Automation Discovery Recalculation for Noisy Desktop Actions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for identifying automation opportunities in business processes are often subjective, time-consuming, and computationally costly, and fail to achieve an optimal cost-to-performance ratio, particularly when dealing with incomplete or noisy data from human-computer interactions.
Innovation Solution
A semi-supervised approach using dynamic time-window optimization to recalculate transition probabilities between routines, allowing user-edited processes and incorporating predefined thresholds to enhance the identification of automation opportunities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If unsupervised learning automation discovery procedures are used to analyze noisy and untagged user desktop actions, then broad data gathering capability is achieved, but the computational cost becomes formidably high
Solution Approach 1:
The patent segments the computational task by dividing the dataset into multiple batches and processing them iteratively. The unsupervised learning procedure is applied in stages, with each batch contributing to the refinement of process models and transition probabilities, thereby reducing the computational burden compared to processing the entire dataset at once.
Solution Approach 2:
The patent applies partial action by using a predefined threshold to filter transitions. Only transitions with probabilities exceeding the threshold are retained for further analysis, discarding noisy or irrelevant transitions. This partial processing approach reduces computational cost while maintaining the quality of automation opportunity identification.
2Reliability
If transition probabilities are calculated using traditional methods, then automation opportunities can be identified, but the cost-to-performance ratio is suboptimal
Solution Approach 1:
The patent employs dynamic time-window optimization where the time window size is not fixed but adaptively adjusted based on the characteristics of the data and the specific transition being analyzed. This dynamic approach allows the system to optimize transition probability calculations for different contexts, improving accuracy while avoiding the excessive computational cost of using a single large window for all cases.
Solution Approach 2:
The patent changes key parameters such as time window size and probability thresholds based on the specific automation opportunity being evaluated. By dynamically adjusting these parameters rather than using fixed values, the system achieves better performance across diverse scenarios without requiring overly complex computational resources.
3Ease of operation
If manual automation discovery procedures are used, then subjective bias is introduced, but the process becomes time consuming and expensive
Solution Approach 1:
The patent incorporates feedback mechanisms where the results of automated analysis are used to refine and improve subsequent analyses. Transition probabilities are recalculated based on accumulated data, and the system learns from patterns identified in the dataset, progressively improving accuracy while maintaining high speed and low cost.
Solution Approach 2:
The system performs automated discovery of automation opportunities without requiring manual intervention. The unsupervised learning procedure automatically identifies processes, routines, and transitions from the data, eliminating the need for human analysts to manually examine each potential automation candidate, thereby achieving high productivity and cost efficiency.
Data Source
AI summary
A system and method may identify computer-based processes which may be candidates for automation. Embodiments may involve a semi-supervised approach for identifying processes as automation opportunities. Transition probabilities for pairs of routines within a candidate process may be calculated based on a set of instances of the process (e.g., in a dataset of computer actions) using a dynamic time-window optimization procedure, where transition times may be measured for a plurality of instances of a first and second routines of a given pair of routines, and where statistical distributions may be calculated and used for deriving one or more time windows, describing a predetermined percentile (e.g., the 70th percentile) of the measured transitions and used for estimating a transition probability for the pair of routines. In some embodiments, the input set of transitions and routines may be generated by a user or business analyst using a graphical user interface (GUI).