Nomadica

Project Overview

Nomadica is a direct-to-consumer e-commerce brand specializing in outdoor and travel gear, managing 8,400 SKUs across multiple product categories with $12M in annual revenue. Their inventory decisions were based primarily on intuition and basic spreadsheet analysis, leading to frequent stockouts and excess inventory.

NodeSparks was engaged to build an ML-powered inventory forecasting system that would predict demand with enough lead time to optimize purchasing decisions and reduce both stockouts and overstock situations.

The Challenge

Nomadica's operations team was making purchasing decisions based on gut feel and simple moving averages, which couldn't account for seasonality, promotional effects, or emerging product trends.

The inventory management issues were significant:

• Average of 2.8 stockouts per month on high-velocity SKUs
• $420,000 tied up in slow-moving inventory
• No visibility into demand patterns beyond simple historical averages
• 21-day supplier lead times requiring accurate long-range forecasting

Standard inventory management tools were evaluated but couldn't handle Nomadica's complex seasonality patterns, promotional calendar, or the long-tail nature of their catalog. They needed a custom solution that could learn from their specific data patterns.

Our Approach

We built a time-series forecasting engine using XGBoost, trained on historical sales data combined with external signals like seasonality and promotional calendars.

Data Pipeline:

• Shopify Plus data extraction via custom API integration
• Snowflake data warehouse for historical sales, inventory, and promotion data
• Feature engineering including day-of-week, seasonality indices, and promotional flags

ML Model Architecture:

• XGBoost ensemble model for robust predictions across SKU velocity tiers
• Separate models for high-velocity vs. long-tail products
• Rolling window training for continuous model improvement

Forecasting System:

• 21-day forward demand predictions updated daily
• Confidence intervals for risk-adjusted purchasing decisions
• Automated reorder alerts when projected inventory falls below safety stock

The Solution

• Custom XGBoost forecasting model trained on 3 years of historical data
• Daily automated predictions for all 8,400 SKUs
• 21-day demand forecast with confidence intervals
• Shopify Plus integration for real-time inventory and sales data
• Snowflake-based analytics warehouse for model training and reporting
• Interactive dashboard showing forecast vs. actuals and inventory health
• Automated email alerts for reorder triggers and stockout risks
• Monthly model retraining pipeline for continuous accuracy improvement

Business and Technical Outcomes

• 74% reduction in stockouts (from 2.8 to 0.7 per month)
• $267,000 in recovered annual revenue from prevented stockouts
• 87.3% forecast accuracy at the SKU-day level
• 32% reduction in slow-moving inventory within 6 months
• Purchasing decisions now data-driven with 21-day forward visibility
• Operations team saved 8+ hours/week previously spent on manual analysis
• Model improves automatically with each monthly retraining cycle

The Impact

Nomadica's operations team transformed from reactive firefighting to proactive inventory management. The 21-day forecast window gives them ample time to place orders with suppliers, even accounting for shipping delays and seasonal demand spikes.

The 74% reduction in stockouts directly translated to recovered revenue—customers who would have bounced to competitors now complete their purchases. The $267K annual revenue recovery represents a significant ROI on the forecasting system investment.

The modular architecture allows Nomadica to extend the system as they grow—adding new product categories, incorporating marketing calendar data, or adjusting for new fulfillment centers as their business expands.