Geospatial Route-to-Market Optimisation

The Challenge

BAT's national field force was responsible for servicing approximately 300,000 retail outlets across Australia. Territory boundaries had been drawn years earlier and had never been systematically reviewed — they reflected historical organisational structures, not the current geography of outlet density or field rep capacity.

The result was predictable: some territories were severely overloaded while others were under-utilised, field reps were travelling excessive distances between stops, and Route-to-Market costs were rising without a corresponding increase in coverage quality. The problem was poorly understood because no one had looked at it spatially.

Approach

I built a geospatial analytics pipeline that combined outlet location data with field rep home base coordinates and historical visit records. The analysis unfolded in three stages:

• Spatial audit: Mapped all 300,000 outlets using GeoPandas to visualise territory boundaries against actual outlet density. Identified regions where territory lines created irrational travel paths.
• Clustering: Applied K-Means clustering to group outlets by proximity, treating each cluster as a natural service unit. Compared cluster boundaries to existing territories to quantify misalignment.
• Route redesign: Generated revised territory proposals based on cluster membership and rep capacity, minimising total travel distance while maintaining even workload distribution across the field force.

The analysis was visualised in Matplotlib with side-by-side maps of current vs. proposed territories, which made the inefficiencies immediately legible to senior stakeholders who had not previously seen the problem represented spatially.

Key Results

The revised territory design was adopted nationally. Field reps reported meaningfully shorter travel days, and the cost savings were verified in the following quarter's operational budget review.

Lessons Learned

• Visualisation was the most important deliverable. The maps made the problem self-evident in a way that no table of statistics could have achieved — leadership approved the redesign within a week of seeing the side-by-side comparison.
• Geospatial problems often hide inside conventional data. The cost inefficiency was visible in aggregate financials, but its root cause was only discoverable once the data was viewed spatially.
• K-Means clustering requires deliberate tuning of k. I iterated across multiple values and validated cluster coherence against business rules (minimum outlets per territory, maximum travel radius) before settling on the final groupings.