+ All Categories
Home > Marketing > What's Next: Cloudy with a chance of AI 2

What's Next: Cloudy with a chance of AI 2

Date post: 21-Jan-2018
Category:
Upload: ogilvyred
View: 624 times
Download: 1 times
Share this document with a friend
101
Powered by CLOUDY WITH A CHANCE OF AI
Transcript

Powered by

CLOUDY WITH A CHANCE OF AI

Hello!

Andrew Van Aken Senior Consultant,

OgilvyOne Worldwide

José Arteaga Creative Digital Strategist,

OgilvyRED

Michael McCarthy Senior Consultant,

OgilvyOne Worldwide

What’s the weather like in your city?

Tell us where you’re dialing in from!

Want this deck?

It will be available for download shortly after the webinar on: slideshare.net/socialogilvy

Ogilvy staff: It’s also on The Market! themarket.ogilvy.com

Are you on the go? You can join our webinars on mobile, too! Download the GoToWebinar app from the App Store or Google Play

This Talk

• A Look at Chatbots

• Intro to Reinforcement Learning

• Ogilvy Case Study

A Look at Chatbots

We’ll Talk About…

• What are Chatbots and why are they so popular?

• What are some of the pitfalls? Why do some Chatbots Fail?

• How can a business evaluate if a Chatbot is worth the investment?

We’ll Talk About…

• What are ChatBots and why are they so popular?

• What are some of the pitfalls? Why do some Chatbots fail?

• How can a business evaluate if a Chatbot is worth the investment?

We’ll Talk About…

• What are ChatBots and why are they so popular?

• What are some of the pitfalls? Why do some Chatbots Fail?

• How can a business evaluate if a Chatbot is worth the investment?

What are Chatbots?

Chatbots are automated interfaces designed to simulate an intelligent conversation with a human user via text

Sephora

• Facebook Messenger Platform

• Designed for a specific purpose

• Conversations are linear

Replika

https://www.popsugar.com/news/Replika-Bot-AI-App-Review-Interview-Eugenia-Kuyda-44216396

• Mobile App (iPhone and Android)

• An “AI Friend”, no specific purpose

• Pushing the limits of NLP and intelligent learning

GymBot

• Facebook Messenger

• Drive to fitness videos

• Conversational, but with limits

GymBot

GymBot

Spectrum of Chatbots

?

Spectrum of Chatbots

Chatbot may start conversation, but it can only gather information and will quickly re-route to an agent

Spectrum of Chatbots

Chatbot may start conversation, but it can only gather information and will quickly re-route to an agent

Chatbot is programmed to handle certain types of task to completion, but relies on agent for more complicated inquiries

Spectrum of Chatbots

Chatbot may start conversation, but it can only gather information and will quickly re-route to an agent

Chatbot is programmed to handle certain types of task to completion, but relies on agent for more complicated inquiries

Chatbot can handle all but the most complicated issues, and can learn from new data and

customer interactions

But how do they work?

How do they work?

I. Rule Based

II. Machine Learning

III. Artificial Intelligence

How do they work?

I. Rule Based

II. Machine Learning

III. Artificial Intelligence

Input

Intent

Intent

Intent

Intent

Response

Response

Response

Response

I. Rule BasedHi! How can I help you? You can say (1) Get a Quote (2) Question About Billing or (3) Other

Chatty the Generic Auto Insurance Bot

I. Rule BasedHi! How can I help you? You can say (1) Get a Quote (2) Question About Billing or (3) Other

Get a quote.

Chatty the Generic Auto Insurance Bot

I. Rule BasedHi! How can I help you? You can say (1) Get a Quote (2) Question About Billing or (3) Other

Get a quote.

Chatty the Generic Auto Insurance Bot

Chatbot tries to map user input to a set of pre-defined intents

Quote Billing Other

I. Rule BasedHi! How can I help you? You can say (1) Get a Quote (2) Question About Billing or (3) Other

Get a quote.

Chatty the Generic Auto Insurance Bot

Chatbot tries to map user input to a set of pre-defined intents

Quote Billing Other

I. Rule BasedHi! How can I help you? You can say (1) Get a Quote (2) Question About Billing or (3) Other

Get a quote.

Ok great! To start, what is your first and last name…

Chatbot tries to map user input to a set of pre-defined intents

Quote Billing Other

Response

Chatty the Generic Auto Insurance Bot

I. Rule BasedHi! How can I help you? You can say (1) Get a Quote (2) Question About Billing or (3) Other

I’d like to insure an automobile.

Chatty the Generic Auto Insurance Bot

I. Rule BasedHi! How can I help you? You can say (1) Get a Quote (2) Question About Billing or (3) Other

I’d like to insure an automobile.

Chatbot has to be manually programmed for each intent and response

Quote Billing Other

Chatty the Generic Auto Insurance Bot

I. Rule BasedHi! How can I help you? You can say (1) Get a Quote (2) Question About Billing or (3) Other

I’d like to insure an automobile.

I’m sorry, I didn’t quite understand. Would you like to (1) Get a Quote (2) Question About Billing or (3) Other help.

Quote Billing Other

Error

Chatty the Generic Auto Insurance Bot

Chatbot has to be manually programmed for each intent and response

II. Machine Learning

Hi! How can I help you today?

Chatty the Generic Auto Insurance Bot

II. Machine Learning

Hi! How can I help you today?

Chatty the Generic Auto Insurance Bot

Chatbot can now accept a free-form response

II. Machine Learning

Hi! How can I help you today?

Chatty the Generic Auto Insurance Bot

I’d like to insure an automobile.

Chatbot can now accept a free-form response

II. Machine Learning

Hi! How can I help you today?

Chatty the Generic Auto Insurance Bot

I’d like to insure an automobile.

Chatbot tries to map user input to a set of pre-defined intents

Chatbot can now accept a free-form response

II. Machine Learning

Hi! How can I help you today?

Chatty the Generic Auto Insurance Bot

Chatbot tries to map user input to a set of pre-defined intents

“like to, automobile, insure ”

Prob

abili

ty

I’d like to insure an automobile.

Chatbot can now accept a free-form response

II. Machine Learning

Hi! How can I help you today?

Chatty the Generic Auto Insurance Bot

Chatbot tries to map user input to a set of pre-defined intents

“like to, automobile, insure ”

Prob

abili

ty

Quote Billing Other

I’d like to insure an automobile.

Chatbot can now accept a free-form response

II. Machine Learning

Hi! How can I help you today?

Chatty the Generic Auto Insurance Bot

Chatbot tries to map user input to a set of pre-defined intents

“like to, automobile, insure ”

Prob

abili

ty

Quote Billing Other

I’d like to insure an automobile.

Response

Ok, I can help you with that. To start, what is your first and last name…

Chatbot can now accept a free-form response

III. Artificial Intelligence

Hi! How can I help you today?

Chatty the Generic Auto Insurance Bot

III. Artificial Intelligence

Hi! How can I help you today?

Chatty the Generic Auto Insurance Bot

I’d like to know more about your automobile insurance.

III. Artificial Intelligence

Hi! How can I help you today?

Chatty the Generic Auto Insurance Bot

I’d like to know more about your automobile insurance.

Natural Language Processing

III. Artificial Intelligence

Hi! How can I help you today?

Chatty the Generic Auto Insurance Bot

I’d like to know more about your automobile insurance.

Natural Language Processing

“like to, know more, automobile”

III. Artificial Intelligence

Hi! How can I help you today?

Chatty the Generic Auto Insurance Bot

I’d like to know more about your automobile insurance.

Natural Language Processing

“like to, know more, automobile”

Prob

abili

ty

III. Artificial Intelligence

Hi! How can I help you today?

Chatty the Generic Auto Insurance Bot

I’d like to know more about your automobile insurance.

Natural Language Processing

“like to, know more, automobile”

Prob

abili

ty

Certainly. We offer insurance for both commercial vehicles and privately owned…

Was this information helpful?

Response

III. Artificial Intelligence

Hi! How can I help you today?

Chatty the Generic Auto Insurance Bot

I’d like to know more about your automobile insurance.

Natural Language Processing

“like to, know more, automobile”

Prob

abili

ty

Certainly. We offer insurance for both commercial vehicles and privately owned…

Was this information helpful?

Response

Yes! Thank you.

III. Artificial Intelligence

Hi! How can I help you today?

Chatty the Generic Auto Insurance Bot

I’d like to know more about my auto insurance bill.

Natural Language Processing

“like to, know more, automobile”Certainly. We offer insurance for both commercial vehicles and privately owned…

Was this information helpful?

Response

No, I’m looking for something else.

Prob

abili

ty

III. Artificial Intelligence

Hi! How can I help you today?

Chatty the Generic Auto Insurance Bot

I’d like to know more about my auto insurance bill.

Natural Language Processing

“like to, know more, automobile”Certainly. We offer insurance for both commercial vehicles and privately owned…

Was this information helpful?

Response

No, I’m looking for something else.

Prob

abili

ty

Rule Based Machine Learning Artificial Intelligence

Powering a Chatbot

• Doesn’t try and do too much • Proven technology /

interface • Easier analytics

• Not conversational • Not likely to impress

customers

• Still requires careful programming to handle mapping of intents

• Analytics are more complicated

• Not accessible to the vast majority of companies

• Newer technology, which has its own pitfalls

• Can be conversational and engaging

• Technology has advanced considerably and is now more readily available

• Allows for innovation • Can drastically improve

customer service costs • Potential to be leader in

industry

So… why are Chatbots so popular?

The Rise of the ChatbotMessaging Apps are Outpacing Social Networking Apps

https://learn.octaneai.com/the-complete-beginners-guide-to-creating-a-bot-on-facebook-messenger-1b557b971fb9

The Results Are Promising

100,000 bots within first year of launch

80% of customers re-used Chatbot within 2-weeks of

launch

266% increase in conversion rate versus

average social ad

https://www.smartmessage.com.tr/impressive-numbers-stats-chatbot-dominance-far/

Why Do Some Chatbots Fail?

:(

Some potential causes for concern…

2016 2017March 2016

Microsoft’s Tay Chatbot on Twitter goes on spamming rampage. Begins to spout racist, genocidal and

misogynistic messages to users.

Some potential causes for concern…

2016 2017March 2016

July 2017

Microsoft’s Tay Chatbot on Twitter goes on spamming rampage. Begins to spout racist, genocidal and

misogynistic messages to users.

Facebook bots start talking to each other and forming their

own language.

Some potential causes for concern…

2016 2017March 2016

July 2017

August 2017

Microsoft’s Tay Chatbot on Twitter goes on spamming rampage. Begins to spout racist, genocidal and

misogynistic messages to users.

Facebook bots start talking to each other and forming their

own language.

Two chatbots pulled from Chinese messaging app (QQ) after questioning

the rule of the Communist Party

• Rule based logic is insufficient to meet customer inquiries

• Chatbot doesn’t work with existing business systems

• Weak use case

What Else Can Go Wrong?

Companies pull back

• Facebook shrunk its AI efforts after its bot API fail rate hit 70%

• Everlane, announced that it’s removing several features of its Messenger bot and returning to email

Is a Chatbot Right for Your Business?

Determining if a Chatbot is Right for Your Business

1. Sooner you deploy, sooner you can iterate and learn

2. Chatbots are seeing record engagement rates

3. Customer service bots can alleviate other costs and have a direct ROI

Deploying a Chatbot

Analyze Current Data

1

2

Deploying a Chatbot

Analyze Current Data

Have a Clear Purpose

1

2

Deploying a Chatbot

Analyze Current Data

Have a Clear Purpose

Define Key Metrics

1 3

2

Deploying a Chatbot

Analyze Current Data

Have a Clear Purpose

Define Key Metrics

Test and Optimize

1 3 4

Intro to Reinforcement Learning

First, let’s review last webinar

To Review

To Review

To Review

New Example: David Ogilvy

How can we learn without data?

Enter Reinforcement Learning

• An “educated” guess and check, with rewards

• No starting data needed, just the rules of the road…

Every Time You See Breakthrough

• Book published in 1998…

What is Reinforcement Learning?

Agent

What is Reinforcement Learning?

Agent Takes an Action

What is Reinforcement Learning?

Agent Takes an Action Environment

What is Reinforcement Learning?

Agent Takes an Action Environment

Observation/Reward

+1

What is Reinforcement Learning?

Agent Takes an Action Environment

What is Reinforcement Learning?

Agent Takes an Action Environment

Observation/Reward

Observation/Reward

What is Reinforcement Learning?

Agent Takes an Action Environment

Observation/Reward

-100

What is Reinforcement Learning?

Agent Takes an Action Environment

Observation/Reward

What is Reinforcement Learning?

Agent Takes an Action Environment

Video Games!

AlphaGo Zero

Getting to Greatness

Getting to Greatness

VS

Getting to Greatness

VS

Data/Accuracy

Data/Accuracy

Getting More Efficient

Will This Method Take Over?

• If we just knew how the stock market works…

• Or how proteins folded…

Yann LeCun

• “If intelligence was a cake, unsupervised learning would be the cake, supervised learning would be the icing on the cake, and reinforcement learning would be the cherry on the cake. We know how to make the icing and the cherry, but we don’t know how to make the cake.”

Look a Like Modeling

What is it?

• Building larger segments from smaller ones

Ogilvy Use Case - Everywhere

• B2B/B2C

• Across all mediums

Pretzel Company

• How would a pretzel company increase sales?

Pretzel Company

Company City Employees Alcohol Pretzels

Pete’s Bar 1 45 1 12

Cinema 8 0 23 0 4

Cues and Shoes

0 140 1 1

Popcorn Palace

1 7 0 1

Collect New Data

Company City Employees Alcohol Lbs of Pretzels

Disco Dave’s

0 29 0 ?

Millenium Movies

0 3 0 ?

Gutter Bowl 1 56 1 ?

Chips and Dips

1 40 0 ?

Collect New Data

Company City Employees Alcohol Lbs of Pretzels

Disco Dave’s

0 29 0 24

Millenium Movies

0 3 0 3

Gutter Bowl 1 56 1 13

Chips and Dips

1 40 0 4

What’s Important?

Variable City

City -0.123

Employees 0.3456

Alcohol 0.982

What are the Results?

• Often can see 2-3x increase in performance over regular targeting, according to Exelate

Applying It

• Need to aggregate and plan your data

• Need competitive data

• Need to execute

Want to Stay Present?

• We write weekly on machine learning, artificial intelligence, cloud computing and other technology

• Cloudy with a Chance of AI - Subscribe today!

Questions?

Andrew Van Aken Senior Consultant,

OgilvyOne Worldwide

José Arteaga Creative Digital Strategist,

OgilvyRED

Michael McCarthy Senior Consultant,

OgilvyOne Worldwide


Recommended