Azure Monitor & Application Insight to monitor Infrastructure & Application
The rise of th enterprise chatbot
1. The rise of the
enterprise chatbot
Gaurav Nigam
Director, Application Development | WorkBoard Inc.
Open at: AIDevWorld 2020
2. About Gaurav
• More than 12 years of experience in the S/W and
cloud industry
• Have been contributing in the distributed
systems, platforms, serverless cloud services,
migrations etc.-
• Have been a code, engineer, and software
architect – implementing microservices, chatbots
and enterprise scale APIs
• Early engineer to a leader in the current company
• Hiking/Biking enthusiast
• Committed foodie to the local burritos and pizzas
• Husband and dad of 1
5. Spend your day like a boss, no
meeting scheduled for today!
How does my day look like?
What about tomorrow?
I found these two meetings in your
calendar for Oct 29th
select to
continue…
Intents and
entities Oct 29th 1:00 pm Oct 29th 2:00 pm
Can you reschedule the first one?
6. NLP: Natural language processing
NLU: Natural language understanding
TA: Tone analyzer
Can you create a meeting with Jessica, David at 10:00 am, this Monday?
NLP
Intents: Intents are the goals that the users
want to achieve through conversation.
e.g.
create a meeting, get a meeting, delete a
meeting, update a meeting, etc.-
Entities: Entities are the key information to
fulfill the intents.
e.g.
meeting(custom), team name (custom),
username, date, time, etc.-
System entities: The system entities help us to
extract the information like date, day,
numbers, boolean, etc.-.
Custom entities: User defined entities as per
the application
NLU
Subject: You
Action: create
Object: meeting
Keywords: Jessica…
The systems sometimes fail to fulfil
the custom entities.
e.g. Most of the system, they don’t
recognize the Indian names.
Train the model as per application
context.
e.g. You create custom entities as per
your platform, you need to extract
those keywords to act on the user’s
intent.
TA
Create multiple responses as per the
sentiments
e.g.
- Sorry to hear that, let me reschedule
it for you.
- I’m glad, I was helpful, let me
reschedule it for you.
Strategize the dialog as per the tone
of the conversation
e.g.
- It looks like a hectic schedule; do you
want to hear something light?
- Happy Monday, do you want to set
up the snooze on this meeting, I can
remind in 10mins?
7. Spend your day like a boss, no
meeting scheduled for today!
Can you schedule a meeting?
Can you schedule a Monday
meeting with Jon at 2:00 pm?
I found two Jon in your calendar,
select to schedule…
It’s with Jon Page
Dialog and
Slots
Jon Doe Jon Page
8. The dialog is a
script that you
program to help
the users for their
end goals
An enterprise bot
provides the
natural language
dialog with the
conversation hints
The conversation
hints are the bot
ability to keep the
users on the script
ü Context is the KEY!
ü The context object of the conversation is the
most important during the conversation
ü Direct context
ü Lateral context
ü The click handlers on the conversation UI
(hints), provide complex dialog flow.
ü Event clicks
ü Carry over context
ü A real world chatbot conversation is-
ü A directed graph, as it moves with the
context
ü The scripted solution solves a set of
workflows
9. Slots
The slots let your bot to ask relevant questions to fulfill your entities’ values.
Can you create a meeting with Jessica, David at 10:00 am, this Monday in
conference room c-1?
Meeting
Meeting name
Meeting participants
Meeting day
Meeting date
Meeting room
ü System defined entities’ slots
ü Day, Date, Location, Users, etc.-
ü User defined slots
ü Workflow name, description, saved replies,
preferences , etc. –
ü Slots digressions
ü Prepare questions for question e.g. I see
Jessica has a conflict at 10:00 am, do you
want to consider another time?
10. Today, October 28th
, you have these
following meetings scheduled
today…
Friday, Oct 16th , 8:00 AM – 9:00 AM
Q4 Strategy planning
Friday, Oct 16th , 4:00 PM – 5:00 PM
Engineering Happy Hours
Retention Friday, Oct 16th , 10:00 AM – 10:30 AM
Kiasko Partner weekly sync up
Can you add Mike too in the 10:00
AM meeting?
11. The retention
is keeping your
users engaged
with the bot.
Welcome
message
Proactive
Notifications
Reminders
and Snoozes
Notifications
channel
Feedback
13. Average
conversation length
Average msg per
conversations
Max conversations Top intents and
entities
Cohorts of the bot
ambassadors
Weak
understandings
Other success metrics
ü Measure and build the funnel pipeline
for all the happy paths of the bot.
ü Measure and build, the steps skip
analytics pipeline, to horn the bot
conversational hints.
100%
•Find meeting time for the
participants
90%
•Create a meeting
75%
•Add an additional participants
62%
•Reschedule the meeting
13%
•Cancel the meeting
ü Use the conversational hints follow up
to build the conversation flow pipeline
ü Keep the backward contexts, to
converse the conversation, happened
earlier
14. Personality
Let me check on that for you! Did
you mean Charlie Brown or Charlie
Sue?
Do I have a meeting with Charlie
today?
You have not clue about what am I
asking?
Sorry about that, sometimes, I
overprotect your schedule :-00
Do I sound like a monkey?
I always wish, I could hear! And I
doubt, I’ve an answer for that.
15. Personality
Direct
Goal Oriented
Amiable
Agreeable-
ness
Awareness
ü Prefer
performance over
personality
ü Simple
Conversation path
ü Must provide
the consistent
responses
ü Should fulfill
the request
ü Must avoid
conflicting
responses
ü Must avoid
unethical
conversation
ü Must listen to the
feedback
ü Must be helpful
ü Must be aware of:
ü Users
ü Customer
ü Must have
assertiveness in
the responses
PrebuiltDialogs
• About YOU
• CONTEXTUAL
welcome
• General
conversations
• CONTEXTUAL
ending
• Favorable
conversations
• Negative
conversations
• SECURITY
assurance
• Conversational
quotes