Real-Time Personalized Offer Generation Using Recursive User Profiles
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current purchase propensity models are not scalable for real-time offer generation at a granular level, such as Stock Keeping Unit (SKU) level, and require frequent human intervention for updates, limiting their ability to provide relevant offers due to discrete time hazard models and memory inefficiencies.
Innovation Solution
A scalable, adaptive, and real-time system that processes user transaction data to generate propensity scores using advanced machine learning techniques, with dimensionality reduction and recursive profile updates, enabling automated and self-learning capabilities for personalized offer generation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If purchase propensity models are built at coarse granularity (category level), then model development is simpler and faster, but the models cannot provide granular predictions at SKU level which are needed for personalized offers
Solution Approach 1:
The patent segments the prediction task by building separate propensity models for different item hierarchies (category level and SKU level). This allows the system to maintain simpler category-level models while also creating granular SKU-level models for specific items, thereby achieving high prediction granularity without overwhelming complexity in any single model.
Solution Approach 2:
The patent introduces a hierarchical dimension to the prediction structure, operating at multiple levels (category → subcategory → SKU). This dimensional approach enables the system to leverage patterns at coarser levels to inform predictions at finer levels, achieving granular SKU-level predictions while benefiting from the structure and simplicity of higher-level categorizations.
2Productivity
If discrete time hazard models are used for purchase propensity prediction, then the modeling approach is well-established, but the models require frequent human intervention for updates and cannot operate in real-time
Solution Approach 1:
The patent implements self-service through automated model training and deployment pipelines that continuously learn from new transaction data without human intervention. The system automatically retrains propensity models, updates parameters, and deploys new versions, enabling real-time adaptive predictions while eliminating the need for manual model maintenance.
Solution Approach 2:
The patent establishes continuous model training and updating processes that operate continuously rather than in discrete batches. This allows the propensity models to adapt in real-time to changing purchase patterns, maintaining prediction accuracy without requiring periodic human intervention for model updates.
3Reliability
If traditional batch processing is used for offer generation, then the system is simpler to implement, but the offers do not take into account recent purchase transactions and lose relevance
Solution Approach 1:
The patent applies preliminary action by pre-computing propensity scores and caching prediction results before actual offer generation occurs. This allows the system to have purchase propensity information ready in advance, enabling real-time offer personalization based on the most recent transactions without adding significant computational complexity during the offer delivery moment.
Data Source
AI summary
A system and method for scalable, adaptive, real-time generation of personalized offers is disclosed. A profile of a user is generated, the profile being a summarized representation of historical behavior of the user, the profile containing recursively updated variables. The profile is updated for each new transaction and/or a time dependent event, the new transaction and/or time dependent event including purchase transaction data, user, item hierarchy, and offer data. A affinity scores is generated for the user based on the updated profile, and for each new transaction, one or more offers are generated for the user based on the updated user profile and the affinity scores.