Key Takeaways
- POI data complements audience data by showing real-world behavior, not just demographics or intent. Combining both creates stronger, higher-intent audience segments.
- Building footprint polygons provide the most accurate store visit attribution. Radius-based geofences often create attribution errors, especially in dense areas.
- POI data quality directly impacts AdTech performance. Inaccurate or outdated data can affect audience targeting, campaign execution, and attribution.
- Geo-contextual targeting offers a privacy-friendly alternative to cookie-based targeting by using offline location behavior as a durable signal.
- Store visit attribution connects digital campaigns to real-world outcomes, helping marketers measure and justify advertising spend with greater confidence.
Programmatic advertising has gotten remarkably precise in how it targets users by demographics, behavior, and intent. But there is a persistent gap that audience data alone cannot close: understanding what people do in the physical world before, after, and between digital touchpoints.
That gap is where POI data earns its value.
This guide is written for AdTech practitioners, DSP teams, and media strategists who are already working with programmatic infrastructure and want a concrete framework for integrating location-based data into campaign planning, audience segmentation, geo-targeting, and online-to-offline attribution. We will not spend time on basics like “what is programmatic advertising.” What we will go deep on is how POI data, building footprints, and geofencing work in practice, and where the common pitfalls are.
Why Audience Data Has a Physical World Blind Spot
Audience-based data is excellent at telling you who someone is: their age bracket, household income estimate, interest categories, and browsing behaviors. It is less equipped to tell you what that person actually does when they put down their phone and go about their day.
Location-based marketing fills that gap. When you know that a device regularly pings at specialty running stores, you have a much stronger signal about an active buyer than if you only know they visited a sporting goods brand’s website. The first signal reflects real-world behavior. The second might be aspiration, curiosity, or a single browsing session.
This distinction matters enormously for targeting precision in programmatic campaigns. Layering POI data onto audience segments lets AdTech teams move from descriptive profiles to behavioral ones, built on where people actually spend time and money in the physical world. As of 2026, approximately 75% of marketers in North America have adopted location-based strategies to increase customer engagement, a signal that the practice has moved well past early adoption into standard operating procedure for serious performance marketing teams.
The signal quality from location data has also improved significantly over the past five years. The expansion of GPS-capable devices, combined with better data hygiene practices among data providers, means that modern POI datasets are substantially more reliable than the location data that circulated in the early programmatic era.
What POI Data Actually Is (and What It Is Not)
Before integrating location-based data into a campaign, it helps to be precise about what POI data is. “Location data” is a broad term that gets applied to several distinct things: raw GPS pings or device-level signals, point of interest records describing physical places, building footprint polygons defining the physical boundaries of structures, and behavioral movement data inferred from those signals.
SafeGraph’s products cover two of those categories directly.
SafeGraph Places (POI Data)
SafeGraph Places is a global dataset of point of interest records covering the places where people spend time and money. Each record includes the place name, street address, geocoordinates, category, brand affiliation where applicable, and other attributes relevant to geospatial analysis and audience segmentation.
For AdTech applications, POI records are what allow a DSP or data platform to make sense of a raw GPS coordinate. A latitude/longitude pair from a device ping is not inherently meaningful. When you can join that ping to a confirmed POI record, you know the device was at a specific coffee shop, gym, car dealership, or retail location. That context is what powers behavioral audience building.
The Places dataset is updated monthly, which matters for campaign accuracy. A POI database that is six or twelve months stale will attribute visits to businesses that have closed, moved, or rebranded, introducing noise into any audience or attribution model built on top of it.
SafeGraph Geometry (Building Footprints)
SafeGraph Geometry provides precise building footprint polygons for points of interest. This is the dataset that determines not just whether a device was “near” a location but whether it was actually inside the physical structure of that location.
For attribution work, the difference is significant. A centroid-based geofence drawn around a point coordinate will often bleed into adjacent businesses, especially in dense retail environments like malls, strip centers, or urban corridors. A precise building footprint polygon eliminates that bleed, ensuring visits are attributed to the correct POI rather than a neighboring one.
Read more about how building footprints support accurate geospatial analysis in SafeGraph’s blog post: Building Footprints: Essential Data for Accurate Geospatial Analysis.
How Location-Based Data Strengthens Audience Segmentation
The most immediate application of POI data in AdTech is audience building. Rather than relying entirely on modeled demographic segments or online behavioral data, teams can construct segments anchored in real-world visit behaviors.
A few patterns that work well in practice:
Category-based audience construction.
By identifying devices that have pinged at POIs within a specific category (QSR chains, luxury automotive dealerships, home improvement retailers), teams can build behavioral audience segments with high purchase-intent signals. The key is having a POI database with accurate category taxonomy so that segment definitions are consistent and defensible.
Competitive conquesting.
POI data enables brands to target consumers who have been observed visiting competitor locations. This requires a reliable, brand-level POI dataset. Generic or open-source POI databases often have significant gaps at the brand or chain level, which makes competitive conquesting segments unreliable. SafeGraph Places covers brand affiliation explicitly in its schema, making this use case directly actionable.
Cross-category behavioral profiling.
Understanding what other places a consumer visits alongside a target POI adds behavioral depth that no demographic model can match. A device that regularly appears at high-end grocery stores, yoga studios, and specialty wine retailers tells a more precise story about purchasing behavior than a household income quintile alone.
Case Study: How RainBarrel Uses POI Data to Build Behavioral Audience Segments
RainBarrel, a Canadian AdTech company, built its entire audience graph around this principle. Rather than relying on browser-based behavioral data, RainBarrel uses SafeGraph Places to anchor its 2,600+ audience segments in documented offline behaviors. A consumer who visits a Nordstrom store at least twice a month is treated as a confirmed active luxury shopper, not an aspirational one. The platform is compatible with leading DSPs and designed to activate across digital advertising channels, all without relying on personal identifiers.
POI data quality was a specific requirement, not a nice-to-have.
“When it comes to data, Canada is an underserved market,” noted David Choi, RainBarrel’s Product Manager.
“Most data providers are based in the U.S. and treat Canada like an afterthought.”
SafeGraph’s global coverage addressed that gap directly, giving RainBarrel the foundation to build audiences at national scale in a market where reliable POI data had previously been hard to source.
How to Use Geofencing and POI Data for Geo-Targeting
Geofencing in AdTech refers to defining a geographic perimeter around a location or set of locations and using that perimeter to trigger ad delivery, capture visit signals, or segment audiences. POI data and building footprint geometry are what determine how precise and effective those fences are.
There are two primary modes of geofencing in practice.
Point-radius geofences draw a circular boundary around a coordinate at a fixed distance. These are easy to implement but have significant limitations in dense environments where multiple businesses cluster together. A 50-meter radius around a restaurant centroid in an urban center might overlap with three competing establishments next door.
Polygon-based geofences use the precise shape of a building footprint to define the boundary. This is materially more accurate because the fence corresponds to the actual physical structure rather than an arbitrary radius. For any attribution work where visit credit matters, polygon-based geofencing is the standard that serious AdTech platforms should be using.
The practical implication: if your geo-targeting or attribution work uses point-radius fences built on POI centroids, you are systematically introducing errors that scale with the density of the environments you are working in. Building footprint polygons is correct for this, and the fix is straightforward when you have access to a Geometry dataset that pairs directly with your POI records.
Learn more about building a geofencing strategy in SafeGraph’s guide: A Complete Guide to Geofencing Data.
Case Study: How Mobsta Uses POI Data to Power Geo-Contextual Campaign Planning
UK-based location marketing specialist Mobsta built its TraffiQ platform on SafeGraph Places data to help ad agencies plan geo-contextual campaigns with real-world accuracy. The platform lets agency teams identify and visualize precisely where target audiences are concentrating, what kinds of places they frequent, and which ad inventory locations are best positioned to reach them.
“We need to have a solid understanding of where places are and how people are moving from one place to another,” said James Sexton-Barrow, Mobsta’s Head of Planning. “Without the right data to fuel those insights, we wouldn’t have any leg to stand on.”
The monthly refresh cadence of SafeGraph Places was a specific operational requirement.
“Receiving SafeGraph data each month in an S3 bucket means the POI data in our platform can automatically refresh each month with minimal development required,” explained Jack Burton, Mobsta’s Head of Product. The result was faster agency sales cycles and enough platform confidence that clients now routinely run the TraffiQ tool themselves during planning.
Closing the Online-to-Offline Attribution Gap
Online-to-offline attribution is one of the most operationally difficult problems in performance marketing. A user sees a display ad on Tuesday, visits a physical store on Thursday, and makes a purchase at the register on Saturday. Connecting those events into a single journey requires precise location data at every step.
When store visit attribution is working well, it validates the spend efficiency of location-targeted campaigns and gives planners the confidence to bid more competitively in the programmatic auction for impressions that demonstrably drive in-store behavior. When it is working poorly, it overstates or understates campaign impact, leading to misallocated budgets.
POI data and building footprints address the physical-world leg of that attribution chain. If you can confirm, with precision, that a device was inside a specific store after being served an ad, you have a credible store visit conversion event.
Want to see how building footprints and POI data work together in attribution workflows?
Test SafeGraph Geometry in your own environment before your next campaign build.
Building a Store Visit Attribution Model: The Right Approach
AdTech firms looking to offer best-in-class visit attribution typically build their own models rather than relying on packaged solutions. SafeGraph Places and Geometry support that approach directly. Here is a breakdown of the four-step process SafeGraph’s own attribution methodology follows, and what to watch for at each stage.
Step 1: Clean the GPS data.
Raw GPS signals require preprocessing before they are usable for attribution. Common issues include GPS drift, spiky horizontal accuracy readings, and jumpy pings that appear to teleport a device between distant coordinates in seconds. Cleaning this data before any spatial analysis prevents systematic errors downstream.
Step 2: Cluster GPS pings into visit candidates.
The goal is to take a stream of individual pings and group them into coherent visit events. Pings that cluster in space and time represent a dwell event at a location. Scattered or transient pings represent movement through an area. Proper clustering separates signal from noise.
Step 3: Spatially join clusters to POI polygons.
Once clusters are identified, a geospatial join matches each cluster to the POI polygons in the Geometry dataset, creating a candidate list of places the cluster could be associated with. In dense environments like malls or urban retail corridors, a single cluster may overlap with multiple candidate POIs, which is why Step 4 matters.
Step 4: Classify the best-fit POI using machine learning.
With multiple candidate matches possible, a classification model incorporates additional variables (time of day, dwell duration, device behavior patterns, spatial overlap percentage) to identify the single best POI match. Machine learning at this stage significantly outperforms rule-based approaches, especially for small-footprint POIs nested within larger structures.
See the complete technical breakdown in SafeGraph’s Technical Guide to Visit Attribution.
Attribution Methods to Avoid
Most AdTech teams evaluating an attribution build will consider simpler approaches before investing in the full ML-based pipeline. Here are three common shortcuts and why they underperform.
Closest centroid wins : This approach assigns a visit to whatever POI centroid is geographically closest to a GPS ping, provided it falls within a threshold distance. It works reasonably well for large, standalone stores. It fails for smaller footprints and anywhere businesses cluster together in shared structures, because the “closest centroid” often belongs to an adjacent business rather than the one the device actually entered.
Any ping inside a polygon : This method flags a visit whenever a GPS ping falls inside a POI’s polygon boundary. GPS drift means a signal can wander outside the physical building even when the device is stationary inside. It also struggles with shared structures, where a device might appear to bounce between two adjacent POI polygons due to signal noise.
Padded custom geofences : Adding a buffer around a POI polygon to account for GPS drift reduces missed visits but introduces a different problem: the padded zone overlaps neighboring businesses, increasing false positive attributions. The fundamental issue is that the fence geometry is imprecise, and precision has to be bought back with more complex processing.
The lesson across all three: simpler attribution geometry requires more complex compensatory logic. The ML-based approach with precise building footprints produces cleaner output that you can actually defend to clients.
Not All Location Data Is Created Equal: A Practitioner’s Evaluation Framework
POI data quality varies substantially across providers, and the difference is not always visible until you start running analysis. Before committing to any location dataset for production use in AdTech, apply these four evaluation criteria.
First, understand the sourcing pipeline: what primary sources feed the dataset, how frequently it updates, and what quality controls govern ingestion. Second, assess coverage completeness relative to your specific use cases. A dataset that looks strong in aggregate may have significant gaps at the category or brand level that only surface when you query for specific chains or venue types. Third, evaluate immediate usability. Will the data require substantial cleaning or normalization before it can join to your existing infrastructure? Fourth, verify schema consistency across monthly refreshes. Unexpected field changes in a monthly update can break downstream pipelines without warning.
For a structured version of this evaluation, see SafeGraph’s Data Evaluation Checklist.
Closing Thoughts
Location-based marketing has moved from a targeting tactic into something closer to core infrastructure for AdTech teams that take offline consumer behavior seriously. The teams getting the most out of it are not treating POI data as a layer they bolt onto existing campaigns. They are building it into audience construction, campaign planning, and attribution from the start.
The common thread across the best implementations in this guide is that data quality is the deciding variable. Precise building footprints, well-sourced POI records, and consistent monthly updates are not premium features. They are the baseline requirements for any attribution or segmentation model that has to hold up under scrutiny. Get those foundations right, and the rest of the stack gets easier to build on.
AdTech teams working on audience building, geo-targeting, or store visit attribution need POI data they can rely on in production. SafeGraph Places and Geometry are updated monthly, well-documented, and structured for direct integration into the platforms and pipelines AdTech practitioners already use.
FAQ’s
1. What is location-based marketing and how does it work in programmatic advertising?
Location-based marketing uses real-world location signals to improve audience targeting and campaign measurement. In programmatic advertising, advertisers use POI data and geofencing to build audience segments based on visit behavior, activate those audiences across advertising platforms, and measure store visits after ad exposure.
2. What is the difference between POI data and foot traffic data?
POI data describes physical places, including their location, category, and brand information. Foot traffic data measures how many people visit those places over time. POI data provides the geographic context, while foot traffic data provides the behavioral activity associated with those locations.
3. How does geofencing differ from geo-targeting?
Geo-targeting serves ads within a broad geographic area such as a city, ZIP code, or region. Geofencing creates a specific boundary around a location and triggers advertising actions based on movement into or out of that area. Building footprint geofences generally provide greater accuracy than simple radius-based geofences.
4. What SafeGraph datasets are most relevant for AdTech use cases?
SafeGraph Places provides POI data, including business names, categories, brand affiliations, and coordinates. SafeGraph Geometry provides building footprint polygons that support precise geofencing, audience creation, and store visit attribution. Together, they form the foundation for many location-based advertising workflows.
5. How should AdTech teams evaluate POI data quality before deployment?
Teams should assess data freshness, coverage, schema consistency, and ease of integration. High-quality POI data reduces targeting and attribution errors, improves campaign performance, and minimizes the operational work required to maintain location-based advertising systems.