AI Platform Company
Big Data Analytics, Deep Learning, Artificial Intelligenceⓒ 2016, MindsLab. All Rights Reserved
Contact Center Solution Case Study
using Artificial Intelligence
Minds Lab Inc.
ⓒ 2016, MindsLab. All Rights Reserved - 1 -
Paradigm Shift
“ It’s the beginning of a brand new era - there was web, then mobile apps
and now bots. It is an incredibly powerful paradigm shift ”
David Marcus, Head of Facebook Messenger
Paradigm
PC
( mid – 80s )
Web
( mid – 90s )
Smartphone
( mid – 00s )
Messaging
( mid – 10s )
Platform
Examples
Applications
Examples
UI/UX
S/W Dev
Desktop
DOS, Windows, Mac OS
Clients
Excel, PPT, Lotus
Native Screens
Client-side
Browser
Mosaic, Explorer, Chrome
Website
Yahoo, Amazon
Web Pages
Server-side
Mobile OS
iOS, Android
Apps
Angry Bird, Instagram
Native Mobile Screens
Client-side
Messaging Apps
WhatsApp, Messenger, Slack
Bots
Weather, Travel
Conversation
Server-side
by Beerud Sheth, Founder & CEO of gupshup.io
ⓒ 2016, MindsLab. All Rights Reserved - 2 -
Contact Center Evolution : From VOC Analytics to AI Assistant
Contact center's primary goal is to shift the paradigm from human to artificially
intelligent agents, thereby meeting the highest standards of today's industry.
Call Center
Virtual
Assistant
Social
Big
Data
VOC
Analytics
• (Real-time) Voice Recognition
• (Real-time) Risk Detection
through Text Analytics
Artificial Intelligence Virtual Assistant
VOC-
STT
• Natural Language
Processing
• Dialog Management
• FAQ, Account Inquiry
• Real Time Speech to Text
• Using STT Data as
learning Data
VoC Analysis
VOC-
TA
• Text Analytics of
Call Center
Counselling
• Reputation
Analysis of Social
Big Data
• Interface
with Legacy
System
Crawling
ⓒ 2016, MindsLab. All Rights Reserved - 3 -
Virtual Assistant (So called AI Assistant) is …
Virtual Assistant utilizes Voice Recognition, Natural Language Processing and ChatBot
Technology to process and to undertake the agent-to-customer interaction.
Virtual AssistantCustomer
Text Chatting Voice Chatting
Customer Human Agent
Web / Mobile App
Machine
Learning
Voice Recognition
& NLP
Intention
Classification
Answer Generation
ⓒ 2016, MindsLab. All Rights Reserved - 4 -
Ideal VA System Architecture and Service Process
Solution for VOC analysis and sentiment analysis:
Natural-Language-Processing, Deep Neural Network, and Dialog Management
Integrated
Dialog
Management
Dialog Scenario
Natural
Language
Processing
Domain /
Intention
DNN*
Classifier
Voice
Recognition
(STT: Speech-
to-Text)
ChatBot
HAD (Human
Assisted Dialog)
Legacy System
ChatBot (MINDs VA) in Financial Industry
* Deep Neural Network
Slot Filling
User/Session
Management
Log Monitoring /
Statistics
Dialog Modeling /
Scenario Manager
Dialog /
Classification
Learning
FAQ
External Service
Text Analysis
Sentiment
Analysis
Dialog DB
Voice
Synthesis
(TTS: Text-
to-Speech)
FAQ DB
API
Human Agent PC
• Speech to Text
• Natural Language
Processing
• Text Analytics
• Sentiment Analysis
MINDs VOC
Machine
Learning
Image
ⓒ 2016, MindsLab. All Rights Reserved - 5 -
MINDs VA Application– 1) Web/Mobile Text based Counselling
Service Inquires including counselling and many other transactional inquires can
be done both on the web and mobile using Chatbot.
Customer
Web
Mobile
Human
Agent
Chat Bot
In-charge
FAQ
Classifier
FAQ
General
Counselling
Automatic Assignment
Transactions
Expert
Counselling
Feedback
Complaints
DM
Dialog
Monitoring
* Human-Assisted Dialog
HAD
API
Legacy system I/F
• Account / Order / Reservation Inquiry
• Product Order, Reservation , Payment, Transfer
Domain
Classifier
Quick & Unified Communication
And Direct Access
ⓒ 2016, MindsLab. All Rights Reserved - 6 -
MINDs VA Application – 2) Human-Assisted Dialog
When VA fails to detect unknown contexts, human experts can intervene and
provide further information.
• Domain Classif.
• Intention Classif.
• FAQ Classif.
Dialog Session
Confidence level
Monitoring
Warning
Human
Agent
VA
Counsellor (Managerial)
Answer Recommendation
score
 XXXXX 65
 YYYYY 63
 ZZZZZ 58

Join
Client
History
Customer
Scale of
complaints
ⓒ 2016, MindsLab. All Rights Reserved - 7 -
MINDs VA Application– 3) Simple inquires at the call center
Virtual Agents can process and communicate in parallel with human agents,
further improving the center's performance
PBX
IVR
CTI
Call
“ Hi, how
can I help
you today? “
Domain
Classifier
STT/NLP
Human Agent
Human Agent
Human Agent
Virtual Agent
Virtual Agent
Virtual Agent
TTS
STT/NLP
TTS
STT/NLP
TTS
Call Assignment
Decoding
Encoding
ⓒ 2016, MindsLab. All Rights Reserved - 8 -
Performance Enhancement by using Reinforcement Learning
Any insufficient data can be remedied using reinforced-learning technique to
accelerate the growth of the VA.
Session Monitoring
• Call disconnection
• HAD Transferred Counselling
• Long wait time
• Repeated Request
Confidence Level Monitoring
• Voice Recognition
• Domain Classification
• Question Intention Classify
• FAQ Classification
VA Performance Monitoring Tools
Customer Feedback
Monitoring
• Call Satisfaction
• Errored Answer
• Complaints
• TA: Pattern
Detection
• Session Log
• Statistics
• Dialog
Generation
Confidence Level
Log
• Customer
Feedback per
Session
• Complaints on
VA
Reinforcement
Learning Data
Gathering
Human Assisted
Dialog
Counselling KMS /
Manual / Script
Reinforcement
Learning
ⓒ 2016, MindsLab. All Rights Reserved - 9 -
MINDs VA Benefits
Cost Effective
• Web/mobile covers 5% of the call
• When work with designated call center, over
10% of coverage.
• Replaces E-Mail / SMS Consultation
Customer Satisfaction
• 24 / 7 Service
• Convenience with Voice Interfaced system
• Easy and Quick Access
• Simplified Customer Contact Channel
• Less waiting time
Specialized Agent
• Focus on complicated and sensitive inquires.
Omni-Channel CRM
• Integrated Analysis on various Customer
Channel
Customer Service
• Expand the service to customized marketing
and video counselling etc.
Efficiency with Maintenance
• System Upgrade through Periodic
Reinforcement Learning
• Minimize Manual Maintenance
Copyright © 2016 Minds Lab. All rights reserved
Dasan Tower 6F, 49 DaewangpangyoRo, BundangGu Seongnam City, GyeonggiDo, Korea
T.031-625-4340 F.031-625-4119 | www.mindslab.ai www.mindsinsight.co.kr
No part of this publication may be circulated, quoted, or reproduced for distribution outside the client organization without prior written approval.
Thank You

Minds Lab Contact_Center_Solution_Using_ai_v1.0

  • 1.
    AI Platform Company BigData Analytics, Deep Learning, Artificial Intelligenceⓒ 2016, MindsLab. All Rights Reserved Contact Center Solution Case Study using Artificial Intelligence Minds Lab Inc.
  • 2.
    ⓒ 2016, MindsLab.All Rights Reserved - 1 - Paradigm Shift “ It’s the beginning of a brand new era - there was web, then mobile apps and now bots. It is an incredibly powerful paradigm shift ” David Marcus, Head of Facebook Messenger Paradigm PC ( mid – 80s ) Web ( mid – 90s ) Smartphone ( mid – 00s ) Messaging ( mid – 10s ) Platform Examples Applications Examples UI/UX S/W Dev Desktop DOS, Windows, Mac OS Clients Excel, PPT, Lotus Native Screens Client-side Browser Mosaic, Explorer, Chrome Website Yahoo, Amazon Web Pages Server-side Mobile OS iOS, Android Apps Angry Bird, Instagram Native Mobile Screens Client-side Messaging Apps WhatsApp, Messenger, Slack Bots Weather, Travel Conversation Server-side by Beerud Sheth, Founder & CEO of gupshup.io
  • 3.
    ⓒ 2016, MindsLab.All Rights Reserved - 2 - Contact Center Evolution : From VOC Analytics to AI Assistant Contact center's primary goal is to shift the paradigm from human to artificially intelligent agents, thereby meeting the highest standards of today's industry. Call Center Virtual Assistant Social Big Data VOC Analytics • (Real-time) Voice Recognition • (Real-time) Risk Detection through Text Analytics Artificial Intelligence Virtual Assistant VOC- STT • Natural Language Processing • Dialog Management • FAQ, Account Inquiry • Real Time Speech to Text • Using STT Data as learning Data VoC Analysis VOC- TA • Text Analytics of Call Center Counselling • Reputation Analysis of Social Big Data • Interface with Legacy System Crawling
  • 4.
    ⓒ 2016, MindsLab.All Rights Reserved - 3 - Virtual Assistant (So called AI Assistant) is … Virtual Assistant utilizes Voice Recognition, Natural Language Processing and ChatBot Technology to process and to undertake the agent-to-customer interaction. Virtual AssistantCustomer Text Chatting Voice Chatting Customer Human Agent Web / Mobile App Machine Learning Voice Recognition & NLP Intention Classification Answer Generation
  • 5.
    ⓒ 2016, MindsLab.All Rights Reserved - 4 - Ideal VA System Architecture and Service Process Solution for VOC analysis and sentiment analysis: Natural-Language-Processing, Deep Neural Network, and Dialog Management Integrated Dialog Management Dialog Scenario Natural Language Processing Domain / Intention DNN* Classifier Voice Recognition (STT: Speech- to-Text) ChatBot HAD (Human Assisted Dialog) Legacy System ChatBot (MINDs VA) in Financial Industry * Deep Neural Network Slot Filling User/Session Management Log Monitoring / Statistics Dialog Modeling / Scenario Manager Dialog / Classification Learning FAQ External Service Text Analysis Sentiment Analysis Dialog DB Voice Synthesis (TTS: Text- to-Speech) FAQ DB API Human Agent PC • Speech to Text • Natural Language Processing • Text Analytics • Sentiment Analysis MINDs VOC Machine Learning Image
  • 6.
    ⓒ 2016, MindsLab.All Rights Reserved - 5 - MINDs VA Application– 1) Web/Mobile Text based Counselling Service Inquires including counselling and many other transactional inquires can be done both on the web and mobile using Chatbot. Customer Web Mobile Human Agent Chat Bot In-charge FAQ Classifier FAQ General Counselling Automatic Assignment Transactions Expert Counselling Feedback Complaints DM Dialog Monitoring * Human-Assisted Dialog HAD API Legacy system I/F • Account / Order / Reservation Inquiry • Product Order, Reservation , Payment, Transfer Domain Classifier Quick & Unified Communication And Direct Access
  • 7.
    ⓒ 2016, MindsLab.All Rights Reserved - 6 - MINDs VA Application – 2) Human-Assisted Dialog When VA fails to detect unknown contexts, human experts can intervene and provide further information. • Domain Classif. • Intention Classif. • FAQ Classif. Dialog Session Confidence level Monitoring Warning Human Agent VA Counsellor (Managerial) Answer Recommendation score  XXXXX 65  YYYYY 63  ZZZZZ 58  Join Client History Customer Scale of complaints
  • 8.
    ⓒ 2016, MindsLab.All Rights Reserved - 7 - MINDs VA Application– 3) Simple inquires at the call center Virtual Agents can process and communicate in parallel with human agents, further improving the center's performance PBX IVR CTI Call “ Hi, how can I help you today? “ Domain Classifier STT/NLP Human Agent Human Agent Human Agent Virtual Agent Virtual Agent Virtual Agent TTS STT/NLP TTS STT/NLP TTS Call Assignment Decoding Encoding
  • 9.
    ⓒ 2016, MindsLab.All Rights Reserved - 8 - Performance Enhancement by using Reinforcement Learning Any insufficient data can be remedied using reinforced-learning technique to accelerate the growth of the VA. Session Monitoring • Call disconnection • HAD Transferred Counselling • Long wait time • Repeated Request Confidence Level Monitoring • Voice Recognition • Domain Classification • Question Intention Classify • FAQ Classification VA Performance Monitoring Tools Customer Feedback Monitoring • Call Satisfaction • Errored Answer • Complaints • TA: Pattern Detection • Session Log • Statistics • Dialog Generation Confidence Level Log • Customer Feedback per Session • Complaints on VA Reinforcement Learning Data Gathering Human Assisted Dialog Counselling KMS / Manual / Script Reinforcement Learning
  • 10.
    ⓒ 2016, MindsLab.All Rights Reserved - 9 - MINDs VA Benefits Cost Effective • Web/mobile covers 5% of the call • When work with designated call center, over 10% of coverage. • Replaces E-Mail / SMS Consultation Customer Satisfaction • 24 / 7 Service • Convenience with Voice Interfaced system • Easy and Quick Access • Simplified Customer Contact Channel • Less waiting time Specialized Agent • Focus on complicated and sensitive inquires. Omni-Channel CRM • Integrated Analysis on various Customer Channel Customer Service • Expand the service to customized marketing and video counselling etc. Efficiency with Maintenance • System Upgrade through Periodic Reinforcement Learning • Minimize Manual Maintenance
  • 11.
    Copyright © 2016Minds Lab. All rights reserved Dasan Tower 6F, 49 DaewangpangyoRo, BundangGu Seongnam City, GyeonggiDo, Korea T.031-625-4340 F.031-625-4119 | www.mindslab.ai www.mindsinsight.co.kr No part of this publication may be circulated, quoted, or reproduced for distribution outside the client organization without prior written approval. Thank You

Editor's Notes

  • #3 To start off the presentation this afternoon, I would like to point out this movement of paradigm, “ It’s the beginning of a brand new era….” Many of us here might already recognize the shift. Starting from PC Era, we are living in the era which you can easily see people using smart phone and messaging on the street. From the early 2000, we have been effected by messaging paradigm. Among the broad spectrum of messaging paradigm, I would like to talk about its application to a specific sector, call center.
  • #4 VOC plays major role on contact center evolution. Started with VoC analytics, the contact center’s primary goal, today in 2016, is to shift the paradigm from human to artificial intelligent agent to meet the high standards of today’s industry. We all familiar with the fact that VoC analytics and the call center have successfully worked together. With the success of VoC analytics, we are moving on to the next level, which is AI Virtual Assistant.
  • #5 Virtual Assistant is an AI agent who studies conversation done by human agents and assists human agents with simple orders. Virtual Assistant utilizes voice recognition, natural language processing and chatbot technology to process the agent to customer interaction.
  • #6 It is total solution for VOC analysis and sentiment analysis using Minds VOC’s original technology. Once, it recognizes the client voice, it goes through natural language processing, then it analyzes the client’s intention through Deep Neural network. The key here is categorizing the intension of the calls. At the Integrated Dialog Management stage, the calls are categorized into 4 different streams. And, when the call can not be recognized by VA, It goes to Human Assisted Dialog system.
  • #7 For the next 3-4 slides, I would like to introduce different application of VA. First of all, It can be applied to web/mobile based counselling through text msg. Service inquires including counselling and many other transactional inquires can be done using Chatbot. It is Quick and unified communication and direct access. And, again by going through domain classifier, the inquires go into 5 different categories. As an example of expert counselling, the client can communicate with VA through dialog management. For the general counselling, client can ask questions to chatbot, and chatbot answers. For more sensitive and complicated requests, such as complaints, domain classifier sends the requests to human agent in -charge.
  • #8 Second application is HAD, Human-Assisted Dialog. When VA fails to detect unknown contexts, human experts can intervene and provide further information. By using confidence level monitoring and scale of complaints, human agent receives warning, when he or she needs to join. 노란색 파란색 bar 과 answer recommendation 파트는 이해가 잘 가지 않았습니다.
  • #9 The third application is simple inquires at the call center. Virtual agents can process and communicate in parallel with human agents, further improving the center’s performance. Here is the abstraction of the voice analytic procedure. Calls go through recording device and domain classifier. Then, the calls distributed to human agents and virtual agents.
  • #10 Since, the virtual assistant needs to be consistently trained, reinforcement learning is crucial to performance enhancement. Any insufficient data can be remedied using reinforced-learning technique to accelerate the growth of the VA. First, three parts of monitoring tower monitors session, confidence level and customer feedback. (좀더 부가설명 필요하신 부분중에 번역이 필요한부분 알려주세요) Second, human assisted dialog and counselling kms are gathered for the learning. Then, reinforcement learning is done using this routine. The goal of this learning is to increase the accuracy and speed. Just as human agent needs consistent training, virtual assistance continuous learning for better quality service.
  • #11 Then, my last point is how the virtual assistant benefits companies and customers. It is cost effective solution. It covers 5%-10% of the call depending on different business settings. And, it replaces emails and SMS consultation with the lower cost. Second, it satisfies clients. Because the service runs 24/7, client have more access to the service It is convenient because it is voice interfaced system, your hands are free, which allows easy and quick access. And, the system simplifies customer contact channel, (usually it takes a lot of time and men power to do sorting job) Then, we expect less waiting time for the client when using VA. Third, since VA takes simple orders, we save our men power for complicated and sensitive inquires. Fourth, it is more efficient with maintenance, because the system upgrades through periodic reinforcement learning, and it minimizes manual maintenance. Fifth, it is omni-channel crm, not multi communication channel, it allows integrated analysis on various customer channel. Lastly, it enhances customer service, since it can be expanded to various customized services.