The making of the TfL TravelBot

We recently launched our first ever Chatbot – the “TfL TravelBot” on Facebook, which uses artificial intelligence to help answer customer queries expressed in everyday language. The bot was launched just two weeks ago and we have already received lots of great feedback. We wanted to offer you more insight into the thinking behind the TravelBot, and shed some light on how we developed it.

 

The TfL TravelBot launched earlier this month – let us know what you think of it in the comments section below.

Why the TfL TravelBot?

Millions of people already use our website to help them get around London, and we’re constantly seeking new channels to make the process even easier. Research indicates that more than half of the world’s population is now online, and more than 50% of those online are active social media users*. Facebook is comfortably the biggest social media platform, and hence we wanted to take the opportunity to provide them with information via their channel of choice.

Why now?

Instant messaging has emerged as the primary platform for communication these days**. With the advent of digital solutions making it easier to provide conversational platform, we felt it was the right time for us to enter the world of bots. We pride ourselves on being early adopters of technology, and wanted to leverage the potential of existing solutions to come up with a product which is one of the first of its kind in the world of travel.

How was it made?

We designed the logic behind the chatbot and it is hosted in the cloud. Every customer message passes through our logic, and the bot then seeks to deliver the best response. We use artificial intelligence enabled by the machine-learning framework to process the customer messages (Natural Language Processing). It works by understanding intent rather than phrases. Once the message is processed, the bot replies with either a response from our unified API or a friendly retort. The bot is intelligent and has the potential to learn over time.

How does it help?

Apart from being the channel of choice for receiving information, our bot will help the customers in many ways. It will help our customers get the information in the quickest possible time with a 100% response rate. For instance, queries like ‘When is my next bus due?’ can be easily automated, saving customers time and meaning they don’t need to wait for a customer services agent to get a response. In the case of more complex queries, the chatbot can prompt you to speak with an agent.

As a business, this frees up the time of our customer service agents and helps them focus on more complex customer queries. We are also be able to handle many more queries in the same time, therefore improving our response rate.

What next?

We’re constantly looking for feedback to improve our products. If you haven’t it tried yet, search for ‘TfL TravelBot’ on the Facebook Messenger app or go to http://m.me/tfltravelbot on your desktop/laptop. More details on how to use the bot can be found in our previous blog.

Please keep your feedback coming in the comments section below. We know there are more things you would like us to include, and we’re really keen to hear from you.

References
* https://wearesocial.com/uk/special-reports/digital-in-2017-global-overview
** http://www.businessinsider.com/the-messaging-app-report-2015-11?IR=T

5 Comments

  1. Have you got any more details of how it runs, more than saying it’s “hosted in the cloud”? This sounds like a perfect use-case for serverless technologies like lambda. Is that what you used, or did you take a different approach?

    This blog used to go into lots of technical detail, I was hoping that this “making of” post was going to be a deep dive. Particularly as natural language processing and chatbots are all the craze these days. I’d love to hear how you are going about it in a way that scales to your numbers of customers.

    1. Hi Luke, we investigated various technologies during development, but settled on a fairly standard auto-scaling server group for initial deployment. We continue to monitor technologies to use the most appropriate for various workloads but also have to consider our overall infrastructure management before using different technologies. Hopefully, that helps 🙂

Comments are closed.