Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

Speech Recognition, Knowledge Graphs, and AI for Intelligent Customer Operations

47 views

Published on

In the typical sales organization the contents of the actual chat or voice conversation between agent and customer is a black hole. In the modern Intelligent Customer Operations center (e.g. N3 Results - www.n3results.com) the interactions between agent and customer are a source of rich information that helps agents to improve the quality of the interaction in real time, creates more sales, and provides far better analytics for management.

In this presentation we describe a real world Intelligent Customer Operations center that uses graph based technology for taxonomy driven entity extraction, speech recognition, machine learning and predictive analytics to improve quality of conversations, increase sales and improve business visibility.

Watch the recording on YouTube - https://www.youtube.com/watch?v=mUZq_-HHD4g&t=1597s

Published in: Technology
  • Be the first to comment

Speech Recognition, Knowledge Graphs, and AI for Intelligent Customer Operations

  1. 1. No part or process to be used without permission.© Speech Recognition, Knowledge Graphs, and AI for Intelligent Customer Operations
  2. 2. No part or process to be used without permission.© • Sales Skillset + Mindset + Culture • Integrated End-To-End Sales Capabilities • Technology + Software Focused (Cloud) • Corporate + Partner Ecosystems Expertise • Strategy + Design + Execution • Proprietary Technology Platforms • Atlanta (USA) HQ + 12 Global Offices • SiriusDecisions Premier Partner • 12+ year Track Record of Growth 2
  3. 3. No part or process to be used without permission.© 3
  4. 4. No part or process to be used without permission.© 4 CHAT TRANSCRIPT RECORD ID: https://n3result.com/00Q3400001zUSXh Chat Started: Friday, December 01, 2017, 07:55:50 (-0800) Chat Origin: us-sales-english Agent Yannick ( 5s ) Yannick: Hello Todd, thank you for choosing Cisco, my name is Yannick, how may I help you today? ( 52s ) Todd: I need help choosing the correct access points for our businesses. I am planning on purchasing two RV345P routers and I need access points to go with them ( 1m 21s ) Yannick: Sure Todd. I will be happy to assist you. Can you please tell me what prompted this initiative? ( 2m 10s ) Todd: We are adding a second location and I want to connect the two with a VPN. ( 3m 0s ) Todd: Location 1 is smaller and only requires 1 access point. Approximately 40 clients, I need 2.4 GHz and 5 GHz. I want to use POE so that I don't need power where I put the access point. ( 3m 49s ) Todd: Location 2 is larger and will require 2 or 3 access points. Approximately 60 clients, I need 2.4 GHz and 5 GHz ( 4m 12s ) Todd: I only use 2 or 3 wired connections at each locations. ( 4m 51s ) Yannick: Ok. I see. Do you have partner you currently working with or is it the first shopping with Cisco? ( 5m 30s ) Todd: I've tried to contact partners myself and through Cisco but I've never received a call back from anyone and I need to make this purchase today so I need help picking the access points.
  5. 5. No part or process to be used without permission.© 5
  6. 6. No part or process to be used without permission.© 6
  7. 7. Contents  Industry Knowledge Graph (KG)  Around companies, subsidiaries, contacts, products and services, industry and product taxonomies, competing products, current technology stack, propensity to buy, company news, hiring?, white papers – updated over time  Taxonomies  Taxonomy Based Entity Extraction &Sentiment Analysis  Examples of analytics  Text classification: find personas, demand scenarios, industry type  Product Recommendations  Speech recognition  Custom services and how to make life for application developers easier
  8. 8. Foundational shared taxonomy N3 universal sales-cycle and IT terms & concepts taxonomy Campaign Notes & Chats Taxonomies in N3 Knowledge graph The Taxonomy enables identification of - • Words • Terms • Titles • Concepts • Products • Buyer intent • BANT (not the word “BANT”, but the actual sales qualifying stages of BANT) • and Sentiment in the Chats
  9. 9. N3 - Sales Cycle Foundational Shared Taxonomy Hierarchy of SALES concepts and terms Synonyms, Broader, Narrower, Related concepts and terms This application is where the Taxonomies are created – these are built to the W3C SKOS industry standard
  10. 10. Hierarchy of Cisco Product concepts and terms Synonyms, Broader, Narrower, Related concepts and terms N3 – Cisco Specific Products Taxonomy Here we see the Cisco Products terminology organized in the taxonomy
  11. 11. Taxonomy based entity extraction  Regular taxonomy editors pretty good at entity extraction when prefLabels and altLabels are regular words.  AllegroGraph has entity extractor with specializations for product names.  Providing altLabels for every product too time consuming, can be automated  Also needed for post processing when doing speech technology  Built in from AG 6.6  When you need place names, people names, organizations, currencies, etc. we use specialized entity extractors like Cogito or IBM Natural Language Understanding  Also come with automatic linking to dbpedia, geonames, etc…
  12. 12. Taxonomy based entity extraction [2]  Python Spacy if you need NLP capabilities -> POS, special language models  All of the above offer some form of Sentiment Analysis
  13. 13. Taxonomy based entity extraction [3]  Shows example in Gruff!
  14. 14. Examples of analytics
  15. 15. 0 5 10 15 20 25 30 Tiona Hill Christopher Spade Devin Smith Sidney Carr Laura Pugh Billy Young Jacob Holmes Yannick Souna George Hanna Leah Wagner Mary Rowland-Doud Robert Edwards Paul Worley Brad Mcdougald Daulton Tyler Allison Slocomb Ashley Etheridge Elise Nemeth Meghan Bush Wesley Sites Nicole Robinette Jeff Thompson Number of Negative, Neutral and Positive Chats 1200 1000 800 600 400 200 0 Ratio of Positive to Negative Chats Query Results- Chat Sentiment – AI deduced Some BDR agents are more positive than others
  16. 16. Quantity Category Query Results “What high-level technology categories do people chat about most ?”
  17. 17. Query Results “What sellable products (SKUs) are mentioned the most?” Quantity Product
  18. 18. 0 10 20 30 40 50 60 70 80 90 100 Percentage of mention of sellable products to overall Chat Query Results Some BDR agents talk more about sellable/SKUs than others %
  19. 19. Product % Product % Are BDRs spending their time on the right products? Query Results – for BDRs Laura vs Sidney % mention of specific SKUs for overall product mentions
  20. 20. Query Results – for BDRs Laura vs Sidney % mention of specific SKUs for overall product mentions Product % Product % or…let’s automatically route in real-time specific product, sales-oriented Chats to select BDRs, and product support-oriented Chats to other BDRs
  21. 21. Query Results “Who is the most experience on Product X?”
  22. 22. Query Results Top Sellable Product (SKU) by Industry
  23. 23. Product Recommendations  If a customer talks about X a BDR should bring up Y  Based on Oddsratios  Temporal co-occurrence corrected for frequency of each element in pair  Used in Logistics, Health Care, Chomsky Graph
  24. 24. The Green boxes are AI created likelihoods that these other Products (connected blue boxes) will also be discussed.
  25. 25. 0 20 40 60 80 100 120 Touch Screens Systems Engineer Sparkboard 55'/70' Network Security Managed Services Energy Retailing Wireless Solutions DX 80 Optical Likelihood of other product mentioned with Touch 8”/10” tablet Query Results If someone talks about Product X (Touch 8”/10” tablet), they will also very likely talk about Product Y
  26. 26. Text classification and ML Customer Persona Demand Scenerio Prospect Industry Network Buyer Replace/Replenish Construction DataCenter Buyer Upgrade Education Security Buyer Optimize Financial Services Collaboration Buyer Enhance/Pioneer Government SMB Buyer Non-Sales Related Inquiry Healthcare CIO Buyer Immediate Purchase RequestLogistic/Distribution Real Estate Manufacturing Retail/Wholesales Energy/Utilities Categories - UseCase
  27. 27. Captures: Fname and email Chat begins on Cisco website N3 Cisco Chat AI Process flow N3 Patents Pending
  28. 28. CHAT TRANSCRIPT RECORD ID: https://n3result.com/00Q3400001zUSXh Chat Started: Friday, December 01, 2017, 07:55:50 (-0800) Chat Origin: us-sales-english Agent Yannick ( 5s ) Yannick: Hello Todd, thank you for choosing Cisco, my name is Yannick, how may I help you today? ( 52s ) Todd: I need help choosing the correct access points for our businesses. I am planning on purchasing two RV345P routers and I need access points to go with them ( 1m 21s ) Yannick: Sure Todd. I will be happy to assist you. Can you please tell me what prompted this initiative? ( 2m 10s ) Todd: We are adding a second location and I want to connect the two with a VPN. ( 3m 0s ) Todd: Location 1 is smaller and only requires 1 access point. Approximately 40 clients, I need 2.4 GHz and 5 GHz. I want to use POE so that I don't need power where I put the access point. ( 3m 49s ) Todd: Location 2 is larger and will require 2 or 3 access points. Approximately 60 clients, I need 2.4 GHz and 5 GHz ( 4m 12s ) Todd: I only use 2 or 3 wired connections at each locations. ( 4m 51s ) Yannick: Ok. I see. Do you have partner you currently working with or is it the first shopping with Cisco? ( 5m 30s ) Todd: I've tried to contact partners myself and through Cisco but I've never received a call back from anyone and I need to make this purchase today so I need help picking the access points. ( 6m 52s ) Yannick: Oh i am sorry to hear that. What i can do is having someone follow up with you and provide you with an estimate. I would just need to gather some information from you. Is that ok? ( 9m 11s ) Yannick: I can send the info via email if you want to. ( 10m 13s ) Todd: That would be great as I have a meeting I need to leave for. todd@myalliedpediatrics.com ( 11m 10s ) Yannick: You are welcome. Thank you for choosing Cisco. Captures: Fname and email Chat begins on Cisco website N3 Cisco Chat AI Process flow N3 Patents Pending
  29. 29. AI identified key terms, phrases in Chat CHAT TRANSCRIPT RECORD ID: https://n3result.com/00Q3400001zUSXh Chat Started: Friday, December 01, 2017, 07:55:50 (-0800) Chat Origin: us-sales-english Agent Yannick ( 5s ) Yannick: Hello Todd, thank you for choosing Cisco, my name is Yannick, how may I help you today? ( 52s ) Todd: I need help choosing the correct access points for our businesses. I am planning on purchasing two RV345P routers and I need access points to go with them ( 1m 21s ) Yannick: Sure Todd. I will be happy to assist you. Can you please tell me what prompted this initiative? ( 2m 10s ) Todd: We are adding a second location and I want to connect the two with a VPN. ( 3m 0s ) Todd: Location 1 is smaller and only requires 1 access point. Approximately 40 clients, I need 2.4 GHz and 5 GHz. I want to use POE so that I don't need power where I put the access point. ( 3m 49s ) Todd: Location 2 is larger and will require 2 or 3 access points. Approximately 60 clients, I need 2.4 GHz and 5 GHz ( 4m 12s ) Todd: I only use 2 or 3 wired connections at each locations. ( 4m 51s ) Yannick: Ok. I see. Do you have partner you currently working with or is it the first shopping with Cisco? ( 5m 30s ) Todd: I've tried to contact partners myself and through Cisco but I've never received a call back from anyone and I need to make this purchase today so I need help picking the access points. ( 6m 52s ) Yannick: Oh i am sorry to hear that. What i can do is having someone follow up with you and provide you with an estimate. I would just need to gather some information from you. Is that ok? ( 9m 11s ) Yannick: I can send the info via email if you want to. ( 10m 13s ) Todd: That would be great as I have a meeting I need to leave for. todd@myalliedpediatrics.com ( 11m 10s ) Yannick: You are welcome. Thank you for choosing Cisco. Captures: Fname and email Chat begins on Cisco website Help choosing, planning on purchasing, purchase today N3 Cisco Chat AI Process flow N3 Patents Pending N3 Enterprise Knowledge Graph
  30. 30. Customer Persona Demand Scenerio Prospect Industry Network Buyer Replace/Replenish Construction DataCenter Buyer Upgrade Education Security Buyer Optimize Financial Services Collaboration Buyer Enhance/Pioneer Government SMB Buyer Non-Sales Related Inquiry Healthcare CIO Buyer Immediate Purchase RequestLogistic/Distribution Real Estate Manufacturing Retail/Wholesales Energy/Utilities Categories - UseCase AI identified key terms, phrases in Chat CHAT TRANSCRIPT RECORD ID: https://n3result.com/00Q3400001zUSXh Chat Started: Friday, December 01, 2017, 07:55:50 (-0800) Chat Origin: us-sales-english Agent Yannick ( 5s ) Yannick: Hello Todd, thank you for choosing Cisco, my name is Yannick, how may I help you today? ( 52s ) Todd: I need help choosing the correct access points for our businesses. I am planning on purchasing two RV345P routers and I need access points to go with them ( 1m 21s ) Yannick: Sure Todd. I will be happy to assist you. Can you please tell me what prompted this initiative? ( 2m 10s ) Todd: We are adding a second location and I want to connect the two with a VPN. ( 3m 0s ) Todd: Location 1 is smaller and only requires 1 access point. Approximately 40 clients, I need 2.4 GHz and 5 GHz. I want to use POE so that I don't need power where I put the access point. ( 3m 49s ) Todd: Location 2 is larger and will require 2 or 3 access points. Approximately 60 clients, I need 2.4 GHz and 5 GHz ( 4m 12s ) Todd: I only use 2 or 3 wired connections at each locations. ( 4m 51s ) Yannick: Ok. I see. Do you have partner you currently working with or is it the first shopping with Cisco? ( 5m 30s ) Todd: I've tried to contact partners myself and through Cisco but I've never received a call back from anyone and I need to make this purchase today so I need help picking the access points. ( 6m 52s ) Yannick: Oh i am sorry to hear that. What i can do is having someone follow up with you and provide you with an estimate. I would just need to gather some information from you. Is that ok? ( 9m 11s ) Yannick: I can send the info via email if you want to. ( 10m 13s ) Todd: That would be great as I have a meeting I need to leave for. todd@myalliedpediatrics.com ( 11m 10s ) Yannick: You are welcome. Thank you for choosing Cisco. Captures: Fname and email Chat begins on Cisco website Help choosing, planning on purchasing, purchase today N3 Cisco Chat AI Process flow N3 Patents Pending N3 Enterprise Knowledge Graph AI/ML computed Demand Scenario
  31. 31. Speech recognition  Only 5 % of interactions are chats, the rest spoken conversations  Speech recognition getting close to perfection for natural language  BUT: Names are still hard, let alone product names  We explored many different platforms  Google and IBM and several others are ‘fabulous’ but lack the capability to train for thousands of product names  We now use a platform that is trainable  Using sounds-like help from developer  Example  Getting to 5000+ products in noisy environment using telephony quality > 80 % accuracy
  32. 32. Microservices and json(ld)  The typical application developer doesn’t want to know about graph databases, let alone semantic graph databases  We provide a javascript, lisp, prolog compiler that can be used to write REST services  We are close to having all input and output to the knowledge graph be json-ld Application developers DON’T have to learn Triples and SPARQL
  33. 33. Taxonomy BuildingEntity Extractor AGraph Sem-Ingest Analytic Tools Real-time Decision support Ad Hoc Queries AG Sentiment Analyzer Machine Learning Text Classifier CCOR Generator: CoOccurence, Correlation, Odds Ratio REST-Semantic Search Taxonomy Tools AI Labelers Guided Advisors Taxonomy Entities Sentiments Classification Statistical Relations Labels Event Knowledge Graph Text Text, email, chats Voice to TextDocuments (PDF, PPT, Word) AllegroGraph KG Platform Knowledge Graph Platform
  34. 34. Taxonomy Tools AI Labelers Guided Advisors Taxonomy Building Taxonomy BuildingEntity Extractor AG Sentiment Analyzer Machine Learning Text Classifier CCOR Generator: CoOccurence, Correlation, Odds Ratio REST-Semantic Search Taxonomy Entities Sentiments Classification Statistical Relations Labels Event Knowledge Graph Text AllegroGraph KG Platform
  35. 35. Taxonomy Tools AI Labelers Guided Advisors Entity Discovery and Extraction Taxonomy BuildingEntity Extractor AG Sentiment Analyzer Machine Learning Text Classifier CCOR Generator: CoOccurence, Correlation, Odds Ratio REST-Semantic Search Taxonomy Entities Sentiments Classification Statistical Relations Labels Event Knowledge Graph Text AllegroGraph KG Platform
  36. 36. Taxonomy Tools AI Labelers Guided Advisors Sentiment Analysis Taxonomy BuildingEntity Extractor AG Sentiment Analyzer Machine Learning Text Classifier CCOR Generator: CoOccurence, Correlation, Odds Ratio REST-Semantic Search Taxonomy Entities Sentiments Classification Statistical Relations Labels Event Knowledge Graph Text AllegroGraph KG Platform
  37. 37. Taxonomy Tools AI Labelers Guided Advisors Classification Taxonomy BuildingEntity Extractor AG Sentiment Analyzer Machine Learning Text Classifier CCOR Generator: CoOccurence, Correlation, Odds Ratio REST-Semantic Search Taxonomy Entities Sentiments Classification Statistical Relations Labels Event Knowledge Graph Text AllegroGraph KG Platform
  38. 38. Taxonomy Tools AI Labelers Guided Advisors Statistical Relationships Taxonomy BuildingEntity Extractor AG Sentiment Analyzer Machine Learning Text Classifier CCOR Generator: CoOccurence, Correlation, Odds Ratio REST-Semantic Search Taxonomy Entities Sentiments Classification Statistical Relations Labels Event Knowledge Graph Text AllegroGraph KG Platform
  39. 39. Taxonomy Tools AI Labelers Guided Advisors Semantic Exploration “Search” Taxonomy BuildingEntity Extractor AG Sentiment Analyzer Machine Learning Text Classifier CCOR Generator: CoOccurence, Correlation, Odds Ratio REST-Semantic Search Taxonomy Entities Sentiments Classification Statistical Relations Labels Event Knowledge Graph Text AllegroGraph KG Platform Semantic Exploration
  40. 40. Taxonomy Tools AI Labelers Guided Advisors Applications Taxonomy BuildingEntity Extractor AG Sentiment Analyzer Machine Learning Text Classifier CCOR Generator: CoOccurence, Correlation, Odds Ratio REST-Semantic Search Taxonomy Entities Sentiments Classification Statistical Relations Labels Event Knowledge Graph Text AllegroGraph KG Platform Analytic Tools Real-time Decision support Ad Hoc Queries
  41. 41. No part or process to be used without permission.© Speech Recognition, Knowledge Graphs, and AI for Intelligent Customer Operations

×