Andrea Enterprise Digital Employee

Critical Issues:

  • Multichannel information
  • Unhomogeneous requests
  • 80% of data from 20% of requests
  • Frequent low added-value activities
  • Repetitiveness and inattention

Concept

Andrea - Enterprise Digital Employee helps users and companies by performing tasks previously done manually, using Artificial Intelligence techniques:

  • Semantic Analysis
  • Deep Learning
  • Natural Language Processing
  • Predictive Models

Andrea

  • Assists people and automates tasks
  • Learns from past behaviour
  • Builds and maintains data models
  • Predicts and recommends actions
  • Performs actions
  • Can act as a Virtual Assistant to the customer
  • Can act as a Virtual Assistant to contact center operators

Andrea's target

Andrea is aimed at companies:

  • That are customer-centric
  • With a highly developed front-end
  • That need to optimize customer support
  • That require multichannel integration
  • With large volume processes

The main benefits of Andrea

  • More timely communications with the customer
  • Virtual clerks capable of performing very repetitive and low value-added tasks
  • Relieves front-desk operators to focus on high value-added tasks
  • Customer satisfaction

Key functionalities

  • Gathers customer communications from different channels
  • Classifies communications
  • Recognizes attached documents
  • Extracts important information from communications and documents
  • Identifies operations requested by customers
  • Trains virtual clerks
  • Performs operations requested by virtual clerks

Andrea: Workflow

  • Email
  • PEC
  • Fax
  • Call
  • Chatbot
Ingestion

  • Automatic pattern detection
  • Pattern selection
  • Pattern refinement
  • Named entity recognition
  • Sentence detection
Features extraction

Prediction Model
  • Word embedding
  • Algorithm choise
  • Parameters identification
  • Category tree analysis
Prediction Model
Confusion Matrix
  • Quality check on initial categorization
  • Deviations analysis
  • Identification of incorrect predictions
  • Analysis of category tree inconsistencies
Confusion Matrix
Feedback Management
  • Increased predictive accuracy
  • Multiple predictions
  • Best prediction choice
  • Correcting wrong predictions
  • Self-learning of the predictive engine
Feedback Management

Self Organized Map
  • Unsupervised cluster identification from communication texts
  • Variable size grid
  • Homogeneity index
Self Organized Map
UMAP
  • Unsupervised cluster identification from communication texts
  • Geometric clustering in two-dimensional space
  • Homogeneity index
Confusion Matrix

  • NLP
  • Dependency tree
  • Semantic analysis
Linguistic Analysis

  • Named entity recognition
  • Meta data extraction
  • Legacy system link
  • RPA feeding
  • API utilization
Action Detection