B2B Territory Optimization (b2b_territory_optimization)

An open-source Python framework designed for Enterprise RevOps and Data Strategy teams to mathematically design, carve, and manage B2B sales territories.

This package bridges the gap between raw CRM data and the start of the forecasting cycle. It dynamically groups accounts into strict Taxonomic Buckets (e.g., "Enterprise AMER"), greedily balances TAM and Workload across territories, and generates human resource staffing matrices spanning multiple roles (AEs, SDRs, Managers) with configurable coverage ratios.

It perfectly complements its sister package, b2b-revenue-forecasting, which cascades quotas and forecasts pipeline after territories have been defined.

🚀 Features

💻 Seamless Forecasting Integration

This package was built to feed directly into b2b-revenue-forecasting. Once your territories are designed and staffed, you can export the hierarchy automatically:

from b2b_territory_optimization import SellerAssignmentMatrix
from b2b_revenue_forecasting.hierarchy import SalesHierarchy

# 1. Staff your territories
matrix = SellerAssignmentMatrix(territory_list)
matrix.add_role('Regional_VP', 5.0)
matrix.add_role('Sales_Manager', 2.0)
matrix.add_role('Account_Executive', 1.0)

# 2. Export to a flat hierarchy dataframe
df_hierarchy = matrix.export_to_hierarchy(
    hierarchy_columns_order=['Regional_VP', 'Sales_Manager', 'Account_Executive']
)

# 3. Feed directly into the forecasting package!
hierarchy = SalesHierarchy()
hierarchy.from_dataframe(df_hierarchy, path_cols=['Global', 'Regional_VP', 'Sales_Manager', 'Account_Executive', 'Territory_ID'])

📦 Installation

pip install b2b-territory-optimization

🧠 Key Concepts

Strict Taxonomy Buckets

Unlike generic K-Means clustering, B2B territory carving requires hard constraints. If a territory is designated for "Mid-Market EMEA", the algorithm physically cannot place an "Enterprise APAC" account into it, regardless of mathematical balance. TaxonomySchema enforces this by slicing the DataFrame into sub-graphs before allocation occurs.

Greedy LPT Balancing

The TerritoryAllocator uses an approach inspired by the Longest Processing Time (LPT) multiprocessor scheduling algorithm. It sorts accounts by the target metric descending, then iteratively places the largest unassigned account into the territory with the current lowest total sum. This consistently yields $< 0.1%$ variance on B2B workloads.

Capacity-Driven Prediction

Instead of hardcoding headcount numbers (e.g., "I want 3 territories in AMER"), the TerritoryAllocator supports predictive capacity modeling via allocate_by_capacity(). Provide a Target Capacity (e.g., "$1.5M TAM per territory") and the engine mathematically analyzes the Total TAM of every taxonomy bucket, predicts the exact headcount needed (using math.ceil()), and automatically carves it into that precise number of territories.

Manager Override Suggestions

When an Account Executive has a pre-existing relationship with a buyer, managers will often manually override the algorithmic territory design to keep them together. When an account is moved, ReassignmentEngine calculates the exact delta required to restore balance and evaluates itertools.combinations of smaller accounts to suggest the perfect trade.

Intelligent Bipartite Matching (Hungarian Algorithm)

Instead of assigning sellers blindly, the IntelligentAssigner models human allocation as a classic Operations Research Linear Assignment Problem. Using scipy.optimize.linear_sum_assignment, it optimally pairs a roster of real human sellers to carved territories.

• Hard Constraints: Dynamically reads taxonomy columns. If a seller's region is AMER, they are strictly locked out of APAC territories.
• Soft Constraints (Fit Optimization): Mathematically aligns Seniority with Territory TAM (Tier 1 reps get Tier 1 territories) and rewards Domain Expertise matches.

🤝 Contributing

Built explicitly for RevOps analysts, Data Scientists, and VP Revenue Operations executing scaling go-to-market strategies.

📄 License

MIT License