Customer Behavior Change Detection Using Normalization Value
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Change detection systems face challenges in identifying changes in customer behavior when the output distribution of customer actions is unknown, requiring the determination of both normal output and changes over time.
Innovation Solution
A computer-implemented method and system that detect customer action data, calculate customer parameters such as mean and standard deviation, generate a normalization value, calculate deviation from expectation, and compare cumulative sum values with preselected thresholds to generate informative electronic communications for merchants.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If change detection is performed when output distribution is unknown, then the system can detect changes in customer behavior, but the complexity of determining both normal output and changes increases
Solution Approach 1:
The system performs preliminary actions by collecting customer action data over a specified period before change detection begins. This preliminary data collection establishes a baseline of normal customer behavior, allowing the system to detect deviations without requiring pre-known distribution parameters. The preliminary action phase includes calculating initial mean and standard deviation values that serve as reference points for subsequent change detection.
Solution Approach 2:
The system applies self-service by automatically determining its own baseline parameters (mean and standard deviation) from collected customer action data without requiring external input or pre-programmed distribution knowledge. The system serves itself by generating the normalization values and control limits needed for change detection directly from its operational data, reducing the need for complex external configuration.
2Measurement precision
If customer action data is collected over a period of time to determine parameters, then detection accuracy improves, but the time required for change detection increases
Solution Approach 1:
The system applies partial action by collecting customer action data for a predetermined, limited period rather than continuously or indefinitely. This predetermined period is sufficient to establish accurate baseline parameters (mean and standard deviation) without excessive data collection. The system balances the need for accurate parameter estimation with the constraint of detection time by using just enough historical data to establish reliable baselines.
Solution Approach 2:
The system performs preliminary data collection and parameter calculation before the actual change detection process begins. This preliminary phase establishes the baseline parameters needed for detection, allowing the main detection process to proceed efficiently without repeatedly accessing historical data. The preliminary action separates the parameter estimation phase from the detection phase, reducing overall detection time.
3Reliability
If normalization values are generated based on standard deviation thresholds, then false alarms are reduced, but the complexity of calculating deviation from expectation increases
Solution Approach 1:
The system applies parameter changes by dynamically adjusting the normalization value based on the calculated standard deviation of customer action data. When the standard deviation falls below a predetermined threshold, the system generates a normalization value that accounts for this low variability, thereby reducing false alarms. This parameter adjustment allows the control limits to adapt to the actual data characteristics rather than using fixed thresholds.
Solution Approach 2:
The system introduces an intermediary calculation step that computes deviation from expectation using the mean, normalization value, and a noise factor. This intermediary value serves as a mediator between the raw customer action data and the final change detection decision. By introducing this intermediate calculation, the system reduces false alarms while managing complexity through a structured, multi-step approach rather than direct threshold comparison.
Data Source
AI summary
Apparatuses, methods, and systems for detecting changes in customer behavior are disclosed. One method includes detecting customer action data, receiving, by a marketing platform server, the customer action data over a period of time, determining, customer parameters including a mean, and a standard deviation of the customer action data, generating a normalization value when the standard deviation is detected to be less than a deviation threshold, calculating, by the marketing platform server, a value of deviation from expectation based at least on the mean, the normalization value, and a noise factor, calculating a current cumulative sum value of the customer action data based on a prior cumulative sum value and the value of the deviation from expectation, comparing the current cumulative sum value with a threshold, and generating an electronic communication for the merchant server when the current cumulative sum value satisfies a compared condition with the preselected threshold.