PiinPoint uses Mobile Location Data (MLD) both within its software platform and in its predictive modelling. A recommended approach to interpreting results for MLD in the context of custom projects as well as in-app use, is to have a strong knowledge of the dataset and its best practices.
This guide describes how PiinPoint uses MLD to provide context and understanding to our customers on the dataset:
Mobile Location Data has taken the real estate and analytics space by storm in the last half-decade and become one of the most important datasets for the retail industry. Given that it is rooted in and adapts to up-to-date human movement data, its addition to the retail geospatial data toolbox has been both exciting and important.
As there has been increasing pressure to both acquire, distribute, purchase or apply MLD, an increasing level of confusion and distrust around the data has grown regarding its accuracy, limitations, and most appropriate applications. When interpreted without appropriate context and understanding, it can easily lose trust of the business user. There have been limited attempts at building credibility around MLD across the industry, as many do not fully grasp the gaps it fills and the potential pitfalls of its misapplication. At the heart of a solution based on trust and transparency, PiinPoint strongly believes that user education on MLD is critical.
A Sample of the Population
Despite some of the confusion and mistrust which has been associated with MLD, the dataset does indeed have great value potential. What is perhaps most important to consider when using MLD is the fact that it is a sampling of the population within a given area, which gets extrapolated to define a general understanding of many.
Many other key datasets in geospatial analysis follow a similar application. For example, the industry has long accepted in the past that a blanket radius or drive times polygon can represent a retail trade area. This inadvertently assumes that every single person in that boundary is their customer, when that is not the case. The people within that boundary are simply a sampling of their customer, and a broad one at that. There is a wide margin of error. MLD is also only a sampling of the population, where there will always be a margin of error. However, it remains so much more useful than simple trade areas to understand a retail business. Not only does it narrow in on areas surrounding a location to understand visitors, it also is an increasingly growing and improving dataset.
Before diving into the appropriate and powerful applications of MLD, let’s understand how it works...
How MLD Works
MLD is a big data set that looks at the movement of humans using GPS-enabled technology. Smartphone users leveraging location-based apps consent to the tracking of their location. Those location-based apps collect, aggregate, and distribute the location data to third-party data MLD providers, like NEAR (formerly Ubermedia), Placer.ai, Cubiq, Pelmorex, Veraset, and SafeGraph, among other providers.
Vast amounts of data are delivered to PiinPoint daily, and updated into the PiinPoint application for study on a monthly basis. The monthly cadence of updates in PiinPoint ensures that PiinPoint can account for lagging observation delivery, which means that additional observations continue to be added to the main dataset for up to 5 days after delivery to PiinPoint.
PiinPoint has four main applications for MLD, as outlined later in this Guide, including Traffic Volumes, Visitor Reporting, Predictive Modelling, and Custom Mobile Data Pulls.
PiinPoint’s methods for managing the data and producing insights on MLD include multiple data filtering and cleaning steps to make sure our outputs are accurate. PiinPoint filters out observations to detect and ensure things such as:
Privacy protection of consumers,
Accuracy and precision in the GPS observation results (e.g. removing points with an accuracy of less than 10 metres), and
Insights reflect intuitive dynamics of human movement patterns
PiinPoint also cross-validates its model outputs for visits, pedestrian, and vehicular data points against real-world retail and entertainment visit observations, and third party datasets such as municipal traffic studies. Improvements are being made through increasing our truth set of visit counts and cohort testing to compare accuracy in urban areas vs. rural, standalone buildings vs. attached, shopping malls, etc.
Interfacing with Mobile Data
MLD in the retail analytics industry has generally been applied to business problems in one of two ways. First, it is applied through a self-serve manner, in which the data is licensed in some form or other within a software application. For example, users can see observations by aggregated areas, such as specific places, custom drawn geofences, Block Group / Dissemination Areas, and so on. Second, it is applied through predictive models or analytics, which takes a more customized approach to using the data within modelling techniques and forecasts to help predict performance.
Both of these applications require education, support, and interpretation to properly understand the impact of the user's findings. With self-serve applications, the user has to figure out the impact on their business themselves. How many observations are being represented at this location? How does this compare with my other locations? How does it compare to my competition? While predictive models ensure a more tailored application and assessment of the data, it often leaves the customer with questions as to its impact on their results, given its use is wrapped under proprietary modelling techniques.
Telling the Stories of MLD
PiinPoint allows users to make data-driven retail and real estate decisions through both a self-serve and tailored approach, by visualizing and predicting patterns of MLD. Our goal is to offer transparency in its applications to predictive modelling, and a user interface that will offer an intuitive and accurate experience for our users.
PiinPoint’s applications of MLD can help support business narratives for a myriad of departments. It really shines when it is aggregated across time to understand the capture area of a location, to describe the characteristics of people who visit there, their cross-shopping behaviours, or when observed from a high-level over time to understand changes to visits due to seasonality.
Below are some examples of popular use cases PiinPoint has supported through our Mobile Data Products:
Real Estate / Leasing
What side of the road should my location be on to capture the highest volume of traffic?
What other brands are my visitors going to, that I might want to consider as a co-tenant?
How far are customers willing to travel to this site?
How does visitor activity/traffic volume at this location compare to other sites or my competition?
What are the demographics of people visiting this location?
What days of the week are busier, and may require more staffing in store?
How does nearby road traffic fluctuate and inform my hours of operation?
What are the demographic trends of visitors coming to our site?
What kinds of geosocial segments are occurring by visit reports?
What other places do our visitors like to go to?
Factors Influencing Observations
There are a number of technical factors at play that influence the effectiveness of MLD on business decision-making. Some of these factors are within the mapping or forecasting vendor’s control while others constraints are based on the nature of the data itself. Both are important to understand for your own analysis.
Outside of PiinPoint Control:
Capture Rate represents ~5% of visitors. Using mobile data, the user must have location settings turned on and have one of the 1000s of apps that Veraset partners with.
Building Elevation & Footprint. Innovation on verticality is limited to the providers themselves. Visits to a location that is multi-story will not represent visits to just one level. Larger building footprints with a storefront of 10 meters or more provides greater accuracy.
Precise Geofencing. In a self-serve platform model, the end-user is sometimes responsible for drawing their own geofence to delineate a report area, as is the case in PiinPoint. If the quality of the drawn geofence varies, the results may be easily skewed. For instance, if you're geofencing two big-box store and a lot of the parking lot is captured in geofence B, your results may be skewed.
Low Volumes: A minimum number of observations are required to visualize or understand movement patterns by aggregated areas. If no data is available, there are no observations that can be extrapolated.
Within PiinPoint Control:
Privacy. Data is de-identified, aggregated to a Dissemination Area (DA), and there must be a minimum of 15 people within the geofenced area to be visualized. See more in our Guiding Principles section below.
Differentiating home & work locations. Determined based on time of day and frequency classifications for work and home location detection.
Dwell Time. PiinPoint is able to observe the length of time a visitor was observed at a location in order to discount or approve valid observation data. PiinPoint uses a minimum time period of 2 minutes to determine whether an observation should be a visitor or not.
Velocity. The period of time along with distance between observations illustrates and informs movement dynamics such as whether an observation is a pedestrian or a vehicle driver.
PiinPoint allows users to apply MLD through one of four ways. Each one is outlined below with additional information.
Application 1. Traffic Volumes
Query any major road segment across Canada to find average annual vehicles or pedestrians per day, including coverage on all major roadways: highways, primary, and secondary roadways. Each volume summary includes a breakdown of the traffic observations over time, as well as directionality (i.e. volume dynamics from one side of the road to the other).
Figure 1. Road Segment for IKEA Burlington showing vehicular traffic estimations on highlighted segments.
Application 2. Visitor Reporting
Visitor Reporting is a powerful self-service application to drive the analysis on different sites according to a geofence created by the user. It offers insights on the visitors to a select property in the form of one-off online summaries in the PiinPoint app, called Visitor Reports, which analyze trade area size, visit counts across time, demographics, and related visit activity for a specific location.
Given that the report can be built on any drawn shape in PiinPoint, there is lots of room for creativity and assessment, but also room for confusion and error. For example, the user has the flexibility of geofencing multiple buildings together, a building plus its parking lot, a larger area such as a train station, and so on. The user also has the ability to geofence a very small building with multiple stories, or an in-line retail location whose geofence may imprecisely overlap with a neighbouring store, or an area that is so large and so frequent with visits that it cannot be run in-app.
It is important to understand that these limitations of the data require that some necessary best practices in the app are followed. PiinPoint recommends following the below guidelines for drawing geofences for Visitor Reports:
Building Footprint is greater than 10 m wide:
If you have a footprint under 10m in width and there are shops on either side, you will capture and lose some visitors with the shops on either side due to the limits of the GPS accuracy.
Building is single-storey:
If there are offices or residential units above the unit that you are geofencing, the workers / residents will be captured as well.
Building is a standalone site:
If the location is within a shopping mall, it will be very hard to draw a geofence around a single store location within the mall without also measuring nearby visits.
Due to the potentially immense amount of data that could result, geofences cannot be created on locations that are larger than 2 square kilometres.
Figure 2. User In-App Exploration on a Custom-Drawn Geofence for an IKEA in Burlington.
Figure 3. Trade Area insights from the corresponding Visitor Report for Burlington IKEA Location.
Application 3. Predictive Modelling
Predictive Models are one of the most sophisticated applications of MLD for retail clients with forecasting models. Predictive models using MLD are closely analyzed and measured toward the purpose of the user; i.e. often understanding network performance.
The data is fed into model fitting in various forms and aggregations. Very simple aggregations are included, representing total mobile data activity in a range of localities, for example 50 to 200 meters. Further data aggregations are done to isolate unique devices, returning devices, etc to further inform the model of certain dynamics that might be relevant to the modelling task at hand. Lastly, highly specialized and complex aggregations are calculated and included to represent dynamics very specific to the problem that is being modelled. Examples of the specialized aggregations include traffic volumes along road segments, % shared traffic between two points, demographics of nearby devices, and many more.
Educating Stakeholders on MLD
MLD is both a highly powerful, yet highly nuanced dataset. As previously stated, its best applications are observed when the user and the audience of the data understand its context, limitations, and applicability to their business problem.
PiinPoint is committed to being a supporting and educating partner to interpret and develop a strong understanding of the dataset and its applications to your business. In addition to offering guiding principles with which to frame your presentation of the dataset, our Customer Experience and Data Science team is also available to provide support and input to you as needed.
Guiding Principles to Frame your Presentation of Mobile Location Data
1. New Insights that Exceed the Status-Quo
Surprisingly, many businesses, especially in retail, struggle to understand their customer, despite their customer being a critical piece to strategic success. MLD unlocks insights by helping retail brands get access to customer data. The alternative to date has been running in-store customer surveys, requesting postal code data at checkout, or creating loyalty programs; all of which are generally a small time period sample, or offer limited insight into location behaviours. MLD offers insights on the customer consistently over time without requiring a costly and extensive in-store implementation of customer data collection.
2. Anonymity and Aggregation
There is concern from retailers regarding the ethics behind MLD and its application. While the MLD dataset is enabled only through consumers consenting to share their location, the general impression toward location-enablement is contentious, particularly regarding how much of an effort apps make to ensure the user understands what they are opting-in to. PiinPoint takes additional precaution in the delivery of MLD.
First, the data is anonymized. PiinPoint ensures the data is de-identified, where the observations have only a device ID, disconnected to any individual user. There is no information collected regarding the individual, contrary to popular assumption that the insights include information on the individual’s usage on the phone, what apps they have, or their contact information.
Second, the data is aggregated for display or understanding in PiinPoint products and services, so there is no way you can look at one individual - even a de-identified one - and their unique activity. Data is aggregated to either a Dissemination Area (DA) or a road segment. For geofencing purposes, there must be a minimum of 15 people within the geofenced area to be visualized.
3. "One Piece of the Puzzle"
In geospatial analysis, a generally accepted principle is that the more datasets you use to understand a given market, the more resilient and effective your analysis will be. MLD is most powerful when paired with additional datasets such as demographics, behavioural segmentation, customer survey data, and if available, in-store or online transaction history. Promoting the use of MLD should fit within an overall strategy to be increasingly data-driven as opposed to solely relying on a single dataset.
4. A Sample of Data
A very popular application of MLD is to better understand the customer of a location through the development of a trade area. Just as traditional trade areas of drive times or rings are based on a sample of the population, this is the case with MLD except that it is more granular. However, MLD is still a sample with extrapolated data insights. Understanding that helps to not only support the understanding of your business, but it also provides reassurance to you and the people of your organization - your customers, partners, suppliers, etc. - that they are respecting everyone's privacy.
Still have questions? We'd love to share more with you about Mobile Location Data and talk through this together. Connect with your Customer Success Manager in the chat or email firstname.lastname@example.org for more conversation.