carto-territory-planning

Territory Planning (Territory Balancing + Location Allocation)

Builds CARTO Workflows that solve two related spatial optimization problems: dividing areas into balanced territories (e.g. sales regions, service zones) and finding optimal facility locations that maximize coverage or minimize cost.

Prerequisites: Load carto-create-workflow for the development process, JSON structure, and validation. It covers both access paths — the MCP server's workflow tools (create_workflow, validate_workflow, run_workflow) when attached, the carto workflows CLI otherwise. Routing signals: carto-basics/references/access-paths.md. Both components require the Territory Planning Extension Package installed on the connection.

Territory Balancing

Divides a study area into spatially contiguous territories that are balanced by a demand variable (e.g. workload, population, revenue).

Pipeline

Source Data (points) -> Spatial Indexing -> Aggregate per Cell -> Territory Balancing -> Save

Step 1: Load Source Data

Use native.gettablebyname. Input is typically point data with one or more numeric metrics to balance.

Success: Node outputs a table with a geometry column and numeric metric columns.

Step 2: Convert to Spatial Index

Use native.h3frompoint (or native.quadbinfromgeopoint) to assign each point to a grid cell.

Resolution guidance: Higher resolution = smaller cells = finer-grained territories, but more computation. H3 resolution 8-9 is typical for city-level analysis.

Success: Output has a spatial index column (e.g. h3).

Step 3: Aggregate per Cell

Use native.groupby to produce one row per cell with the metrics to balance:

• Group by: the spatial index column (h3)
• Aggregation: e.g. geom,count,revenue,sum,population,avg

Success: One row per cell with numeric columns for demand and optional similarity features.

Step 4: Run Territory Balancing

Use native.territorybalancing with:

Success: Output contains all input rows with an additional territory assignment column. Each cell is assigned to exactly one territory.

Step 5: Save

Use native.saveastable to persist results. The spatial index column is directly visualizable in CARTO Builder, colored by territory ID.

Success: Validated workflow uploadable via create_workflow (MCP) or carto workflows create (CLI).

Location Allocation

Finds the optimal subset of candidate facility locations that either maximize demand coverage or minimize total cost/distance.

Pipeline

Demand Data (grid/points) -> Candidate Locations -> Location Allocation -> Save

Step 1: Load Demand Data

Use native.gettablebyname to load a grid or point dataset representing demand (e.g. population per H3 cell, customer locations).

Success: Table with a spatial index column and a numeric demand column.

Step 2: Load Candidate Locations

Use a second native.gettablebyname for potential facility locations (e.g. existing infrastructure, candidate sites).

If candidates need filtering, use native.wheresimplified or native.limit to reduce the set -- fewer candidates = faster computation.

Success: Table with a spatial index column representing candidate sites.

Step 3: Run Location Allocation

Choose the variant based on the objective:

Maximize Coverage

Use native.locallocallocation_maximizecoverage:

Success: Output identifies which facilities to open and which demand cells each facility serves, maximizing total covered demand within the radius.

Minimize Total Cost

Use native.locallocallocation_minimizetotalcost. Same inputs as Maximize Coverage but without coverageradius — the objective is total weighted travel distance, not coverage within a radius.

Success: Output identifies which facilities to open, minimizing total weighted travel distance between demand and assigned facilities.

Step 4: Save

Use native.saveastable to persist results.

Success: Validated workflow uploadable via create_workflow (MCP) or carto workflows create (CLI).

Gotchas

• Provider casing & SQL dialect. This skill uses lowercase column names (h3, geom_count, population, sentiment_avg, etc.) — BigQuery / Databricks / Postgres / Redshift convention. On Snowflake, unquoted identifiers surface UPPERCASE — reference them as H3, GEOM_COUNT, POPULATION, SENTIMENT_AVG. See carto-create-workflow/references/providers/<provider>.md for casing rules and SQL dialect equivalents.
• Extension package required: Both territory balancing and location allocation require the Territory Planning Extension Package installed on the specific connection being used. This is NOT available by default -- the user must have it installed. Validation will fail if the package is missing.
• Location allocation is computationally expensive with many candidates. Pre-filter candidates (by geography, capacity, or other criteria) to keep the candidate set small. Hundreds of candidates can be slow; thousands may time out.
• Coverage radius units depend on the spatial reference system -- typically meters when using geography types. Verify the unit matches your intent.
• Territory balancing optimizes a trade-off between spatial contiguity and balance. Perfect balance is not guaranteed -- territories will be approximately equal in the demand variable, not exactly equal.
• Grid resolution sensitivity: Results change significantly with the input grid resolution. Finer grids give more precise boundaries but increase computation. Start with H3 resolution 8 for city-scale and adjust.
• The component names use native.territorybalancing (no dot separator in "territory balancing") and native.locallocallocation_maximizecoverage / native.locallocallocation_minimizetotalcost (note the double "local" prefix).
• Demand data must cover the study area. Gaps in the demand grid mean those areas are invisible to the optimizer. Use polyfill + COALESCE(value, 0) if you need full coverage.

Reference Templates

Academy Tutorials

• Territory Balancing tutorial -- step-by-step walkthrough of the Milan POS use case
• Location Allocation tutorial -- telco network management with location allocation
• AI Agent + Location Allocation -- using AI agents with location allocation for emergency response hub placement

Common Variations