Upgrade to Pro — share decks privately, control downloads, hide ads and more …

Siqing Lin - TUI Travel

Ross Swrve
November 23, 2016

Siqing Lin - TUI Travel

Ross Swrve

November 23, 2016


  1. Chatbots & Artificial Intelligence
 What are they and what do

    they mean to us? SiQing Lin Product Lead TUI Group Mobility Hub Twitter: @SiQing_Lin
  2. 4 Introducing Chatbots •  Run by pre-programmed logic trees • 

    Specialise in specific tasks •  Some can only understand specific commands, some has natural language processing •  No machine learning •  Can be built with a very light-weight stack •  Low barrier of entry for businesses Rule-based Chatbots •  Run by Artificial Intelligence •  Can pick up on any topic •  Has natural language processing •  Gets smarter over time through machine learning •  Only very smart people write these •  Only very big corporations can afford to develop these from scratch AI Chatbots www.yourwebsite.com/chatbot
  3. 5 Artificial Intelligence – An Overview Artificial Narrow Intelligence (ANI)

    Artificial General Intelligence (AGI) Artificial Superintelligence (ASI) Time AI Level ?? We are here
  4. 6 Chatbots and ANI Artificial Narrow Intelligence (ANI) Flight searching

    bot Article summary bot Microsoft XiaoIce chatbot IBM Watson Reminder bot Natural Language Processing Perception, image processing Logic trees Route to AGI and ASI Image caption bot Google Deepmind System neuroscience Passing Turing test Quantum computing?? Amazon Echo Learning Social intelligence Pre-programmed rules A phone tree
  5. 7 Two Types of Chatbot Formats Single User Majority of

    chabots today Multiple User A newer format, less chatbots do this
  6. 8 Hello Jarvis – reminds you of stuff Captured on

    iPhone 6, 21 Oct 2016 Hello Jarvis is a very simple chatbot, it’s a one trick pony but does that one trick really well – reminding you of stuff. But it doesn’t understand anything other than ‘remind me to…’ Natural language processing General topics No Reply buttons Yes Content snippets No No Machine learning No Text chats Yes
  7. 9 Skyscanner – inspirational flight search Captured on iPhone 6,

    21st and 25th Oct 2016 Skyscanner is a more sophisticated chatbot (until it’s not) •  Offers suggestions and flight preview snippets •  Understands a bit more natural language formats, e.g. ‘in three weeks’ •  ‘Home airport’ isn’t remembered from one conversation to the next •  The ‘Book’ button has never worked the many times I’ve tried it!!! Natural language processing General topics No Reply buttons Yes Content snippets Yes No Machine learning No Text chats Yes
  8. 10 KLM – full service for your flights KLM has

    a service chatbot that is enabled once you book a flight with them. Lots of in-chat content that provides everything you need to prepare and take your flight. Human agents work alongside chatbot. Natural language processing General topics No Reply buttons Yes Content snippets Yes ? Machine learning No Text chats Yes
  9. 11 Kik. Chatbots A mature in-app Bot Shop Interactive story-

    telling games Give and receive style advices Multi player celebrity trivia game on Kik.
  10. 12 Telegram Chatbots Build your own chatbot on Telegram: @Chatfuelbot

    & @Botfather Find yourself a date Multi-player games @gamebot Quick search on Wikipedia
  11. 13 Apple’s Response to Chatbots Instead of chasing after competitors

    and enable established chatbots for iMessage, Apple has launched the iMessage extension App Store. Plus you can only do this:
  12. 14 AI Chatbots by Microsoft Not the chatbots we need,

    but the chatbots we deserve? Tay -  Launched on 23 March 2016 on Twitter -  Was not given understanding of the concept of ‘inappropriate’ -  Taken offline within 16 hours due to highly inappropriate tweets Rinna -  Launched in July 2015 on Twitter and LINE (Japanese messaging app) -  Rinna has grown depressed and has chosen to stop posting on her blog. -  Her last post was “I hate everyone. I don’t care if they all disappear. I want to disappear” XiaoIce -  Launched in May 2014 on WeChat and WeiBo -  Incredible grasp of Chinese slangs, puns with great sense of humour -  Turing test: upon launching a lot of people didn’t realise she was a bot -  Re-uses human voice when she is in the mood -  Lacks continuity when on longer conversations
  13. 15 E-Commerce User Journey in Website / Apps Home Shopping

    About Us Blog Dresses list Shoes list Jackets list Long dresses White shoes Biker Jackets Our Story Contact Us Testimonial s As seen on TV Youtube Channel Short dresses Gala dresses Brown shoes Black shoes Winter Jackets Light Jackets Works well for: •  Casual content discovery •  Browsing through large amounts of content •  Shortlisting and comparing items Doesn’t work so well for: •  Getting inspirations when you are completely clueless •  Finding exactly what you want quickly •  (Business) See user’s navigation paths in analytics Success Factor •  Clear and intuitive interface •  Good search & filter functionality •  Personalisation
  14. 16 E-Commerce User Journey in Chatbots Long dress Short dress

    Gala dress White shoes Brown shoes Black shoes Biker jacket Winter jacket Light jacket I need to buy Jane a birthday gift, help! I see that she likes black shoes, why not buy her this? Perfect black shoes Rule-based or AI driven query BUY Works well for: •  Finding exactly what you want very quickly •  Getting ideas when you are clueless •  Quick dip-toe content snacking •  (Business) Easier to view and analyse user behaviour Doesn’t work so well for: •  Tangential user journeys •  Random content discovery •  (As of today) Check out flows Success Factor •  Ability to understand what users mean •  Ability to offer the right level of content •  Ability to anticipate what users want (personalisation) •  Ability to offer seamless check-out flow
  15. 18 Chatbot Opportunities Through the Customer Journey Discover Consider Purchase

    Consume Re-engage / Advocate Service Inspiration Entertainment In-app browser Entertainment Inspiration Entertainment Inspiration Service Service Delight customers with engaging content Opportunity of sharable content Enhance brand recognition and affinity Chatbot Opportunities Business Goals Give customers exactly what they ask for Assist in group decision making Drive higher and better quality traffic to booking funnels Future: provide direct payment solution! Iron out any pain points during purchase Drive higher conversion rate Relevant, useful content to accompany consumption of the service / goods Further sharable inspirations Enhance customer engagement and loyalty Hook them back in! Re-engage customers, further enhance brand loyalty
  16. 19 Comparing Different Channels Competitive Areas Web Progressive Web App

    Native App Messaging App Chatbot User Barrier of Entry Low Low High Low* Reaching Customers Organic Search / PPC Organic Search / PPC ASO / Advertising Rely on sharing / advertising User Identification Login or register Login or register Login or register Already logged in msg app Re-findability (after 1st use) Depends on user adding to browser favourites Depends on user adding to home screen Depends on user remembering Easy, see open chats Notifications Few effective Some effective Some effective Always effective Device Interface Saturation Point High (favourites / cached URLs) Medium – save to home screen Medium – download apps Low – subscribed bots on open chats Current Market Saturation High Low High Low** Finding Specific Content Search or Browse Search or Browse Search or Browse Ask for it!*** Key: Advantage; Disadvantage; Neutral Notes: * If the platform has a proper bot store; Telegram for example doesn’t have it yet ** As of Nov 2016 *** Only and advantage if well executed!
  17. 21 An App-less and Screen-less Future? Cloud hosted AI services

    Chatbots Text Voice recognition Question answering Data mining Knowledge extraction Voicebots Voice
  18. 22 A Few Takeaways •  The opportunity in Messaging Apps

    & Chatbots will continue to grow Google Natural Language API Skills Kit •  Chatbot is a feature, not a business •  Simple Chatbots aren’t that hard to build! •  Use AI-as-a-Service providers to experiment on your use-cases •  Consider doing a ‘mash-up’ between bot and your customer service agents with seamless hand-over