- May 19
- 6 min read
BIZZIREX ARTICLE
Why Hexbin Aggregation Fails for Site-Level Location Intelligence
and the Emergence of On-Demand Location Intelligence Reports
By Dr. Nicholas Patorniti, Emma Tremble, Murilo Colin and Raiza Benedecti
BizziRex Co-Founders
The hidden accuracy problem in modern location intelligence.
Over the last decade, location intelligence has evolved rapidly alongside the exponential growth of mobility data. Billions of mobile and device signals are now processed daily to estimate visitation patterns, consumer behaviour, and spatial interaction across cities.
To handle this scale, the industry widely adopted hexagonal spatial aggregation — commonly known as hexbin mapping or hexbin binning — a spatial method of grouping or aggregating data into a hexagonal cell. Hexbin binning is used as a practical solution for managing massive datasets, to reduce computational complexity, enable visual density analysis, and allow large-scale spatial comparisons. It can also be used to desensitise sensitive data for display on self-service software platforms.
However, as location intelligence moved from macro-scale urban analysis toward operational decision-making, a fundamental problem emerged:
Aggregation optimised for scalability does not necessarily preserve spatial truth.
When insights must be generated at the level where real-world decisions occur — individual stores, commercial buildings, venues, or parcels — small geometric distortions translate directly into incorrect conclusions about human activity.
This article investigates these limitations and demonstrates why precise spatial attribution requires a different analytical paradigm.
Why hexbin became the industry standard
Hexagonal grids became dominant in location intelligence because they solve a genuine computational challenge. Continuous geographic space is subdivided into equal-area cells, reducing visual clutter and enabling efficient processing of extremely large mobility datasets.
Hexbins provide several advantages:
• Scalable processing of billions of location pings;
• Simplified spatial indexing;
• Consistent neighbourhood structure;
• Efficient rendering for interactive analytics environments.
These characteristics made hexbin aggregation particularly suitable for large-scale spatial platforms and broad location-based analyses.
Yet these advantages come with an important trade-off:
aggregation replaces exact geographic measurement with approximation.
When raw mobility data is grouped into predefined hexbin cells, the original geographic precision of each point is partially lost. While acceptable for regional density visualisation, this loss becomes critical when analysing specific real-world locations.
Structural limitations of hexbin-based location attribution
1. Loss of spatial fidelity through aggregation
Hexbin aggregation converts precise georeferenced location point into grouped and averaged spatial representations. Once points are aggregated into hexbin cells, it becomes impossible to determine exactly which points fall within a specific area of interest (AOI).
For applications that depend on accurate attribution — such as estimating visits to a retail location or evaluating venue performance — this distinction is essential to ensure the analysis doesn't miss critical areas in the AOI or capture unnecessary surrounding areas occurring from the mismatch of AOI boundary and data. The analytical question shifts from where density occurs to who actually visited a defined place.
Aggregation answers the first question well.
It struggles with the second.
2. Boundary error: the underreported industry limitation
A critical limitation of centroid-based hexbin selection lies in boundary error — the geometric mismatch between the AOI polygon boundary and the artificial boundary implied by selected hexagonal cells.
Under the common centroid containment rule, a hexbin cell is included only if its centroid lies inside the AOI. Cells intersecting the boundary inevitably introduce two simultaneous errors:
Commission error — areas outside the AOI incorrectly included;
Omission error — portions of the AOI not represented by any selected hexbin cell.
These errors are structural, unavoidable consequences of aggregation rather than implementation mistakes. Their magnitude depends on the interaction between polygon size, shape complexity, grid orientation, and cell resolution.
Importantly, neither error is visible to the analyst using aggregated outputs.
Omission error removes valid visitors,underestimating activity by missing a portion within the AOI. | Commission error adds false visitors,inflating performance metrics by capturing passer bys. |
Because both occur simultaneously, aggregated results can be both undercounted and contaminated at the same time — without any explicit indication in the final analysis.
Empirical evaluation across real-world urban sites
To quantify these effects, four AOI were selected representing typical commercial analysis contexts:
• Commercial building with square shape;
• Stadium with regular shape;
• Restaurant with irregular complex shape;
• Large shopping centre with irregular shape;
Polygons of the AOIs were digitised from high-resolution imagery to represent true geographic boundaries. Three hexagonal grid resolutions (10m, 12m, and 20m edge lengths) were tested, covering relatively fine spatial configurations compared to common industry practice.
Selected 10m edge length hexagonal cells to represent: (a) Commercial Building, (b) Stadium, (c) Restaurant-scale, (d) Large shopping Centre.
Selected 12m edge length hexagonal cells to represent: (a) Commercial Building, (b) Stadium, (c) Restaurant-scale, (d) Large shopping Centre.
Selected 20m edge length hexagonal cells to represent: (a) Commercial Building, (b) Stadium, (c) Restaurant-scale, (d) Large shopping Centre.
For each configuration, five metrics were computed:
Polygon area of the AOI;
Selected hexbin area;
True geometric overlap;
Commission area;
Omission area.
Additionally, mobile device location points retained under each method were compared against exact polygon filtering using a point-in-polygon approach.
Results: accuracy degrades as spatial resolution becomes commercially relevant
The analysis confirms that boundary errors occur in every tested configuration. Even under relatively fine grid resolutions, centroid-based hexbin selection systematically misrepresents real geographic boundaries.
Large Sites vs Small Sites: A Critical Industry Blind Spot
Large locations containing many hexbin cells show relatively modest error levels. The shopping centre maintained high data retention rates, demonstrating why aggregation performs reasonably well for large-scale urban features.
However, results change dramatically for smaller sites.
At commercial building scale, omission error exceeded 36% even at fine resolution, and reached 53% under coarser grids. In practical terms, more than half of valid location observations may disappear from analysis.
Example of omission and commission errors generated by hexbin-based spatial attribution.
Note: the image includes data from a single day only.
The restaurant-scale site showed increasing commission contamination as grid size increased, with large cells extending beyond irregular boundaries and falsely attributing nearby pedestrian activity to the location. With 10m hexagonal-grid, approximately 11.5% of the AOI is unrepresented (omission) while 16% of the attributed pings originate from outside the AOI (commission); With 20m hexagonal-grid, omission reaches 18.8% while commission-based contamination reaches approximately 45%.
These findings reveal a critical blind spot:
the smaller and more operationally relevant the location becomes, the less reliable hexbin attribution becomes.
The restaurant-scale site showed increasing commission contamination as grid size increased, with large cells extending beyond irregular boundaries and falsely attributing nearby pedestrian activity to the location. With 10m hexagonal-grid, approximately 11.5% of the AOI is unrepresented (omission) while 16% of the attributed pings originate from outside the AOI (commission); With 20m hexagonal-grid, omission reaches 18.8% while commission-based contamination reaches approximately 45%.
More details of the findings are detailed in the table below:
These findings reveal a critical blind spot:
the smaller and more operationally relevant the location becomes, the less reliable hexbin attribution becomes.
How BizziRex is innovating the location intelligence industry
From aggregated analytics to precision site-level location intelligence
The results suggest that site-level location intelligence is entering a methodological transition. Early industry solutions prioritised scalability, enabling large datasets to become accessible. The next phase prioritises precision.
As analytical questions move closer to operational decisions, accurate spatial attribution becomes more valuable than computational convenience.
BizziRex defines a new analytical paradigm:
Precision Location Intelligence by selecting raw mobility points within the user-defined area of interest.
The BizziRex method eliminates both commission contamination and omission gaps, preserving the full informational content of raw mobility data regardless of polygon size or geometry.
Rather than approximating reality through predefined spatial units, analysis is grounded in exact geographic relationships.
Comparison between hexbin aggregation and BizziRex point-in-polygon spatial attribution.
With this approach, BizziRex delivers On-Demand Location Intelligence Reports generated specifically for a decision context, where spatial precision is preserved and insights are derived directly from exact geographic capture rather than aggregated approximation.
Toward a new standard: On-demand location intelligence
The findings highlight an industry-wide transition. Early location intelligence platforms were designed to make large datasets usable. As adoption matures, organisations increasingly require defensible, reproducible measurements tied to specific physical locations. This evolution shifts emphasis from scalable visualisation toward analytical precision.
On-Demand Location Intelligence Reports from BizziRex are a user-defined reports that analyses raw mobility data to show how people visit, stay and move around physical locations. Instead of relying on pre-aggregated spatial grids, analysis is generated dynamically for each location, preserving spatial fidelity and methodological transparency.
No data agreggation nor approximation:
The true human behaviour as it is.
On-demand Location Intelligence Reports from BizziRex can be ordered in minutes and delivered in a few days without requiring software platforms or technical expertise.
Need site-level precision without the flaws of spatial aggregation?
About the Author
Dr. Nicholas Patorniti is the founder of UACS consulting – an urban analytics specialist consultancy services company operating for 10 years. He is a senior adjunct research fellow at Griffith University, Australia. Over 20years experience researching, developing and applying urban analytics methods helped identified this market gap and provided the ‘know-how’ to meet the market demand with BizziRex.