Escalation Decision Engine

The Bottleneck

In Global Support, the biggest operational cost isn't case volume—it's escalation ambiguity. Consultants lack a structured way to determine whether a case should be Resolved, Reviewed, or Escalated, leading to inconsistent triage, unnecessary escalations, and missed signals. Traditional tooling focuses on information retrieval but provides no decision intelligence layer — and no feedback loop to measure whether recommendations are actually followed or effective.

The Architectural Solution

I built a weighted decision engine that integrates knowledge retrieval, historical pattern analysis, and risk scoring into a unified decision workspace with a closed feedback loop. Rather than asking consultants to manually cross-reference documentation, case history, and escalation criteria, the system computes a single escalation score with full reasoning trace — and then tracks whether the recommendation was followed and what the outcome was.

• Dual-Mode Interface: Decision Workspace for frontline execution and Analyst View for transparency, diagnostics, and system-level insight.
• End-to-End Case Tracking: Each analyzed case now carries a case number, enabling tracking from intake to outcome for reporting and auditability.
• Post-Recommendation Follow-Through: After receiving a recommendation, users record whether they followed it, what the actual outcome was (resolve, escalate, review), and optionally add notes — with stronger guidance when the recommendation is not followed.
• Workflow Outcomes Tab: A dedicated operational view with KPI cards (follow rate, override rate, resolution rate), action breakdown chart, recent follow-through records table, and filters for Category, Recommendation, and Followed Recommendation — plus CSV export for BI handoff.
• Weighted Scoring Model: Combines article escalation flags, match quality, historical escalation rates, high-risk language detection, and simple-resolution signals into a single 0–100 score.
• Explainable Output: Every recommendation includes a plain-language reasoning trace showing which factors drove the decision.
• Pattern Integration: Historical case data feeds escalation rate calculations and trend analysis, enabling the system to learn from past outcomes.

Key Upgrades (v0.2)

• Case number intake added for end-to-end tracking and reporting.
• Post-recommendation feedback capture (followed? outcome? reason/notes) with conditional outcome paths.
• Dedicated Workflow Outcomes tab with KPI cards, action breakdown, filterable records, and CSV export.
• Seeded demo data for immediate populated experience in Walkthrough Mode.
• Python compatibility fix (Optional over union syntax) for broader interpreter support.

Command Stack

Python | Streamlit | SQLite | Decision Logic | Explainable AI | BI-Ready Export