Chatboty jako nowoczesny sposób komunikacji

1. History
2. Statistics
3. Trends
4. Potential Benefits
5. Potential Blockers
6. NLP vs Buttons
7. Examples from Poland
8. Technologies
9. Conclusions
10. Demo - buy insurance through chatbot
11. Live Coding - chatbot that answers questions 'What time is it in [CITY]?' and 'What is the weather like in [CITY]?'

Transcript

1. Robert Witkowski, Senior Software Engineer, ASC LAB (Chat)boty jako nowoczesny sposób komunikacji
2. ASC LAB - What we do • Monitoring trends and innovations to enrich our offer and raise working standards • Helping and mentoring internal & customers teams with new technologies • Conducting R&D activities
3. Agenda PART I • History • Statistics • Trends • Potential Benefits • Potential Blockers • NLP vs buttons • Examples from Poland • Technologies • Conclusions PART II • Demo - buy insurance through our chatbot • Live Coding - chatbot that answers questions "what time is it in [CITY]?” and “what is the weather like in [CITY]?”
4. Psychotherapist ELIZA (1964-1966)
5. Paranoid schizophrenic PARRY (1972)
6. When PARRY met ELIZA (1973)
7. A.L.I.C.E. (1995)
8. Why we speak about chatbots?
9. Chatbots in Google Search
10. Activity of chatbots discussion from Facebook IQ https://chatbotsmagazine.com/chatbot-report-2018-global-trends-and-analysis-4d8bbe4d924b
11. Activity of chatbots discussion https://nativemsg.com/wp-content/uploads/2017/10/Chatbot-Usage-Statistics.png
12. Not every bot is done well According to recent reports, 70% of the 100,000+ bots on Facebook Messenger are failing to fulfil simple user requests. https://www.fool.com/investing/2017/02/28/facebook-incs-chatbots-hit-a-70-failure-rate.aspx
13. Maroon 5 bot
14. Trends • According to Oracle research 80% of marketers already use chatbots or intend to implement them within two years. • According to Grand View Research, the market will grow by 24.3% every year. By 2025 it will be worth 1.23 billion dollars. • According to Capgemini’s research, within 3 years the share of bots in conversations with clients may increase to 40%. • According to McKinsey until 2020 about 110-140 million positions in the world will be replaced by robots or software. • According to Accenture #TechVision2018 by 2020, 85% interaction with banking clients will be served by chatbots.
15. Trends “In our banks we have people behaving like robots doing mechanical things, tomorrow we’re going to have robots behaving like people.” John Cryan (Chief Executive of Deutsche Bank)
18. Bot interface types • Free Text or Speech (Natural Language Processing) • Response Buttons (“Quick Replies”) • Hybrid
19. NLP vs Response Buttons
20. NLP vs Response Buttons - User Freedom Users can speak freely and ask any question, also outside the domain https://chatbotsmagazine.com/chatbot-nlp-vs-response-buttons-70bcf0aa6a38 Users have to follow the conversational path, simply clicking on the available options
21. NLP vs Response Buttons - Intent Comprehension Precision is understanding user’s requests is still around 60-70% https://chatbotsmagazine.com/chatbot-nlp-vs-response-buttons-70bcf0aa6a38 Clicking on the options, it’s basically impossible to make mistakes
22. NLP vs Response Buttons - Surprise Effect Natural conversation creates surprise and curiosity https://chatbotsmagazine.com/chatbot-nlp-vs-response-buttons-70bcf0aa6a38 Less natural conversational flow may create limited enthusiasm
24. NLP vs Response Buttons - Response Relevance Users might stop asking questions and miss in-depth information https://chatbotsmagazine.com/chatbot-nlp-vs-response-buttons-70bcf0aa6a38 Users are guided through different levels, deeper and deeper into the topic
25. NLP vs Response Buttons - Domain Clarity It can be unclear what questions the bot can answer https://chatbotsmagazine.com/chatbot-nlp-vs-response-buttons-70bcf0aa6a38 Conversation topics are clear from the beginning
27. Examples - Big Four Tech Company
28. Examples • The Chinese Bank OCBC sold loans worth 7 000 000 dollars in three months using a chatbot. • Axis Bank’s chatbot registered 3 million responses. In 97% of cases, chatbot correctly recognized the context and needs of clients. • HSBC Hong Kong uses the "Amy" virtual assistant for corporate banking - the assistant helps to open an account, offers banking products and clarify doubts. • Marriott allows its Rewards Members to search hotels and book rooms. Customer assistance success rate = 44%.
29. Examples from Poland
30. Examples - Bank Millennium Bank Millennium is the first in Poland to take a risk and introduce a chatbot to the first line of contact with the client in mobile app. https://www.bankmillennium.pl/bankowosc-elektroniczna/bankowosc- mobilna/aplikacja-mobilna-klienci-indywidualni-biznes/milla-chatbot
31. Examples - InPost https://www.youtube.com/watch?v=zLE-mjxRBBA
32. Examples - InPost https://www.youtube.com/watch?v=zLE-mjxRBBA
33. Examples - InPost http://m.me/paczkomaty / https://www.k2bots.ai/blog/polskie-chatboty-2018-raport • ~38 000 Total Unique Users • ~42 000 check package status • ~19 000 add parcel to favourite list • 7,7% retention rate (>1 conversation)
34. Examples - Credit Agricole https://m.me/KrEdytka/
35. Examples - Allianz https://www.facebook.com/messages/t/AllianzPolska
36. Examples - Peugeot https://m.me/PeugeotPolska/
37. Examples - Warta https://m.me/wartaPL/ • ~10 000 Total Unique Users • ~700 claims • ~70% of chatbot conversations were handled by themselves • 20 - average conversation length
38. Examples - Santander (BZWBK) https://www.messenger.com/t/EmplobotOfficial
39. Examples - Jean Louis David https://pragmatists.pl/jean-louis-david/ Jean Louis David is a chain with over 30 hair salons; voice-bot solution that significantly relieved the call-center lines by giving 24/7 access to it.
40. Examples and other 1200+ large corporations... https://www.chatbotguide.org/
41. Examples https://www.chatbots.org
42. Technologies • Ready-made solutions with a predefined domain (Emplocity, Edward.ai, InteliWise) • SaaS platforms (chatfuel for Messenger, botsify for web chat) • Frameworks (Microsoft Bot Framework, Botkit, Amazon Lex) • Natural Language Understanding Services or Libraries (LUIS.ai, wit.ai, Dialogflow, Watson, RASA) • Text-to-Speech/Speech-to-Text Services (Google, Microsoft, Amazon Polly, IBM Watson)
43. Conclusions • Computer are good at being computers (searching, analyzing options, process a large amount of data). • Computers aren’t good at understanding human emotion (no empathy, emotional intelligence) now. • Now bots can continue to help us with automated, repetitive tasks. • The hype is over. You should not overestimated the initial chatbots impact. • Thanks to progress in AI, I believe we are at the beginning of explosive growth.
44. DEMO
45. Demo technologies • Microsoft Bot Builder SDK v3 • JavaScript with Node.js – restify – request • LUIS.ai
47. Remember! a bot that does ONE thing well is definitely better and more helpful than a bot that does multiple things with poor quality.
48. https://github.com/asc-lab/microsoft-bot-framework-poc
49. What’s next? • More and more NLP • Tensorflow Bot • Sentiment analysis • FAQ, product catalog
50. LIVE CODING
51. Live coding technologies • Microsoft Bot Builder SDK v4 • TypeScript with Node.js – restify – request • Google Geocoding & Timezone API • Open Weather Map API
52. npm install -g npm npm install -g yo generator-botbuilder yo botbuilder https://docs.microsoft.com/en-us/azure/bot-service/javascript/ bot-builder-javascript-quickstart?view=azure-bot-service-4.0
53. https://github.com/Microsoft/BotBuilder-Samples/tree/master/generators/generator-botbuilder
54. https://github.com/asc-lab/city-information-bot
55. 55 altkomsoftware.pl github.com/asc-lab @AltkomSC asc-lab@altkom.pl