Tim Walpole Using big data to unlock the delivery of personalized, - PowerPoint PPT Presentation

Tim Walpole Using big data to unlock the delivery of personalized, multi-lingual real-time chat services for global financial service organizations Cognitive Architect

About the Speaker Head of Mobile Cognitive Architect Passionate about designing and delivering complex IT solutions 30 years working as a IT consultant Tim Walpole

About BJSS TH THE UK’S ’S LARGEST T PRIV IVATE TELY-OWNED OWNED I. I.T. T. & BUSIN INESS CONSULTA TANCY >1000 staff in the UK and USA Culture of service and quality

Todays Deck - Who, What and Why ? What Who Cloud Native, vendor agnostic, global chatbot architectures Cognitive architects Chatbot developers Conversational design concepts Conversational designers How to deliver personalized, real time content And now A vision of what's possible..

5

Why should I invest today? Consumers are demanding § Increased Personalisation § Real-Time Insight-Based Decisions § Multi Channel Access (Including Voice) Businesses need to § Become more Efficient & Reduce Costs § Improve Customer Engagement and Loyalty to Drive Revenue § Decrease the risk of Incidents / Compliance Violation § By using the power of Machine Learning

Chatbot Headlines 2018 Gartner IT Hires “By 2019, more than 10% of IT hires in customer service will be writing scripts for conversational bots” Bots will take over By 2021, more than 50% of Production chatbots enterprises will spend more per “By 2020, over 50% of medium to large annum on bots and chatbot creation enterprises will have deployed production than traditional mobile app chatbots” development. “By 2020, 85% of customer interactions will be managed without a human ”

Chatbot Headlines 2018 BJSS BOT Networks Increased Revenue & Loyalty Personalized Conversational BOT’s Personalized BOT’s Complex FAQ Simple FAQ BOT’s Cost to BOT Serve Time Why

Why Chatbots - Summary Deliver 24/7 customer service 1 Reduce the cost to serve 2 Enable personalization 3 Improve customer satisfaction and retention 4 Lead to revenue growth 5

Mizuho Bank and HSBC, Pizza Hut … • Mizuho Bank uses Big Data and Machine Learning to process consumer loan applications in 30 minutes. • Conversational robot workers such as Pepper are being introduced to to to free up human stuff.

Example of a User Story “What will my balance be at the end of the month?” Understand the Present back in AI & NLP question meaningful forms. Process data and Invoke predictive Machine Learning algorithms produce result Big Data Find relevant data

Chatbot Framework Components

Use an extensible framework Why should I use an extensible framework? Business Requirements AI Maturity Vendor Agnostic Tooling Built in Approval & Signoff Processes It’ It’s rapidly changing market Role based Security The There is a need to o evol volve ve as Automated CI / CD & Assurance Sm Small provider ers are e ei either er bei being bo bought Testable Locally AI ser AI ervices es matur ure. e. Blue / Green Deployments ou out, pivot ot or or stop op trading g ! A / B Testing Rich Analytics & Visualisation Tools Te Technologies are evolving rapidly Linked to Change control process for content Mo Most do some things well, No No one provider does it all perfectly Data Security / Compliance bu but not ev ever erything wel ell. GDPR / Personal Data Requirements

BJSS Chatbot Accelerator Key Features BJSS has developed - Low cost Cloud Native architecture - Available on AWS with Azure to follow An enterprise grade - Infrastructure as Code deployment Reference architecture + - Using best of breed AI / ML Services Conversational CMS + - Accelerator Platform Extensible and future proofed To be open sourced later this year

Example Implementations A Vi A Virtual As Assistant for a large A USA A A Educational As Assistant to fi financial provider he help students § Un Unauthenticated FAQ’s § Fi Find the most appropriate college & course § Au Authenticated Ac Account Ac Access § St Start, and compet ete e thei eir chosen en course § Pl Place them in an appropriate Job. § Pl Planning for Worldwide Deployment

Reference Architecture (AWS) Orchestration Step Functions

Orchestration Step Functions

Orchestration Step Functions

Orchestration Step Functions

Regional AI Services Orchestration Step Functions Regional AI Services

Session Context Storage Regional AI Services Orchestration Step Functions

Session Context Storage Regional AI Services Orchestration Step Functions

Data Persistence (Encrypted) Session Context Storage Regional AI Services Orchestration Step Functions

Data Persistence (Encrypted) Session Context Storage Regional AI Services Orchestration Step Functions Client Connection- iOT

Data Persistence (Encrypted) Session Context Storage Regional AI Services Orchestration Step Functions Client Connection- iOT

Data Persistence (Encrypted) Session Context Storage Regional AI Services Orchestration Step Functions Client Connection- iOT

Data Persistence (Encrypted) Session Context Storage Regional AI Services Orchestration Step Functions Client Connection- iOT

Why iOT and MQTT MQ Telemetry Transport Lightweight Message Transport Protocol Low power & Low Bandwidth Bi-Directional Machine 2 Machine Communication Guaranteed Delivery (QOS) Available at very low cost on Standard protocol for most cloud providers Internet of Things Transport

Conversation Design The hardest part of implementing a good Chatbot is the conversation design Our toolchain provides A Conversation Design Methodology A vendor agnostic Conversation Design tool To help you design and document complex Tone of Voice Conversation Flows Smalltalk Multi Language chatbot conversations FAQ’s Quick Replies Fulfilment Handling Failure Personalization Delegation

The Conversation Design Process I want a Pizza Eventually… Elicit Intent Thanks for ordering, your pizza is on its way What type of Crust What toppings Finish Elicit Slots Conversation Lifecycle To confirm, you want Delegate to Confirm a thin pizza with tomato Billing Intent Delegate and cheese Intent

Intent Fulfilment (Personalization) • Dialogflow • Snips.ai • AWS LEX • Microsoft Luis Sentiment Question Detect Analytics Intent(s) Intent Resolution User & Session Context Personalized Response Personalize Response Conversation CMS Live Agent Handoff

Which NLP Engine? Dialogflow Language Snips.ai AWS LEX Detection Microsoft Luis Cost . . . Chatbot Accelerator NLP Minimum Requirements Privacy Intent Detection & Slot Filling Programmatic API’s for Intent selection Slot filling

Intent Fulfilment How does Big Data Help? Machine learning is helping computers Conversations Gone Awry spot arguments online before they happen Sentiment Analysis can be used to detect early Signs of Conversational Failure Personalization Implement a ’toxicity’ filter Big Data can be used to Cornell, Google, and Wikimedia provide personalized, real time researchers train AI to predict when we’ll responses get angry on the internet:

Example - A ’toxicity’ filter https://www.perspectiveapi.com

What do I do next.. - Start with a slim, but full stack solution: UI , NLP , Machine - Focus on a low risk part of your Learning & Big Data running on business to understand what works public cloud. and what doesn’t. - Take an Agile approach to - AI is immature and still developing; deliver new products and the best way to make progress is to services: Prototype, Alpha and rapidly prototype . then Beta.

Final thoughts Script-Writing is one on the Biggest Challenges Make sure you consider the ethics of running your chatbot

To be open sourced later this year Available now - Contact BJSS for further details The BJSS Chatbot Accelerator Any questions Rapidly delivering AI enabled, enterprise scale, strategic chatbots. Tim Walpole Tim.Walpole@bjss.com Cognitive Architect