Our Technology

Crosswise aggregates over one petabyte of activity data, collected from billions of unique devices every month.

This data includes tens of data points, including device IP address, WiFi networks used, GPS coordinates, websites browsed, ads displayed, device type, operating system, browser cookies, mobile device IDs, time of day and many more. This anonymous activity data does not include any personally-identifiable information (such as name, email address or phone number).

Crosswise applies advanced data science and proprietary machine learning techniques to this data, to construct a probabilistic device map that matches multiple devices - including PCs, phones, tablets and digital TVs - to individual users.

A key element of the technology is the use of over 100 million deterministic pairs (cross-device matches confirmed by unique login) provided by third parties. This data is used continuously to test and improve Crosswise’s probabilistic model. In head-to-head comparisons with device map from other providers, Device Map by Crosswise consistently delivers superior results.

 

Detailed information about billions of devices (PCs, tablets, smartphones, etc.) and the activities performed on them, is collected and aggregated from multiple data sources on an ongoing basis.
Devices are grouped according to time and location usage patterns.
Tens of data points—including device type, operating system, IP addresses, WiFi networks used, GPS coordinates, websites browsed, ads served, browser cookies and mobile device ID—are cataloged for each device.
All possible device pairs within each cluster are prepared, along with the features extracted for each device, in a data structure ready for matching.
An advanced statistical model is used to assign a probability score predicting the likelihood that each pair of candidate devices is used by a single person, with unlikely pairs discarded.
Large data files, containing pairs or groups of matched devices, are distributed weekly to clients for use in their own systems.

 
Device & Activity Data
Detailed information about billions of devices (PCs, tablets, smartphones, etc.) and the activities performed on them, is collected and aggregated from multiple data sources on an ongoing basis.
 
Clustering
Devices are grouped according to time and location usage patterns.
 
Feature Extraction
Tens of data points—including device type, operating system, IP addresses, WiFi networks used, GPS coordinates, websites browsed, ads served, browser cookies and mobile device ID—are cataloged for each device.
 
Candidate Preparation
All possible device pairs within each cluster are prepared, along with the features extracted for each device, in a data structure ready for matching.
 
Matching
An advanced statistical model is used to assign a probability score predicting the likelihood that each pair of candidate devices is used by a single person, with unlikely pairs discarded.
 
Device Map
Large data files, containing pairs or groups of matched devices, are distributed weekly to clients for use in their own systems.
 
Performance Evaluation
The device map’s accuracy (precision and recall) is measured and used to improve future iterations of the overall device-matching process.
 
Ground Truth
Anonymous deterministic device-matching data acquired from third parties, identifying devices on which the same login was entered, is used to train the machine learning models and to evaluate the model’s results.
 
Model Training
Machine learning techniques are used to generate and refine a statistical matching model based on common patterns observed across multiple devices known to be used by the same person.
 
Free Trial
Use our complete device map for 30 days, no strings attached!

Learn More about Device Map by Crosswise

The Crosswise Advantages

Discover why our device map
is superior.

Our Technology

See how we crunch a petabyte of activity data into an accurate device map every week.

Our Customers

Check out who’s using
Device Map by Crosswise.