Geofencing sounds simple: draw a boundary, count devices inside it. In production, the geometry of that boundary drives whether your visit counts, audience segments, and incrementality reads survive audit. Radius fences are cheap to generate at scale; custom polygons trace building footprints and are labor-intensive to maintain. The procurement mistake is treating both as interchangeable — they answer different measurement questions. Teams running POI & Geofencing for attribution, audience targeting for activation, or cross-channel measurement for closed-loop reads should decide polygon vs radius before panel sizing, because bad geometry shrinks effective panel depth by contaminating adjacent-business noise. The MRC measurement standards and IAB quality guidelines both treat place-definition accuracy as upstream of any visit metric.
A radius fence takes a lat/long center and a distance (often 50–200 meters) and captures every device event inside the circle. Generation is trivial — which is why self-serve platforms default to it. A polygon fence traces the indoor footprint, drive-through lanes where relevant, and sometimes parcel edges — built from licensed parcel files, aerial imagery, and human correction. GSDSI POI data ships polygon-primary boundaries refreshed on a place lifecycle, not only when mobility panels fluctuate. The operational difference is join quality: radius joins bleed sidewalk, bus-stop, and neighbor-tenant traffic into your "visit."
Measurement teams should document which geometry they used in every benchmark — switching from radius to polygon without relabeling history makes YoY reads un-auditable. For strip malls and downtown blocks, require polygon samples in the pilot; for isolated big-box sites, radius with an explicit confidence caveat may be acceptable. Pair geometry tests with foot-traffic panel sizing so you are not solving a panel problem that is actually a polygon problem.
Platform defaults hide the geometry decision until attribution breaks. When self-serve tools pre-draw 150-meter circles at every centroid, analytic teams inherit bias before the first dashboard loads. Procurement should require geometry metadata with visit counts: fence type, footprint version, and effective date. That metadata reconciles Q3 radius reads with Q4 polygon migrations without rewriting history. Vendors shipping POI & Geofencing should expose fence-type flags in schema docs so warehouse joins do not silently merge mixed geometries. Compare delivery specs against IAB data transparency when legal asks whether visit definitions are stable quarter to quarter.
Circles do not match reality. A 100-meter radius around a coffee shop on a dense block may capture the dry cleaner, the bus stop, and apartments across the street — every pause becomes a "visit." Internal and public methodology work consistently finds 25–45% false-positive reduction when migrating radius to polygon in urban and mixed-use environments; malls and transit hubs can be larger. The devices did not change; the join key did. That gap is why why POI data quality makes or breaks foot-traffic analytics treats polygons as load-bearing infrastructure.
False-positive economics compound in multi-touch attribution. A strip-mall radius fence inflating visits 35% does not only distort store KPIs — it poisons holdout design and media mix models using visit as a conversion proxy. Run a paired geometry test on ten hard sites: same panel, same date range, radius vs polygon, documented side by side. Publish the delta internally before scaling spend. Teams using cross-channel measurement should treat geometry as a model input, not a cartographic detail. When the delta exceeds your tolerance band, fix POI before renegotiating panel breadth — the MRC treats place definition as upstream of any visit metric you will defend in audit.
Radius is not wrong — it is narrower. Large outdoor venues, parks, and event campuses where a generous circle approximates the venue footprint are reasonable radius cases. Awareness campaigns optimizing reach over attributed visits can accept higher false positives if downstream KPIs are impressions, not store lift. Early-stage market scans asking "is there device activity here?" may use radius with documented uncertainty. When no trusted polygon exists, a scoped radius with a confidence flag beats a polygon you cannot validate. Geospatial data quality framework walks the full decision tree.
Document the use-case lane when you choose radius. Activation teams buying audience targeting for broad reach can accept noisier joins if conversion is not the KPI. Measurement teams running incrementality need polygon truth on the same sites — mixing lanes without labeling creates executive dashboards that look comparable but are not. Build a geometry policy matrix: asset type, campaign objective, acceptable false-positive band, and escalation path when pilot deltas exceed threshold. Revisit the matrix when you enter new geographies; suburban radius defaults that worked in Phoenix often fail on Boston mixed-use blocks without anyone changing a setting.
Defensible polygons combine parcel boundaries (where available), satellite or aerial validation, and operator correction for remodels, pad sites, and co-tenancy. The polygon should trace the indoor business footprint, not the full parcel — parking and loading docks follow different attribution rules. Refresh must track openings, closings, rebrands, and remodels; annual POI refreshes fail for weekly foot-traffic products. Census Business Dynamics Statistics and the NAICS 2022 manual help roll polygon-tagged visits to industry benchmarks consistently.
Procurement should request a change-delta file showing recent edits on your chain list, not only a static snapshot. Same-address disambiguation (multiple brands, one street number) is where cheap catalogs fail — see POI quality in depth for test patterns.
Use this diagnostic set in RFPs and pilots:
Answers matter more than headline POI counts. If a vendor cannot explain false-positive methodology, assume radius defaults and price the noise into your model. For production specs on polygon delivery, start at POI & Geofencing and validate on your densest corridors first.