Network Metrics:
Net Negative: A locker where monthly costs exceed monthly revenue (transport savings + ad revenue < concessions + rent + maintenance + depreciation + other).
Annual Loss Exposure: Sum of net_savings_mo × 12 for all net-negative lockers.
DPMO: Defects Per Million Opportunities = (conceded / shipped) × 1,000,000. <8K good, >15K critical.
Throughput: Packages per slot per active day. >0.15 healthy, <0.08 requires intervention.
Priority Tiers:
P0 — Emerging Risk: Currently profitable but 4+ months consecutive decline. Projected to turn negative.
P1 — Critical (95.6%): Near-certainty of continued negative. Slope or maintenance path.
P2 — High (93.8%): Includes "Run Ads" path (operational fix, not structural).
P3 — Elevated (91.1%): Moderate decline detected.
P4 — Low (87.2%): Early throughput erosion.
Key Concepts:
Dual-path: Model classifies via EITHER slope deterioration OR maintenance spikes. Covers both failure modes.
Quality Gate: Good DPMO + good throughput + has ad rights → "Run Ads" not CRITICAL. Fix is operational.
Seasonal Detrending: Removes holiday volume effects from slope estimation. Prevents false positives.
EWS: Two-axis scoring (vs Network + vs Self). HIGH RISK = both axes bad. Can flag positive lockers.
Ad Wrap Cost: $209/mo charged ONLY when an actual ad campaign runs (not when wrap is merely installed).
Dispatches vs Tickets: Dispatches = technician sent ($150-500). Tickets = complaints filed. Not interchangeable.
Project NERVE (Network Earnings & Risk Visibility Engine) is a predictive financial deterioration model for the PARP Controllership team. It monitors 26,167 US Amazon Locker access points across 27 months of data (Mar 2024 – May 2026), identifies kiosks heading toward persistent losses, classifies them by urgency, and recommends operational interventions.
Current state: 9,836 lockers (37.6%) are net-negative, generating $16,649,572 in annual losses. Commercial segment is 58% negative; Residential is 7% negative.
Model performance: CRITICAL 95.6% | HIGH 93.8% | Combined 92.8% precision (3-month grace window). Validated across 5 quarterly walk-forward windows. 70.9% of flagged lockers remain negative every single month.
How it works: Hybrid dual-path classification (slope deterioration OR maintenance cost spikes) with seasonal detrending, quality gate for operational fixes, and actual vendor invoice costs. Built over 13 iterations from a basic dashboard to a 95.6% precision predictive engine.
Click any tab below to navigate. Each serves a specific purpose.
Purpose: How we know the model works. 5 quarterly backtests + miss analysis.
Use when: Explaining methodology or building trust in the predictions.
Purpose: Search any locker by KID, name, address, partner. AI-generated summary + detail card + PnL trend chart.
Use when: Quick lookup — "Tell me everything about this specific locker."
Purpose: Early Warning System. At-risk lockers scored on two axes (vs Network + vs Self). Catch deterioration early.
Use when: Proactive intervention — identify lockers heading toward failure while still positive.
Purpose: Every flagged locker with full PnL breakdown. Filterable by tier, partner, vertical, type, age, generation.
Use when: Operational planning — "which lockers do we act on?"
Purpose: Interactive geographic view. 500m DBSCAN clusters colored by health. Radius search.
Use when: Geographic strategy, dead zone identification, zone-level exit decisions.
Purpose: Root cause analysis across 11 categories. Click-to-expand breakdowns per locker.
Includes: Maintenance Spikes • DPMO Spiral • Volume Drop • High Tickets • Offline History • Delivery Failures • Cannibalization • Proximity Analysis • Partner Health • Renewals • Ad Opportunity
Purpose: All lockers — positive and negative. Single source of truth with complete PnL.
Use when: "How is locker X performing?" or portfolio-wide analysis.
Purpose: 47 metrics defined. Key principles. PnL formula. Model methodology.
Use when: "What does this metric mean?" or onboarding new team members.
Only 15% of commercial lockers earn actual ad revenue. Without ads: 75% are negative (avg -$136/mo). With ads: 40% negative (avg +$281/mo). Activating campaigns on lockers with existing ad rights = ~$6.7M revenue opportunity.
198 clusters (869 lockers) perform 2-5x worse than network average. NYCHA housing (478 lockers, all <2 years) = 55% of bad zones. Single partnership decision created the largest concentration of losses.
49% of lockers have $0 actual maintenance. 6% have >$100/mo. 46% of model misses were caused by maintenance spikes — now captured via the maintenance classification path.
406 residential leases expired without renewal. Partners stopped paying lease fees — $3.16M/yr in lost lease revenue. However, many of these lockers still generate positive PnL from transport savings. Action: Pursue renewal outreach to restore lease payments. Do NOT proactively remove any locker with positive PnL even if lease is expired.