Putting an assistant in people’s pockets
Keen to offer a new digital touchpoint for customer interaction, Lufthansa decided to develop its own chat bot for Facebook Messenger—one of the most widely-used instant messaging apps in Europe.
Ivonne Engemann comments: “Having seen the rise of chat bots in recent years, I wanted to play around with the technology, and see if we could harness artificial intelligence to help improve the Lufthansa customer experience. So, I started working on a prototype on my PC.
“I chose MongoDB as the underlying database for the application because I wanted a platform that we would be able to scale flexibly as the project took off. Another important deciding factor was that MongoDB stores data in JSON format, the same native data-structure used by our Node.js back-end. And because there is no need for a rigid relational schema, it’s very easy to change things as your application grows. This made it perfect for our web-based chat bot, which we needed to develop in a very agile and iterative way.”
When the prototype began to take shape, Engemann needed to decide on the best way to deploy it into production. At around the same time, a Lufthansa team that was working on a different project introduced her to IBM® Bluemix®. She quickly realized the potential for building the chat bot solution on the Bluemix cloud application development platform.
“After learning that we could run our database on Bluemix via IBM Compose for MongoDB, we were sold,” says Ivonne Engemann. “Compose is a fully managed service, which means that IBM manages all of the back-end database administration—leaving us free to focus on development. In addition, we can scale our database instances seamlessly to meet changing requirements very quickly and easily, and the service levels that Compose provides around availability are impressive.
“It was very easy to get started,” she adds. “We were able to set up a Bluemix account and get our development environment up and running in less than an hour. What’s more, I only had to rewrite about five lines of code to move the app from my development environment to IBM Compose—it couldn’t have been simpler.”
The chat bot application Lufthansa Best Price—known as Mildred—integrates data from a number of other services via open and private APIs, including Lufthansa’s in-house services to locate the customer’s nearest airport and the best prices for each flight, as well as Facebook Messenger, Google Geolocation, a natural language processing (NLP) platform, and, of course, IBM Compose for MongoDB. Bluemix makes it easy to integrate these services with the core application logic, which runs on a Node.js server on the Bluemix platform.
The first release of Lufthansa Best Price uses data from these services to calculate the cheapest flights between two airports, which it identifies by parsing text from conversations with customers in the instant messaging app and extracting important data-points, such as cities and dates. As the conversation with the user continues, Mildred stores these data-points in the IBM Compose for MongoDB database, enabling her to keep track of items that have previously been mentioned and respond to each new query more effectively.
Ivonne Engemann elaborates: “Customers can chat to Mildred as though they were communicating with any other of their Facebook contacts. The chat bot processes travel queries such as ‘I want to go to the Eiffel Tower next week’, recognizes the locations that are mentioned, searches for the customer’s nearest airport, and finds the cheapest flights.”
Revolutionizing customer services
Today, Mildred is on hand to help customers find the best-value flights 24 hours a day, 365 days a year. The chat bot responds to queries in real time, ensuring that customers don’t have to wait around for an answer, or spend time calling the Lufthansa helpline.
Ivonne Engemann remarks: “After launching the chat bot, we experienced a huge spike in demand. With thousands of people now exchanging tens of thousands of messages with Mildred, data volumes are growing fast. With IBM Compose, we’re totally confident that we can seamlessly scale to meet demand, ensuring that customers get fast, accurate flight information whenever they need it.
“The ability to store users’ responses is critical as we work on improving Mildred’s cognitive capabilities. Three weeks after launch, we had over 15,000 expressions in the NLP platform—we never expected that there were so many ways of asking for flight information!
“The data stored in IBM Compose helps us train Mildred to recognize new formulations and respond accordingly. We’re currently investigating whether we could integrate IBM Watson® Conversation via Bluemix to enhance our NLP capabilities even further.”
Based on Mildred’s success, Lufthansa has set its sights on making more customer services available via interactive digital channels in the future.
Ivonne Engemann concludes: “Thanks to innovative IBM cloud data services, we’ve been able to turn our chat bot experiment into a game-changing customer service offering. We're looking forward to developing the solution further, and we know that we can count on IBM for support.”