This whitepaper provides an in-depth exploration of the technical methodologies and best practices behind Azira's location measurement platform. It details how Azira constructs representative measurement panels and well-matched control groups to isolate campaign impact, as well as the algorithms used to calculate key performance indicators like footfall lift and incremental visits. The paper also showcases Azira's sophisticated cross-media measurement capabilities, which allow for attribution of campaigns that span multiple channels and devices.

Introduction

A key aspect of Azira’s advertising platform is its advanced location measurement solution that enables advertisers to accurately assess the real-world impact of their digital advertising, out-of-home (OOH) media, and connected TV (CTV) campaigns on driving foot traffic to physical business locations. By processing billions of high-quality, privacy-compliant mobile location data points on a daily basis and leveraging a proprietary library of precise location boundary polygons, Azira offers the most accurate and comprehensive suite of location attribution metrics in the industry.

Azira's location data is of the highest quality and is well-suited for a wide range of applications. By leveraging advanced positioning technologies and fusing data from multiple sources, including GPS, Wi-Fi, and cellular networks, Azira ensures accurate and precise location information. The data undergoes rigorous quality  checks and filtering to remove noise, outliers, and inconsistencies. Azira's location data is frequently updated to reflect real-world changes, ensuring its recency and relevance. Moreover, the data is enriched with valuable attributes such as points of interest, road networks, and building footprints, providing a comprehensive and context-rich representation of locations. With its superior accuracy, consistency, and completeness, Azira's location data empowers businesses and organizations to attract their best customers, make informed decisions, and deliver location-based services with confidence.

This whitepaper provides an in-depth exploration of the technical methodologies and best practices behind Azira's location  measurement platform. It details how Azira constructs representative measurement panels and well-matched control groups to isolate campaign impact, as well as the algorithms used to calculate key performance indicators like footfall lift and incremental visits. The paper also showcases Azira's sophisticated cross-media measurement capabilities,which allow for attribution of campaigns that span multiple channels and devices.

Data Inputs and Quality Assurance

Discover how Azira provides advertisers with a reliable foundation for measuring campaign impact.

  • By leveraging a diverse range of collecting high-quality, diverse location data and applying rigorous cleansing and verification processes, Azira provides advertisers with a reliable and representative foundation for attribution to provide advertisers with a robust and representative foundation to execute and measure the impact of their campaigns

Diverse Location- Data Sourcing

One of the key strengths of Azira's attribution platform is the diversity and scale of its location data inputs. Rather than relying on a single provider or panel, Azira takes a multi-source approach to collecting high-precision, high-frequency device level movement data.

The two main inputs are:

  • Second-party location SDK partnerships: Azira has built direct integrations with several leading location data SDK providers, allowing for aggregation of opt-in, application-derived data.
  • Bidstream data from Azira's advertising DSP: As a buyer on major mobile programmatic exchanges, Azira is able to collect location signals from the bid requests of over 100,000 apps for the purpose of serving geotargeted ads.

This diversity of inputs allows Azira to extract the highest-quality, most stable and representative data from each source, while mitigating the limitations and potential biases of any individual provider. Compared to single-source location platforms, Azira is able to provide a much larger opted-in measurement panel, with greater reach across demographics, geographies, and app categories. This ensures the utmost reliability and generalization of attribution insights.

Ensuring Attribution Accuracy

Achieving a high degree of accuracy and confidence in location-based attribution is not simply a matter of data scale, but also data quality and methodological rigor. Azira focuses on four key pillars to provide trustworthy, defensible visitation and lift metrics:

  • Data Confidence: Azira takes a "quality over quantity" approach to location data ingestion and processing. Before being used in attribution reporting, all data points undergo rigorous cleansing, normalization, and scoring to ensure utmost precision and consistency. Questionable or spurious signals are aggressively filtered out. Azira's data science team continually updates these machine learning-based quality assurance models by validating against a large set of ground truth data. Only the highest-scoring location signals are used in any analysis.
  • Observational Confidence: The reliability of any campaign impact analysis depends on capturing as much of the relevant devices' real-world movement patterns as possible during the reporting period. If too many true visits to the measured locations go unobserved due to gaps in the location data, lift estimates can be severely understated. To maximize observational confidence, Azira analyzes each device's overall location activity level and temporal coverage, assigning a confidence score based on the density, frequency, and hours of the day for which data is available. Only devices that meet strict activity thresholds are included in the final measurement panel.
  • Visit Confidence: Accuracy in attribution also requires a high degree of certainty that any observed visits truly occurred within the confines of the measured locations and not at adjacent or nearby places. To provide this spatial precision, Azira employs an extensive library of manually drawn location boundary polygons, created using high-resolution satellite imagery and on-the-ground point-of-interest data. This polygon-based approach offers much greater reliability than commonly used methods like point-in-radius, contour maps, or user check-in data, which can introduce significant noise and false positives into visitation data.
  • Statistical Significance: As with any data-driven analysis, Azira ensures that all reported insights meet rigorous standards of statistical confidence. All baseline and lift metrics are reported with 99% confidence intervals, and extrapolations are only performed when the underlying visit counts and panel sizes are sufficiently large as to be stable and projectible. Azira's reporting framework is grounded in established statistical techniques and experimentally validated against a range of campaign designs.

Measurement Panel Construction

Learn how Azira constructs customized measurement panels and control groups that enable accurate, reliable assessment of a campaign's real-world impact.

  • With Azira, you can construct customized measurement panels and control groups that enable accurate, reliable assessment of a media campaign's impact on real-world foot traffic. By carefully selecting a representative sample of devices based on factors like geography, data quality, and visit density, Azira ensures that the exposed group provides a statistically robust picture of campaign reach and visitation. Furthermore, by leveraging advanced identity resolution and data science techniques to build tightly matched control groups, Azira provides a credible baseline for isolating the incremental impact of media exposure.

Optimizing Panel Power and Representation

The foundation of any successful attribution study is a well-constructed measurement panel of devices against which the media campaign's location impact will be assessed.

For each study, Azira dynamically builds a custom panel that is optimized for:

  • Maximizing the overall volume and density of devices seen in the campaign locations
  • Ensuring adequate representation of devices across key targeted geographies and audience segments
  • Achieving the highest possible "observational confidence", based on each device's overall location activity level during the reporting period

To construct this panel, Azira first generates a large pool of candidate devices that were seen in the campaign locations during a several-week pre-study period. This ensures adequate foot traffic density to measure against. From this pool, devices are strategically sampled based on their geographic spread (to ensure regional representation in line with the media targeting), as well as their individual quality scores for location precision and activity frequency (to maximize the probability of a visit being captured).

By favoring more active devices, this approach does introduce a bias towards the heaviest users. However, Azira's data science team has determined through careful testing that this is more than offset by the increased statistical power and ability to detect true campaign lift. Trying to measure campaign impact on a panel of devices for which little location data is available leads to a high rate of false negatives, and systematic understatement of attribution results.

Control Group Formation and Matching

Another critical aspect of accurate attribution is the quality of the control group against which the visitation patterns of the exposed group are compared.

Azira offers two options for advertisers:

  • Randomized PSA-based control: This is the default approach, whereby the advertiser allows a portion of the campaign impressions (typically 10-15%) to be randomly served public service content in lieu of the branded ad. This PSA-exposed group forms a natural control that is perfectly matched to the target audience by virtue of the randomization.
  • Synthetic control: In cases where PSA-based controls are not used, Azira leverages its proprietary identity graph to programmatically generate a synthetic control group that mirrors the exposed group across key geographic, demographic, and behavioral dimensions. The goal is to reproduce the visitation patterns the exposed group would have exhibited had they not been exposed to the media. Key variables used in the identity graph-based matching include:
  • Geographic: Using county-level or ZIP code boundaries, control devices are selected to match the regional delivery skew of the exposed group impressions. Further weighting is applied based on the Census block group to ensure the control is geospatially representative.
  • Behavioral: Each device is assigned a rich set of behavioral attributes based on its
    observed pattern of visitation to key pointsof interest. These include measures of its affinities for different retail, dining, and recreational categories. Matching on these variables ensures that the control group not only looks like the exposed group, but also acts like it, with similar lifestyles and purchase habits. This is critical for teasing out the media's impact from the natural, pre-existing differences in visitation propensities
  • Device Location Activity: Perhaps the most important variable to control for is each device's overall location-sharing frequency and temporal coverage during the campaign reporting period. As discussed earlier, high-activity devices are more likely to be served an impression (since they generate more ad requests) and also more likely to be directly observed making a visit to the target locations (since their movements are more completely captured). Matching the two groups on this variable eliminates the potential for bias and ensures an apples-to-apples comparison.

Visit Rate and Lift Calculations 

Understand how Azira calculates key performance metrics to quantify the impact of campaigns on driving visits to physical locations.

  • Azira calculates the key performance metrics that advertisers use to quantify the impact of their campaigns on driving visits to physical locations. By comparing the observed and extrapolated visit rates of devices exposed to an advertisement to those of a matched control group, Azira can measure the incremental lift in visitation attributable to the campaign.

Counting Observed Visits 

Once the exposed and control groups have been defined, the core of the attribution analysis is a comparison of each group's visitation rates to the campaign destinations. The most straightforward calculation is a count of directly observed visits. Using Azira's precise location boundary polygons, each group is monitored for instances where their devices were detected within the defined geocoordinates of the target locations.

To be considered valid and attributable to the campaign, visits must occur within a specified time window following an ad exposure. Typically, advertisers use a 7-day attribution window for high-frequency, low-consideration categories like Quick Service Restaurants (QSRs), and a 14-30 day window for less frequent, higher-consideration purchases like auto dealerships. Azira's platform allows for fully custom windows by campaign objective.

However, due to the inherent sparsity of opted-in location data, and the fact that most consumers do not keep location sharing active at all times, there will always be some number of visits that go uncaptured by this direct observation method. This is especially true for locations that consumers tend to visit for very short durations (e.g. picking up a coffee) or where they are less likely to use their phones (e.g. a movie theater). Therefore, to get a more complete picture of a campaign's impact, Azira employs a proprietary visit data extrapolation model.

Extrapolating Total Visits 

Azira's data science team has built a machine-learning model to estimate the total count of visits to a campaign location, based on the directly observed visits. The model is grounded in well-established statistical methods, but has been extensively customized and trained for the location data use case. At its core, the model fits a Poisson process¹ to each device's observed location updates to calculate:

  • The device's individualized probability of detection within the location, based on factors like its average location-sharing frequency, the inferred visit dwell time, and the type of venue
  • The likely number of unseen devices that visited the location, but whose visits went unobserved based on their detection probabilities

To make these determinations, Azira first profiles each device seen in the campaign locations, assigning it an average "location activity" score that captures the density and frequency with which it was detected during the attribution window. This is calculated by bucketing devices based on their number of recorded data points, and calculating the probability of each bucket based on the overall population density. For instance, if devices that pinged between 100-200 times per day made up 10% of the total visitor pool, any individual device falling into that category would be assigned a 10% probability score.

This location activity score is then combined with the confidence and dwell time scores to determine the final detection probability for the device. Dwell time is defined as the average time spent by visitors at the location type, and is calculated based on Azira's vast historical dataset of hundreds of billions of visits to millions of commercial points of interest.

Confidence is a measure of visit certainty, taking into account factors like the density, geometry, and positioning of data points within the polygon boundaries.

These individual probabilities are then used to infer the total number of visits, including those that were not directly observed, using the following formula:

`Total Visits = Observed Visits / P(Detection)`

Conceptually, this extrapolation credits each observed device with representing `1 / P(Detection)` total devices, since the lower the probability of detection, the more real-world devices each observed device is assumed to represent. For instance, an observed visit from a device with a 10% detection probability would be extrapolated to 10 total visits (1 / 0.10 = 10). These extrapolations are summed across all observed devices to estimate the total visit count.

To ensure statistical stability, visit extrapolation is only performed when the sample size of observed visits is sufficiently large, typically greater than 1,000. Azira also employs Bayesian priors² and caps the extrapolation multiplier at a maximum of 20x to prevent over-inflation. 

The accuracy of this method has been validated by comparing extrapolated estimates to ground truth sales data and foot traffic counters for a range of business types and campaign designs. Across millions of campaigns measured, Azira has observed a median absolute percent error of 10% between extrapolated and actualized visits, and a correlation of 0.85.

This approach provides advertisers with an understanding of the full reach and impact of their campaigns, not just the fraction that can be directly observed. The model also allows for apples-to-apples comparison between exposed and control groups by accounting for any systematic differences in their location-sharing patterns.

Calculating Lift Metrics

With the observed and extrapolated visit counts for each group in hand, calculating the core campaign lift metrics is straightforward:

Visitation Rate = Total Visits / Total Devices Lift = (Exposed Visitation Rate - Control Visitation
Rate) / Control Visitation Rate

Incremental Visits = Exposed Group Lift * Exposed Group Devices

These lift metrics are calculated at the overall campaign level, and can also be broken out by key dimensions like geography, publisher, ad format, frequency, demographics, or audience behavioral segments. To ensure statistical reliability, Azira only reports lift for breakouts that meet a minimum sample size requirement, typically 100,000 exposed devices and 1,000 observed visits.

All lift metrics are reported with 99% confidence intervals, calculated using the standard Gaussian³ formula based on the visit counts and sample sizes of each group. Azira also calculates statistical significance using a two-tailed t-test4 to determine if the difference between the exposed and control visitation rates is distinguishable from random chance.

Cross-Media Measurement

Discover how Azira's cross-media measurement solution combines ad exposure data from various channels to provide a unified view of their total and incremental impact on driving foot traffic.

  • With the growing fragmentation of digital media consumption across devices, publishers, and formats, advertisers increasingly need a unified view of how these channels work in combination to influence consumer behavior. This has given rise to substantial demand for cross-media measurement solutions that can attribute conversions and estimate the marginal lift of each channel.

Azira offers a sophisticated cross-media measurement solution that combines ad exposure data from in-app, web, CTV, and OOH campaigns to measure their total and incremental impact on driving foot traffic. While the measurement is straightforward for in-app impressions (where a common identifier can bridge ad exposures to location visits at the device-ID level), it becomes more challenging for web, CTV and OOH impressions where cookies and device IDs are not reliably available

Probabilistic Attribution for Non-In-App Channels

To connect ad exposures to visits for cookie-less impressions, Azira has built a probabilistic attribution model that leverages the impressions' IP addresses and the location data's network connection information. The core concept is to estimate the probability that each device observed in the campaign locations was exposed to an impression, based on the number of impressions delivered to the network(s) that the device was seen connecting through.

The process works as follows:

  • Divide the web/CTV/OOH impressions into discrete time intervals (e.g. 15 minutes).
  • For each interval, get the count of impressions and distinct IP addresses reached.
  • For any device observed connecting to a campaign-impressed IP address during that interval, assign it a probability of exposure based on:

    P(Impression) = Impressions on IP /Distinct Devices on IP`

  • For example, if an IP address had five impressions delivered and we observed 10 distinct devices connecting to it in that 15 minute span, each device would receive a 50% probability of having been exposed.
  • Sum these probabilities across all time intervals and all networks to which the device is connected.
  • For any device that visited a campaign location, estimate the probability that it was influenced by the campaign by taking the union of its impression probabilities.

    `P(Influenced Visit) = 1 - ∏ [1 - P(Impression)]`

  • This calculates the complement of the probability that the device received none of the impressions, i.e. the probability that it received at least one impression.
  • Sum these fractional probabilities across all visitors to estimate the total count of visits influenced by the campaign

This approach effectively leverages the fact that devices tend to connect primarily to a small number of networks over time, and that those networks vary significantly in size (from home WiFi routers with a handful of devices to cellular towers and public WiFi networks with thousands). The more impressions served to the network, and the smaller the network, the higher the probability that any individual device was reached.

Azira has carefully tested and validated this model by comparing the results to a truth set of campaigns where all channels used deterministic device-ID tracking. Across hundreds of campaigns, the probabilistically attributed incremental visits showed a median absolute percent error of 15% compared to the fully deterministic approach. While not perfect, this provides advertisers with a highly reliable way to include the vast majority of digital ad exposures that happen outside of in-app environments.

Reporting Results by Partner

For multi-partner attribution studies, Azira reports the results at two levels: overall campaign impact and individual partner contribution. The overall campaign report calculates visitation lift using all ad exposures across all partners. This provides a holistic view of the campaign's impact.

To calculate each partner's individual contribution, Azira re-runs the analysis using only that partner's ad exposures, providing a comparative view of how each partner performed in driving visits. However, it's important to note that these individual partner estimates are not additive, and will not sum to the overall campaign total.

Azira employs a probabilistic attribution methodology for non-in-app exposures. For example, consider a campaign with two partners, each of which served 1,000 impressions to a particular network, and a device that was subsequently seen visiting the advertised location after connecting to that network. The overall campaign analysis would assign that visit a 100% probability of being influenced (since the device was exposed to 2,000 total impressions). However, the partner-level analyses would each assign a 50% probability (1,000 / 2,000) to the visit. Summing these partner-level probabilities would double-count that device's visit.

Therefore, Azira recommends focusing on each partner's individual lift metrics (e.g. visitation rate, incremental visits per thousand impressions) to assess their relative performance, rather than summing the incremental visits across partners. The partner with the highest lift should be considered the most efficient at driving visitorship, independent of the absolute scale of impressions delivered

Measuring Incremental Impact by Partner

To quantify each partner's incremental impact in a multi-partner campaign, Azira calculates two additional metrics:

  • Exclusive Visits: These are visits that can be attributed to only that partner because the device was not exposed to any other partner's impressions. These visits provide a clean read on the partner's ability to drive incremental traffic, without any potential duplication from other partners. To calculate exclusive visits, Azira first identifies the subset of devices that were exposed to that partner and that partner alone. It then calculates the visitation rate and lift for this exclusive exposed group compared to the overall control group. This lift is then multiplied by the total number of exclusive exposures to arrive at the number of exclusive visits driven by that partner.
  • Marginal Visits: These are visits that occurred above and beyond what would have been expected based on the other partners' media alone. In other words, they are the additional visits that can be attributed to the partner's participation in the campaign, controlling for the impact of other partners.

To calculate marginal visits, Azira first runs a visitation analysis for all other partners' media excluding the partner in question. This establishes a baseline "rest of campaign" visitation rate. Azira then runs a second analysis for the full campaign including the partner. The difference in visits between these two scenarios is the marginal impact of that partner.

Importantly, marginal visits can be negative, indicating that the partner's impressions were actually less effective at driving visitation than the rest of the campaign. This could be due to suboptimal targeting, creative, or frequency.

By calculating both exclusive and marginal metrics for each partner, advertisers get a comprehensive view of how each partner contributed to the overall campaign impact, and how much value they added above and beyond the other partners. This enables more strategic decisioning about partner selection and budget allocation for future campaigns.

Actionable Insights and Optimization

Learn how Azira's attribution solution delivers actionable insights for campaign optimization, helping advertisers uncover key performance drivers and make data-driven decisions.

Beyond providing trusted measurement of bottom-funnel business outcomes, Azira's attribution solution is built to deliver actionable insights for campaign optimization. Through its self-serve reporting platform and expert client analytics support, Azira helps advertisers answer key questions like

  • Which audience segments are most responsive to my message and most likely to visit after an ad exposure?
  • Which publisher partners and tactics are most efficient at driving visitorship at different points in the consumer journey?
  • What is the optimal frequency range to maximize visitation while minimizing wasted impressions?
  • How does ad engagement (e.g. click rates) correlate with store visits, and what is the true "cost per visit" of my campaign?

To uncover these insights, Azira's platform enables real-time slicing and dicing of attribution metrics across a wide range of dimensions, including:

  • Audience: Using both Azira's proprietary location-based behavioral segments as well as first-party customer data and third-party audience data, advertisers can compare attribution rates across consumer types to identify high-performing audiences for targeting.
  • Geography: Azira's reporting interface allows for attribution analysis at any geographic level, from state and DMA down to county, ZIP code, and even custom trade areas. This enables advertisers to quickly spot geographic pockets of efficiency or struggle.
  • Publishers and Tactics: Results can be broken out by publisher, ad format, device, creative size and format, and other tactical variables to identify top-performing partners and placements. For each partner, metrics like visits per thousand impressions and cost per incremental visit provide a cross-comparable view of media efficiency.
  • Frequency: Azira's platform provides breakouts of visitation rate and lift by impression frequency range, illustrating where frequency caps should be set to optimize efficiency. Frequency analysis can also be helpful for determining the optimal flight duration and spacing between messages. These insights can be pushed directly to DSPs and publisher platforms for in-flight optimization through Azira's API integrations. Azira also offers strategic consulting services to help advertisers interpret results and craft go-forward media plans based on the attribution learnings.

By linking media tactics to real-world results, Azira empowers advertisers to confidently shift spend to the partners, platforms, and audiences that are truly moving the needle on business objectives.
Over time, this test-and-learn approach drives substantial improvements in return on ad spend and overall marketing effectiveness

Conclusion

Location data is an incredibly powerful tool for understanding advertising's impact on consumer behavior. But deriving accurate, actionable insights from this data requires a thoughtful approach to quality control, bias correction, and methodological rigor. Azira's industry-leading attribution platform is built on a foundation of:

  • High-quality data inputs from a diversity of opt-in sources
  • Rigorous data cleansing and filtering to ensure accuracy and precision
  • Carefully constructed measurement panels and control groups to isolate incrementally
  • Proprietary visit confidence scoring and extrapolation models to provide a complete picture of campaign reach
  • Sophisticated multi-touch attribution algorithms to assign fair credit across partners
  • An intuitive, visual reporting interface to surface optimization insights

By combining these elements, Azira delivers foot traffic measurement and attribution that advertisers can truly rely on to guide their marketing decisions. To learn more about how Azira can help you unlock the power of location analytics, contact us today at info@azira.com.

Learn how Azira's attribution solution delivers actionable insights for campaign optimization, helping advertisers uncover key performance drivers and make data-driven decisions.

Location data is an incredibly powerful tool for understanding advertising's impact on consumer behavior. But deriving accurate, actionable insights from this data requires a thoughtful approach to quality control, bias correction, and methodological rigor.