Product Case · Data & AI Product Strategy

HR and Legal sit on the same fragmented-data problem — just with different symptoms.

This is a Product Strategy 0–1 statement.

Self-directed portfolio project (2026), not a live enterprise deployment. Metrics and targets on this page are illustrative, built to show product judgment, not reported as measured results.

HR couldn't see attrition risk until it showed up in an exit interview. Legal couldn't see contract bottlenecks or missed obligations until someone asked. Both ran on manual, spreadsheet-stitched reporting from systems that were never built to talk to each other. I designed one governed semantic layer serving both functions — with lineage, provenance, and a review-lane model so AI outputs get flagged or suggested, never decided unsupervised.

Attrition intervention prototype
Who HRBPs, workforce planners, Legal Ops & counsel Two functions that need trustworthy data to act on, not two separate spreadsheets.
What Query one trusted model, not rebuild reports by hand A shared semantic layer with lineage, provenance, and clear ownership, serving both functions.
Why 4 systems, 1 semantic layer Workday, Agiloft, SFDC, and ServiceNow unified under one governed model, not four disconnected exports.

Two things that were broken

Reactive, not predictive

HR couldn't see attrition risk or skills gaps until they showed up in an exit interview; Legal couldn't see contract bottlenecks until someone asked.

No single source of reference

Every report was rebuilt from scratch by hand, stitched from systems never designed to talk to each other, with no lineage back to source.

What we designed

Product architecture

One semantic layer, governed once, serving two functions — not two disconnected dashboards built on four different exports.

WorkdayHR core
AgiloftContracts
SFDCBusiness context
ServiceNowCase & workflow
Semantic Model & Governance Layer
LineageProvenanceAuditabilityDomain Ownership
Workforce Intelligence Serves HRBPs & workforce planners
Contracts & Compliance Intelligence Serves Legal Ops & counsel
Query APIsDashboardsConversational AI

Figure 1 — shared semantic layer serving two modules through common governance.

AI/ML layer & responsible AI

Two models sit on top of the semantic layer — an attrition-risk score for HR, and a contract/compliance anomaly flag for Legal. Neither acts unsupervised.

Auto-apply High confidence, low risk

Routine compliance updates and standard renewal reminders — logged with a full audit trail.

Suggest Medium confidence

Attrition risk flagged to the HRBP, an unusual clause surfaced to counsel — one-click accept, edit, or dismiss.

Escalate Low confidence or high stakes

A potential compliance breach or high-value anomaly, routed directly to a named owner.

Figure 2 — every AI output routes to auto-apply, suggest, or escalate, based on confidence and stakes.

Every AI output is flagged, suggested, or escalated — never decided unsupervised.

Business case & metrics

Chosen to answer the question every stakeholder actually asks — is this worth what it costs — rather than vanity usage numbers. Illustrative targets, not measured results.

70%
Adoption
of target HRBPs & counsel active monthly by end of Phase 2
−65%
Cycle-time
reduction in time to produce a workforce or compliance report
0.80+
Forecast accuracy
AUC target for the 90-day attrition risk model
<10%
Model quality
false-positive rate on escalated contract / compliance flags
3–4x
Automation ROI
manual-hours saved per hour of platform investment