Methodology

Route Optimization Methodology

How priority scores and routes are calculated

System Overview

The route optimization system transforms raw sales data into actionable daily routes. The process has three stages:

1. Score

RFM scoring assigns each customer to a segment

2. Prioritize

Visit priority calculated from segment + urgency

3. Optimize

TSP algorithm sequences stops efficiently

Visit Priority Score

How customers are ranked for visits

Each customer receives a visit priority score from 1-100, calculated as:

priority = (segment_weight × 40) + (recency_urgency × 30) + (monetary_value × 20) + (initiative_bonus × 10)

Component	Weight	Description
Segment Weight	40%	Champions/Cannot Lose = high, Lost = low
Recency Urgency	30%	Days overdue vs typical purchase frequency
Monetary Value	20%	Historical revenue tier (quintile)
Initiative Bonus	10%	Part of Cannot Lose, At Risk, or Reactivation list

Segment Weights

Segment	Base Weight	Rationale
Cannot Lose	1.0 (max)	Urgent retention
Champions	0.9	Protect best customers
At Risk	0.8	Proactive retention
Loyal	0.7	Maintain relationship
New	0.6	Nurture early
Lost/Hibernating	0.3	Lower priority unless reactivation target

Route Generation Process

When a route is generated for a zone, the following steps occur:

Select Mandatory Stops

All Cannot Lose customers in the zone are included automatically

Fill Remaining Slots

Sort remaining customers by visit_priority, add top N to fill max_stops

Optimize Sequence (TSP)

Use OR-Tools to minimize total travel distance between stops

Attach Context

Add recommended actions, historical data, and contact info to each stop

TSP Optimization

Traveling Salesman Problem solver

The Traveling Salesman Problem (TSP) finds the shortest route visiting all stops exactly once. We use Google OR-Tools with the following configuration:

Distance Metric

Haversine (great-circle) distance between geocoded addresses

Solver Strategy

PATH_CHEAPEST_ARC with guided local search

Time Limit

30 seconds max solve time

Fallback

City centroid used when address geocoding fails

Route Output Format

Each route stop includes the following context:

PARADA 1: TIENDA LA ECONOMIA (CANNOT LOSE)

- Ultimo pedido: 45 dias (Normal: cada 15 dias)

- Valor historico: $3.2MM

- Riesgo churn: 78%

- Accion: Visita de retencion, ofrecer descuento 5%

- Contacto: Juan Perez, 311-XXX-XXXX

Limitations

Geocoding accuracy depends on address quality in source data
TSP assumes direct travel; doesn't account for road networks
No time windows or capacity constraints currently modeled
Churn risk is estimated from RFM scores, not a predictive ML model