ML Workflow Step Prediction from Customer Dialogues
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods fail to accurately identify and document the workflow steps taken by agents to resolve customer issues, especially in scenarios where formal workflows do not exist or are not followed, leading to inconsistencies and inefficiencies in customer service processes.
Innovation Solution
A machine learning-based system that utilizes a text-to-text model to predict workflow steps from dialogues between customers and agents, allowing for the extraction of workflows even in the absence of formal procedures, by conditioning the model with allowable action steps and using domain discovery to handle out-of-domain predictions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing methods are used to identify workflow steps, then formal workflows can be documented, but accuracy fails in scenarios where formal workflows do not exist or are not followed
Solution Approach 1:
The patent replaces traditional mechanical/document-based workflow tracking with a machine learning-based natural language processing system. The ML model processes unstructured dialogue text to automatically identify and extract workflow steps, eliminating the need for pre-defined formal workflows while maintaining accurate step identification through pattern recognition and domain discovery capabilities.
Solution Approach 2:
The system changes the parameter of workflow representation from structured formal documents to unstructured natural language dialogues. By transforming workflow capture from requiring formal documentation to accepting informal conversation, the system adapts to various workflow contexts while maintaining identification accuracy through ML-based pattern recognition.
2Reliability
If manual documentation of workflow steps is performed, then formal workflows can be captured, but time consumption and labor requirements increase
Solution Approach 1:
The system enables self-service workflow extraction where the machine learning model automatically processes dialogue data and generates workflow step documentation without human intervention. The domain discovery and prediction mechanisms autonomously identify workflow patterns, eliminating manual documentation efforts while maintaining consistent and reliable workflow capture.
Solution Approach 2:
The patent substitutes manual documentation processes with automated machine learning-based extraction. The system processes dialogue text through NLP techniques, automatically identifies workflow steps, and generates structured documentation, thereby eliminating time-consuming manual efforts while improving consistency through algorithmic processing.
3Adaptability or versatility
If workflow steps are extracted without formal procedures, then adaptability to informal workflows improves, but consistency and uniformity in step naming deteriorate
Solution Approach 1:
The system applies parameter changes by transforming informal, variable step naming in dialogues into standardized, uniform step names through the machine learning model's prediction capabilities. The domain discovery mechanism learns from dialogue patterns and consistently maps similar actions to standardized step names, maintaining uniformity while preserving adaptability to informal workflow contexts.
Solution Approach 2:
The patent replaces ad-hoc manual naming with automated ML-based step name generation. The trained prediction model consistently produces uniform step names by learning from training data patterns, thereby standardizing workflow step nomenclature while maintaining the ability to extract workflows from informal dialogue sources.
Data Source
AI summary
Content of a dialog between at least two communication parties to resolve a task is received. A specification associated with at least a portion of eligible steps of a workflow is received. Machine learning input data is determined based on the received content of the dialog and the received specification. The determined machine learning input data is processed using a trained machine learning model executing on one or more hardware processors to automatically predict a sequence of workflow steps representing the dialog.