AP analysts manually resolved Shipping Discrepancy tickets — gathering and validating data across five systems in parallel for every case. The result: 4-day first response, 8–11 day resolution, a 6,000+ ticket backlog and 283 new tickets daily. Vendor payments slipped; throughput could only grow with headcount.
What was in the way.
- AP analysts manually resolved shipping discrepancy tickets across five systems in parallel.
- 4-day first response, 8–11 day resolution, and a 6,000+ ticket backlog.
- 283 new tickets daily; throughput could only grow with headcount.
- SCOUT routes every ticket through 16 specialist agents coordinated by a state-machine orchestrator.
- Decision graphs and multi-tier memory ensure recommendations reflect full ticket history.
- A 24×7 feedback loop continuously tunes routing and prompts in production.
The build, step by step.
- 01
Orchestration
State-machine orchestrator coordinates 16 specialist agents per ticket.
- 02
Extraction & Validation
Data extraction, validation, payback, overage, damage and FedEx specialist peers.
- 03
Memory
Multi-tier memory across the full ticket history for grounded recommendations.
- 04
Human-in-the-Loop
AI recommendation ≥90% accepted by AP analysts; no-code workflow designer.
- 05
Observability
Power BI dashboards track agent efficiency and accuracy live from production.
- 06
Feedback Loop
24×7 monitoring refines prompts and routing continuously.
- Agent FrameworkMicrosoft Agent FrameworkAzure AI Foundry
- ReasoningAzure OpenAIAnthropic Claude
- Data LayerMicrosoft FabricData FactoryOneLake Lakehouse
- IntegrationsSAPAS400MPowerMyShipInfoZendeskFedEx
- PlatformAzure Kubernetes ServiceCosmos DBADLSKey Vault
- Last-MileUiPathDownstream actions.
Next-day decisions, four FTE of capacity returned, payback in year one. First response cut from 4 days to ~1.4, resolution ~30% faster, and ≥50% less manual effort across A→P workflows — 104K tickets a year without adding analysts.
The multi-agentic design effectively addresses the full complexity of our end-to-end workflow automation.
