7. Plan
1. Intro
2. AI and current Chatbots
3. Hopes vs reality
4. Natural Language Understanding
5. Our approach
8. “30 or 40% of our clients’ messages are recurrent and could
be partly automated. ”
Axa France
In China, a brand will
create a bot on WeChat
before creating a website.
3 billion
users
An indeniable trend
12. AI and current Chatbots
An undeniable trend
April 2016: 30 000 bots created in 3 months but…
« Bots right now are in the trough of despair. To industry
observers, it feels like they are overhyped and under-delivering. »
Greg John, CEO of Burner
Current chatbot technology is nothing new. It becomes
interesting when chatbots meet AI.
13. Hopes vs Reality
Understanding the limits of AI Chatbots
✓ Bots are not yet intelligent
(language, context)
✓ Questions need to be predicted
✓ Answers need to be written in
advance
✓ Complicated features take time
to develop
Some bad examples
Х Tay: Microsoft’s error in 2016
Х M: Facebook’s perfect bot
14. Natural Language Understanding
The basics
To build a chatbot able to converse with human, you need NLU technology.
✓ Data-powered
✓ Detecting intentions
✓ Focused on keywords and trigger
words
✓ Understanding words in contexts
15. Our approach
Data-based chatbots
✓ We analyze historical datasets
✓ We detect FAQs
✓ We map those questions and detect intentions
✓ We input this data in the bot
17. 17
Bot Admin &
Analytics
Your Data
Pull Data
Push API
Your Data
Push Data
Webhook
Trigger actions and
answers (Conversation
Management platform)
Intent detection
(Clustaar Deep Query)
User input
Integration with data
20. Plan
1. How to envision a bot project?
2. Use cases
3. How to make it great?
21. Envision a chatbot project
Designing a new user experience
✓ Define the objectives of the bot
✓ Adapt to the target
✓ Recreate user habit
✓ Transform them into conversation
✓ Imagine client reactions
22. Possible use cases
All recurrent interactions
Internal
(ex: Nexity)
- HR
- Knowledge
management
- Search in databases
- FAQ
Client acquisition
(ex: Cortex)
- Integrated to a
website
- Automatic lead
generation
- FAQ
- Available 24/7
Service
(ex: Hachette)
- Playful features
- Geolocalisation / Store
locator
- Promotional offers or
news Push
- FAQ
Customer service
(ex: 20 Minutes)
- Integrated to a website
or Messenger
- Automatic FAQ in
Natural Language
- Available 24/7
24. How to make it great
Put a smile on the client’s face
✓ Smalltalk: simulating human interaction and setting a tone
✓ Fallback: building a fluid conversation
✓Playful features: jokes, quiz…
Find the detail that will make the user say « thank you! »
25. Data import Bot training Put the bot online Run, manage & improve
Internal sources (FAQ,
conversations) and external
sources (Google Queries)
Intent detection
Connexion with internal data
Writing scenarios and
answers
UX & integration
Publication
Fine tuning
Machine learning
improvement & Analytics
Improving response
scenarios
25
Project phases