PROD · PP-2026-08 14:42:18 UTC
CSC AI People Ops Platform
An AI-powered orchestration and intelligence layer on top of Workday. Four workflows, one architecture: deterministic detection → AI narrative → validator gate → human disposition → Workday Business Process trigger. Covering payroll, workforce intelligence, talent acquisition, and cross-border compliance across 13 portfolio brands, 6 US states + Ontario, and 78,000+ employees.
Workday v46.0 SOAP + REST
Anthropic Claude Sonnet 4.6
n8n orchestration
Pinecone semantic retrieval
Slack Block Kit HITL
Recent activity
CRITICAL: Garnishment not applied — EMP-48329 (Backcountry, UT). Active child-support order CO-2024-77821 shows $0 deducted. Get_Payroll_Results
2 min ago · Workflow 1 · routed to Compliance
Resolved: Multi-state tax mismatch — EMP-33910 (Z Gallerie, TX). Tax election updated to match work location. Put_Worker_Tax_Election
18 min ago · Workflow 1 · approved by Maria Kowalski
Query answered: "Why is voluntary attrition spiking in Backcountry warehouses?" → HYBRID path, 4 sources, validator PASS. Claude Sonnet
34 min ago · Workflow 2 · asked by VP Operations
TA Alert: Senior Buyer interview (Sur La Table) — interviewer missed "Problem Solving" competency bucket. Real-time prompt delivered. Get_Interviews
1h ago · Workflow 3 · interviewer acknowledged
Cross-border: CAN payroll PP-2026-08 reconciled against US. FX rate locked at 1.3642 CAD/USD. 0 exceptions. Get_Payroll_Results (CAN)
2h ago · Workflow 4 · auto-reconciled
CRITICAL: Missing I-9 — EMP-22107 (Sur La Table, CA). Hire date 2026-03-15; I-9 still incomplete past 30-day deadline. Get_Workers
3h ago · Workflow 1 · escalated to CLC
Platform health · PP-2026-08
Pre-payroll catch rate
91%
↑ 4pts vs prior period · target >95% by month 12
Validator pass rate (1st attempt)
94%
on baseline (target >85%)
AI spend today
$8.42
42% of daily cap · 2,148 calls
Open Critical aging
3
1 past 4h SLA · page sent
Cross-border exceptions
3
CAN provincial tax + FX variance
System self-auditing & self-healing continuous · auto-actions logged to ai_decision_log
Self-auditing (system checks its own quality)
PASS Output drift detection: Validator PASS rate 94.1% (7d trend: +0.3pts). Human override rate 4.2% (stable). No anomaly class exceeds retry threshold.
PASS Shadow run (last prompt update): Re-ran PP-2026-07 findings through v2.4 prompt. 2 of 47 narratives changed (4.3%) — below 5% threshold. Deployed without block.
PASS Gold-set regression: 50/50 passed on current prompt v2.4 + Claude Sonnet 4.6. Last run: 4h ago.
Self-healing (system corrects its own failures)
🔄 Prompt auto-tuning: Garnishment class FAIL rate was 22% (>20% threshold) last week. System auto-generated prompt v2.4.1, gold-set passed 50/50. Staged for human approval.
🔄 Routing auto-correction: "LOA adverse action" cases rerouted from Compliance→HRBP 35% of the time (>30% threshold). Routing rule revision suggested. Awaiting confirmation.
OK Circuit breaker: No LLM degradation in 14 days. Last auto-fallback: Mar 28, 23:41 UTC (Anthropic 502, 8 min, 12 findings routed deterministic-only, auto-resumed).
OK Contract auto-repair: Last schema diff clean. No column renames detected. Next scan: tonight 02:00 UTC.
PROTOTYPE · synthetic data · all employees and entities are illustrative