Logistics & Warehouse Forecasting

Problem

Agricultural last-mile logistics fail when inventory forecasts are late or locally wrong. Duka and warehouse stock must move with seasonal, spatial and behavioural demand.

Context

Applied forecasting work at One Acre Fund during a senior data scientist role: automating real-time logistics, duka and warehouse inventory prediction.

Dataset

Operational inventory, logistics and sales systems from One Acre Fund's field operations.

Methodology

Forecasting pipelines combining statistical methods with operational data engineering so predictions arrive in time to change a shipment, not a slide.

Model / architecture

Time-series and demand models for warehouse and duka inventory, including NLP-based churn and sentiment models for customer engagement.

Results

Automated real-time forecasting pipelines supporting marketing and operational decision-making across the organisation.

Evaluation metrics

Setting

Last-mile agricultural logistics

Challenges

Messy operational data, seasonality, and the organisational work of putting a forecast into a warehouse manager's day.

Impact

Less stock-out, less idle inventory: applied data science in a system that feeds farmers.

Tools / technology

• Python
• R
• SQL
• Forecasting
• Spark
• NLP