Visitor Stitching Models for Unified Analytics
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing website analytics techniques fail to provide a unified view of interactions for anonymous or non-uniquely identifiable visitors across multiple devices and browsers, leading to undercounting of distinct visits by the same user.
Innovation Solution
The implementation of visitor stitching models that quantify unique visits with a certain level of confidence, generating unified records by grouping visits from the same person, and allowing for various analytics to be performed on both unique and unified records based on user-defined confidence levels, using models such as log-in information, IP address, or probabilistic models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If visitor stitching models are implemented to group visits from the same person, then measurement precision of visitor interactions is improved, but device complexity increases
Solution Approach 1:
The system segments visitor identification into multiple independent stitching models (log-in based, IP address based, probabilistic models), each handling specific scenarios. This allows the complex task of accurate visitor attribution to be divided into manageable components that can be selected and applied as needed, improving measurement precision without requiring all models to run simultaneously.
Solution Approach 2:
The system dynamically selects which stitching models to apply based on the specific visit data and confidence level requirements. Rather than using a fixed complex system, the architecture adapts its complexity level by choosing appropriate models for different scenarios, maintaining measurement precision while managing computational resources.
2Reliability
If multiple stitching models are used to increase confidence in visitor attribution, then reliability of visitor identification is improved, but ease of operation deteriorates
Solution Approach 1:
The system provides a universal interface that handles multiple stitching models through a single unified architecture. The same codebase and processing pipeline support log-in based, IP address based, and probabilistic models, making the system versatile while maintaining consistent operation patterns. This multi-functionality improves reliability through model diversity without complicating the user experience.
Solution Approach 2:
The system incorporates confidence levels as feedback mechanisms, where each stitching model provides a confidence metric for its attributions. This feedback allows the system to automatically adjust which models to trust and apply, improving reliability through evidence-based selection while keeping the operation simple as the system self-regulates based on confidence scores.
3Productivity
If visitor stitching is applied to group visits across devices, then productivity of analytics processing is improved, but loss of information increases
Solution Approach 1:
The system applies different stitching strategies to different data sources and visit types, maintaining local quality in the attribution decisions. Rather than applying a uniform stitching approach that might lose information, the system tailors the stitching method to the specific characteristics of each data source, improving productivity through efficient processing while preserving information accuracy.
Solution Approach 2:
The system changes parameters such as confidence thresholds and stitching criteria based on the specific analysis requirements and data availability. This parameter flexibility allows the system to adjust the balance between productivity and information preservation, enabling efficient processing while maintaining information integrity through adaptive parameter settings.
Data Source
AI summary
Techniques are provided for transforming information about unique visits to a website or interactions with a mobile application by a single individual on different devices into unified analytical data representing these visits. One or more visitor stitching models can be used to quantify unique visits as belonging to the same person with a certain level of confidence. In addition to maintaining separate records for each unique visit, a unified record is generated, which represents a combination of the unique visits by the same person. Various analytics can then be performed on the unique and unified records. The unified records are favored over the separate data records wherever the stitching model corresponding to the unified record provides a certainty that meets or exceeds a user-specified confidence level. The visitor stitching model used for the analytics is considered non-committal because different models can be employed or none at all, depending on user preferences.