The challenge
A regional warehouse distributor managing 15,000+ SKUs was stuck in a cycle of overstocking slow movers and running out of bestsellers. Their Excel-based forecasting was 6 weeks behind reality. Stockouts were costing $400K/year in lost sales, while $1.2M sat in dead inventory. Seasonal demand spikes caught them off guard every time.
What we built
We built a machine learning forecasting system that predicts demand at the SKU level and automatically triggers reorder workflows.
- 01
Data Pipeline with Python + Supabase
Consolidated 3 years of sales history, supplier lead times, seasonal trends, and external signals (weather, holidays, promotions) into a clean Supabase data warehouse. Python ETL scripts run nightly to keep the dataset current.
- 02
Demand Forecasting with TensorFlow
Trained a TensorFlow LSTM model on historical demand patterns for each SKU category. The model accounts for seasonality, trend, promotional uplift, and day-of-week effects — generating 12-week rolling forecasts updated daily.
- 03
Automated Reorder Workflows via N8N
N8N workflows monitor inventory levels against forecast thresholds. When a SKU is projected to run low, the system auto-generates a purchase order, sends it to the supplier via email, and updates the ERP — no human intervention needed.
- 04
Dashboard & Alerts
Built a real-time dashboard showing forecast vs. actual demand, inventory health scores, and cash tied up in stock. Daily Slack alerts notify the warehouse manager of upcoming stockout risks and overstock situations.
Built with
Delivered in 6 weeks
Python
Data pipeline & model training
TensorFlow
LSTM demand forecasting model
N8N
Reorder automation & alerts
Supabase
Data warehouse & real-time dashboard
Slack
Inventory alerts & daily reports
Pandas
Data transformation & analysis
The outcome
Stockouts dropped 89% in the first quarter. Excess inventory was reduced by 34%, freeing up $420K in working capital. The model achieves 92% forecast accuracy at the SKU level, and the warehouse team now plans proactively instead of reactively.
- Stockouts
- -89%
- Excess Inventory
- -34%
- Revenue Recovered
- $400K
- Forecast Accuracy
- 92%
Want something like this?
Tell us what's slowing your team down. We'll show you where AI fits.