Building Trust and Explainability into Chatbots: the Partena Ziekenfonds Business Case, presented by Karel Kremer, CTO of Oswald, at the Trustworthy and Ethical AI Conference on Feb 13th, 2020
Chatbots and conversational interfaces are taking over customer service departments by storm. In many companies, they provide first-line support to customers. Based on the Partena Ziekenfonds business case, Karel Kremer shares a few critical success factors...
SpatzAI: An intervention app & platform for resolving spats in teamsDesmond Sherlock
More Related Content
Similar to Building Trust and Explainability into Chatbots: the Partena Ziekenfonds Business Case, presented by Karel Kremer, CTO of Oswald, at the Trustworthy and Ethical AI Conference on Feb 13th, 2020
Similar to Building Trust and Explainability into Chatbots: the Partena Ziekenfonds Business Case, presented by Karel Kremer, CTO of Oswald, at the Trustworthy and Ethical AI Conference on Feb 13th, 2020 (20)
Anypoint Code Builder , Google Pub sub connector and MuleSoft RPA
Building Trust and Explainability into Chatbots: the Partena Ziekenfonds Business Case, presented by Karel Kremer, CTO of Oswald, at the Trustworthy and Ethical AI Conference on Feb 13th, 2020
2. OSWALD, an inside view
PARTENA CASE
TRAIN INTERFACE
BOT BRAIN
CONFUSION MATRIX
TEST CASES
STATISTICS
HUMAN TAKEOVER
3. ”I want to go to Paris in two weeks”
CHATBOTS
OSWALD
Real quick!
Intent
+
Entities
+
Context
=
Response
I want to go to Paris in two weeksDestination:Paris Date:2020-02-27
intent: Plan Trip Context:
- name: John Doe
- Means: Flight
- Location: LA, CA
Ok John, here are some flights you might like…
8. PARTENA CASE
OSWALD
Before
Call center with information in large CMS
system.
Customers with a question immediately end
up with the helpdesk
Helpdesk manually takes customers through
the flows defined in CMS
12. TRAIN INTERFACE
OSWALD
Try out and unanswered
Enter some text here and see how the bot
understands your input.
Also find input sent by your users that hasn’t
been handled as it should.
Filter by problem type, intent score, archive
status
View brain and takeover
14. BOT BRAIN
OSWALD
Have a look at how Oswald
THINKS!
What are the intent and the entities it
understood?
What was in context? How did the response it
gave change that?
Any tags in the session?
Which response did it give?
19. CONFUSION MATRIX
OSWALD
What confuses your bot?
Train your bot on 80% of training sentences
Use the other 20% as input and examine the
results.
This can be a measure for how well your bot
will perform on new input.
Which intents are too much alike, which
simply need more training sentences?
22. TEST CASES
OSWALD
Trust that your chatbot
handles changes
When retraining intents
When adding rules to scenarios
When modifying critical responses
24. STATISTICS
OSWALD
How do your users actually
use your chatbot?
What intents are used?
What scenarios?
What tags?
Load?
26. HUMAN TAKEOVER
OSWALD
When your bot needs a
hand
View and filter conversations
Take over from the bot when things go wrong
How did the user get here?