band × department), plus outlier employees whose base salary is >15% off their cohort median.

Bias & Governance (/analytics/governance)

The adverse-impact report third-party auditors ask for, computed live rather than assembled from exports. Requires analytics.dei.view_pii.

Sparko runs the four-fifths (80%) rule across three decisions:

• Hiring — advancing past AI screening, the automated decision NYC Local Law 144 regulates. Grouped by candidates' own voluntary EEO answers.
• Promotion — recommended for promotion in a completed talent-calibration session. Considered = everyone reviewed; selected = recommended.
• Compensation — a merit increase approved. Considered = everyone whose recommendation was decided; selected = approved.

Each is broken out by gender, race/ethnicity, veteran status and disability status.

The group-first view is the one an export-based tool cannot produce: it shows where a single group is disadvantaged at more than one stage, which is the pattern that matters and the one a per-decision report hides.

Only finished decisions count. A calibration session still in discussion, or a merit recommendation nobody has decided yet, is left out — grading work in progress invents findings.

Small groups are withheld. Under 5 people, the group's numbers are hidden entirely rather than shown; between 5 and 30 they are shown but marked provisional. A withheld group's size cannot be recovered by subtracting the groups that are shown.

When it can't be computed, it says so rather than showing a clean result. If demographic values cannot be read, the report is marked incomplete — on screen and in the export. "Not enough data to audit yet" is never presented as "no bias found".

CSV export for filings under NYC LL144, Illinois HB 3773 and Colorado SB 24-205, carrying the window and a data-quality column alongside each row.

This is a diagnostic, not a determination. Consult qualified counsel before acting on it.

Retention (/analytics/turnover)

Reached from the "More people analytics" cards on People Analytics. Tenure-cohort retention (0–6m / 6m–1y / 1–2y / 2–4y / 4y+), 12-month attrition trend line, department × month heatmap, exit reasons breakdown, and a "High flight risk" rail with scores sourced from the Engagement module.

Model accuracy sits directly under the flight-risk rail, because an accuracy figure on another page is one nobody checks before acting on a name. It grades past risk levels against who actually left:

• A calibration table — for each risk band, how many people were in it and what share of them left within the horizon — plus lift, precision and recall at the high band. A band's rate is readable at small numbers where a single headline percentage would mislead.
• Only finished horizons count. Scoring someone as "correctly predicted to stay" a week after the prediction is how this number gets inflated, so predictions still inside their window are excluded rather than counted as successes.
• One person, one observation, even though scoring runs weekly.
• Departures the model doesn't claim to predict leave the group — a recorded layoff, retirement or end of contract is a different process, not a flight-risk miss.
• Bands under 5 people are withheld, and the derived figures are withheld with them so the hidden counts can't be recovered.

Until a group of predictions has finished its horizon, the page says "Still accumulating" and names the date the first readout becomes available. It does not show a number derived from an unfinished cohort.

Departures with no recorded separation type are counted, and the page tells you how many — recording voluntary or involuntary at offboarding sharpens the figure over time.

Acting on a flight-risk flag

Where to find it: the Take action button on each row of the High flight risk card, on Turnover & Retention.

From a flagged employee, open a retention compensation review or schedule a 1:1 with their manager. Sparko records which flag prompted the action, so the effect of acting can be measured later.

Both actions are safe to repeat: a second click reopens nothing, and the menu marks what has already been done so two managers don't act twice.

The comp review opens as a draft with the person's current salary and title filled in and no proposed figure. The model identifies who is at risk, not what they should be paid, and a pre-filled number would read as a recommendation it never made.

Succession (/analytics/succession)

Reached from the "More people analytics" cards on People Analytics. Succession bench view sourced from the Career module's intelligence data.

Survey Trends (/analytics/surveys)

Cross-survey participation and status board: counts by status, participation-over-time bars, and eNPS per survey. Rows link to each survey's full results in Engagement (Engagement > Surveys > View results). Survey creation and per-survey results live in the Engagement module.

Report Builder (/analytics/builder)

Build custom reports or start from a template. 3-pane workspace:

• Library rail (left) — Reference reports (7 Analytics pages as one-click composites), Templates (16 atomic charts by module), My reports (your saved ones), Recent.
• Center canvas — Editable report name, action row (Set as Home / Save / Share / Schedule / Export / Run), tab strip (Data / Filters / Visualization / Preview), and the live chart preview.
• Right rail — Chart-type picker (Bar / Line / Pie / Donut / Distribution / Heatmap / Table — distribution + heatmap arrive in a follow-up release), data source dropdown, X-axis and Y-axis field slots, and the available-fields list.

Click a field to assign it to X or Y axis. Click Run to execute the query against the live database and render the result.

Known Limitations

• Export formats: CSV and PDF work today. Excel and PNG export options currently return an error — don't rely on them yet.
• Date-range/period selectors on the pre-built analytics pages (Analytics Home, Turnover, People, Performance) are largely cosmetic right now — changing the period doesn't refetch the data behind most tiles. The Performance Analytics cohort tab's own period selector is the one exception that does filter live.
• Turnover and Performance Analytics stat tiles: a few individual tiles on these two pages (e.g. specific attrition and OKR-progress percentages) are still showing placeholder figures rather than your company's live numbers — everything else on the page is live. This is being fixed.
• AI insight generation runs on a weekly cadence (not daily) for Performance, People, Engagement, Recruiting and Skills. The Org Health narrative is monthly. Compensation and Time & Leave are not covered yet.
• POST /analytics/custom-report currently errors on every call (a missing internal service) — don't rely on this endpoint yet.
• Scheduled report delivery via Slack or a shareable link silently falls back to in-app delivery only — if you pick Slack as the delivery method in Schedule Report, no Slack message is actually sent.
• There's no digest or push notification for new AI insights — you have to visit Insights Hub yourself to see what's new.

Permissions

Action	Permission
See the Analytics nav entry	analytics.module.view (HR Admin / IT Admin / Manager / Super Admin)
View DEI Analytics page	analytics.dei.view (HR Admin / IT Admin)
See ethnicity + pay-gap inside DEI	analytics.dei.view_pii (HR Admin only)
Build / edit custom reports	analytics.reports.create (HR Admin / IT Admin)
Schedule recurring delivery	analytics.reports.schedule (HR Admin / IT Admin)
Share saved reports	analytics.reports.share (HR Admin / IT Admin)
Acknowledge insights	performance.insights.acknowledge (HR Admin / Manager)

Employees do not see the Analytics nav entry at all. Analytics is the whole-company view by design: managers with the analytics permission see company-wide aggregates, not a subtree slice. People Analytics is open to HR and managers alike; the deeper breakdowns (DEI, Retention, Pay Equity) additionally require company-wide permissions and show restricted or empty states without them.

Tier availability

See docs/FEATURES_AND_PRICING.md § "6. Reporting & Analytics Module" for the per-tier feature matrix.

Related

• AI features deep-dive: docs/help-center/ai-features/insights-hub.md
• For module owners building new pages that consume Analytics endpoints, see .claude/docs/development/backend-patterns.md § "Scope Helper Trio" — specifically the assert_tenant_scoped walker for any user-config-driven queries.