The chatbot space has seen exponential growth in the past few years and is predicted to have an approximate growth of $3 billion by 2021. While chatbots are the current buzz and technologies such as artificial intelligence and machine learning are evolving at a breakneck speed, there are so many questions that are still unanswered questions about this space:
How do you build a great chatbot? How to scale your bot? What are the best frameworks used for chatbot development? How do bot builders overcome some of the challenges that come with it?
Here is a presentation by the CTO of Haptik, one of the world's largest conversational AI platform which answers most of the questions related to building a great & scalable chatbot.
Artificial intelligence in the post-deep learning era
Chatbot Framework Key Elements
1. Framework behind a Chatbot
Bots-Up, Bangalore, Jan 18
Framework behind a Chatbot
Key Note by Swapan Rajdev (Co-Founder & CTO)
Bots-Up, Bangalore, Jan’18
2. About Haptik
Chatbot platform for publishers, advertisers andenterprises
AI powered conversational interface to drive customer engagement
Reach of 30 Million Users, processing 5 Million Chats per month
One of the world’s largest chatbotplatforms
Started in 2013, global pioneers of chatbots
3.
4. First things First
Establish - Use case
Identify - User base
Choose - Platform of communication
Decide -Tone and Content
6. Approaches to build a chatbot
• Use a Chatbot SaaS Platform
• Chatfuel
• Dialogflow
• Gupshup
• Use ML as a service APIs
• Amazon Lex
• Wit.ai
• Build it yourself
• Haptik
• Goibigo
• Use a solutions partner
7. Things to consider when building a bot
1. Data collection
2. Machine Learning Expertise
3. Speed of bot
4. Scalability
5. Speed of iteration
6. Tools and management platform
8. Things to consider when building a bot
1. Data collection
2. Machine Learning Expertise
3. Speed of bot
4. Scalability
5. Speed of iteration
6. Tools and management platform
9. 1. Data Collection
• Very important for a good chatbot
• Required regardless approach
• Need to spend time cleaning the data
• Tag data depending on approach
• Feed as much data back into the bot to make it smarter
10. Things to consider when building a bot
1. Data collection
2. Machine Learning Expertise
3. Speed of bot
4. Scalability
5. Speed of iteration
6. Tools and management platform
11. 2. Machine Learning Expertise
• Different features required
• Intent detection
• Entity Detection
• Spell checker
• Dialogue flows
• Machine Learning has been Democratized
• Can use a hybrid approach between 3rd party APIs and in-house models
• More expertise in house means more flexibility with bots
12. Things to consider when building a bot
1. Data collection
2. Machine Learning Expertise
3. Speed of bot
4. Scalability
5. Speed of iteration
6. Tools and management platform
13. 3. Speed of Bot
• Make or break for user experience
• Important to have at least one reply in less than 3 seconds
• Keep track of latency when using other APIs
• Optimize speed of models to ensure quick replies
• Serial ensemble of models can slow down the bot
14. Things to consider when building a bot
1. Data collection
2. Machine Learning Expertise
3. Speed of bot
4. Scalability
5. Speed of iteration
6. Tools and management platform
15. 4. Scalability
• Important to predict scale and build for that
• Cost will increase with scale
• Measure cost per chat
• Scalable deployments of machine learning models
• Systems should be auto healing
16. Things to consider when building a bot
1. Data collection
2. Machine Learning Expertise
3. Speed of bot
4. Scalability
5. Speed of iteration
6. Tools and management platform
17. 5. Speed of iteration
• Ability to make changes to content and flows frequently
• Should not require engineers to make small changes
• ML models should be able to adapt to data changes in real time
• Keep an eye out on how much time it takes to train your models
18. Things to consider when building a bot
1. Data collection
2. Machine Learning Expertise
3. Speed of bot
4. Scalability
5. Speed of iteration
6. Tools and management platform
19. 6. Tools and Management platforms
• Important for quality and scale of the bot
• Bot building tool: Easy to create and manage bots
• Analytics tool: is important to gather good insights
• Human chat interface: to take control of the chat
• Balance between flexibility and ease of use
20. Bonus: Some great chatbots to check out
Poncho (FB)
Swell (FB)
Letz (app)
Durex (FB)
KLM (FB)
Goibibo (WhatsApp)
Haptik