2. CRM SOLUTIONS
ABOUT US
AI services as “helpers” in everyday work
• Email classification
• Named Entity Recognition
• Chatbots
Domain-specific solutions (hospitality industry)
• Hospitality CRM
• Contact Center
• Loyalty
• Marketing Automation
3. Knowledge
Knowledge of hospitality, several
industrial sectors and trends.
Long-term work in CRM area, using
Microsoft Technology.
Latest technologies and processes.
6. Use Case – chatbot for
the hospitality industry
Domain-specific question-answering system
Multichannel (Web, WhatsApp, Facebook, …)
Integration with call center (“agent escalation”)
24/7 customer
service
Data sources:
• Hotels’ FAQ’s
• Existing communication via emails
• Client MUST have access to the knowledge base
(a "back office of question-and-answer pairs")
7. MICROSOFT BOT FRAMEWORK
Building a chatbot with Microsoft Bot Framework
Advantages
• easy to use
• pre-built templates and workflows
• integration with other Microsoft services
• good documentation and support
• multichannel communication
Disadvantages
• limited customization options
• not suitable for complex use cases
• requires some technical knowledge
8. Bot service – Azure
MICROSOFT BOT
FRAMEWORK
Building a chatbot with Microsoft bot framework
Cognitive services
Registering multiple channels
Custom integration with our call center software (“agent” intents)
9. KNOWLEDGE BASE
The platform provides a structured way to store and retrieve
information, easier to build conversational bots
Built with Microsoft’s Cognitive services – Why?
Building a custom large language model requires significant
expertise in natural language processing and machine learning,
which may not be available to all organizations
10. KNOWLEDGE BASE
• Based on BERT transformer architecture
• Fine-tuned using smaller dataset of question-answer pairs
• Best answer based on question provided
similarity between question and answer
relevance of entities mentioned in the question
overall coherence of answer
Microsoft Cognitive Services – Question-Answering system
11. Hospitality bot - v1
Multichannel (Web, WhatsApp, Facebook, …)
Question-answering system
Not perfect...
12. Hospitality bot - v1
Implementation and results
Reduction in calls received
(YoY)
(Seasonality)
Chatbot Engagement
13. Use Case 2 – recommending
hotels and creating reservations
Intent Detection
Users wanting to “make a reservation”, “cancel a reservation”,
“modify a reservation”
Service will pass to Question-Answering model all the
other input messages
User preferences
• Predefined list of (mandatory) attributes location,
most-common facilities
• any other important feature the client deems necessary…
Matching availability
Integration with booking systems
- Passing request for the "recommended" objects
- Matching the "best option", along with budget and luxury option
A recommender model that matches the user’s
preference to available objects and facilities
14. Use Case 2 – recommending
hotels and creating reservations
(Phases)
Phase 1
• Using transfer-learning (any basic
language model) and string
matching (e.g., cosine similarity)
• Comparing the user inputs with
available attributes
Phase 2
• Using blind annotations from
agents – data of manual
bookings
Phase 3 – after rollout
Using collaborative filtering and
content-based models together
• Conversion rates
• Feedback from loyalty
members
• Improving existing content
15. USE CASE 2 – RECOMMENDING HOTELS AND CREATING
RESERVATIONS (CHALLENGES)
Using both intents and QnA answers to "navigate" users to a booking
scenario
Complexity:
• Scarce data ("hotel attributes" as seen on popular B2C booking apps)
• Client-specific information
• Domain-specific information ( Boarding, upselling options, room types ... )
• Choosing the right option from availability response
16. Use Case 2 – use chatbot to "get
customers to book hotels"
Implementation
Exemplary attributes:
• Name:Marko Polo
• Location: Korčula
• Rating: 4 stars
• Facilities:private parking, free Wi-Fi, beachfront, family rooms, airport shuttle, non-
smoking rooms, bar, very good breakfast, 2 swimming pools, terrace
Using customer's wishes to match best options
17. Use Case 2 – use chatbot to "get
customers to book hotels"
Step 1
Number of conversions (bookings per conversation): ??
Step 2
Test dataset annotated by agents manually
Step 1&2 ? (no feedback or booking
history)
18. Use Case 2 – use chatbot to "get
customers to book hotels"
Adding and/or changing the content (descriptions, facilities, hotel
attributes)
Do the additional classes and/or content affect recommendations
and/or conversions
• Yes
• How?