Air traveler self-service with NLU

The top e-commerce retailers have set the standard for a modern customer experience. As consumers we take these expectations with us across industry boundaries. And then we encounter this.

Natural language understanding (NLU) technology can untether passengers from airport staff and call centers, enabling them to explore options with voice or text conversation. This is conversation AI.

One popular method of deploying conversation AI is to use a 3rd-party chatbot like WhatsApp or Facebook Messenger. The UI is already there, lots of people use those apps, and other companies have gone first. It’s a good way to get started at the cost of an ever expanding list of UIs to manage.

Another approach is to incrementally enhance your current systems with conversation AI to make them more useful and engaging.

For example, adding a working microphone icon to an airline or hotel mobile app can turn it into an AI-powered voicebot. A minimal UI design change versus ramping up a dedicated chatbot channel. 100% of customers who use your mobile app will have access to AI-powered conversations without going through a 3rd-party interface.

An IBM Watson Chatbot

We built an IBM Watson chatbot which we connected to the British Airways NDC system as an experimentation in using NLU to grab flight information. The NDC API had just been released and was fairly easy to get working with the chatbot.

In the clip, I requested flights from Boston to London and make itinerary changes before finally letting it get the flights. We connected the NLU engine to the British Airways booking system so the results are for BA or OneWorld carrier flights and in Pounds Sterling (GBP).

How many results is optimal?
The booking system sends back dozens of offers, which we display five at a time. I’ve seen other chatbots that display one “best” option. I’m not yet a fan of that due to trust. How do I know that this single result is my best option? Is there something cheaper/faster/better I’m not seeing?

This will change in the future when our trust in AI is built on a history of successful personal experiences. For now, displaying more than one result increases my confidence and proves that I’m the one in control of the transaction.

To zero in the most relevant flight/fare options I asked it for “evening nonstops only to Heathrow” which reduced the list to two flights offering four fare choices.

Whether on a website or an app, the UI team decides what to display and how.

Even at this early stage, conversation AI is having a positive impact on the customer experience, while reducing costs. One approach is to deploy a new customer engagement channel with the caveats noted above. Another method is to incrementally enhance your current systems to make them more useful and engaging.

Feel free to contact us if you need some guidance with your AI or natual language strategy or with IBM Watson chatbot development.

Leave a Reply