French transit develops chatbot with SAP Conversational AI
The transit system in the Paris metropolitan area is tackling an AI-driven chatbot project to create an easy way for riders to access personalized information. Here's a look.
The Paris metropolitan area has 3,200 trains serving 3.2 million riders every day. As anyone who has ridden public transit knows, this leaves plenty of room for delays and annoyed customers, whether they're locals using the trains to get to work or tourists heading to the Louvre. Société nationale des chemins de fer français (SNCF, or the National Society of French Railways), which serves the Paris metropolitan area, wanted to make sure every rider could find the information they needed, including delay information, fares and routes. It set out to develop a chatbot using SAP Conversational AI, still widely known as Recast.AI, for these types of customer service queries.
SAP acquired Recast.AI in January as part of its Leonardo machine learning strategy, as well as its planned investments in France. Recast.AI, renamed SAP Conversational AI, is a platform to develop bots using community templates, natural language processing (NLP) and integration with channels like Slack and Facebook Messenger. Chatbots are nothing new with SAP systems, but the purchase of Recast.AI is another indicator that SAP is dedicated to making inroads with Leonardo technology.
SNCF began using Recast.AI to develop a natural language chatbot in 2016, before the purchase by SAP, according to Olivia Fischer, digital media director for Transilien -- the SNCF-owned suburban railway service for Paris and certain surrounding areas. It took approximately six months to develop a prototype to launch on a small scale for SNCF agents and early adopters, and it is now used by a few hundred people on a regular basis, she said.
"The chatbot helps us bring customer service and personalized information on an industrialized scale," she said. For hearing-impaired riders, the bot helps them access information without having to use voice services, she noted.
Currently, SNCF has programmed the bot to answer queries in three languages: French, English and Spanish. Developing in more languages is expensive, especially with different alphabets, such as Arabic or Russian, Fischer said. However, with the 2024 Summer Olympics coming to Paris -- and the influx of tourists it will bring -- adding support for more languages is on the horizon, she noted.
From Conversational AI 'baby bot' to more sophisticated bot
Olivia FischerDigital media director, SNCF
However, the biggest challenge with using NLP and AI in the chatbot has been for the bot itself to understand the intention of the user, according to Fischer. Customers don't necessarily have realistic expectations for the bot, and SNCF is still developing it to provide end-to-end service information and provide alerts to customers, she said.
Still, SNCF expects to release the chatbot at the beginning of 2019 at the latest with much better understanding of user intentions. Using SAP Conversational AI enabled the agency to develop a bot in nine weeks, but it was a "baby bot" that needed to be trained thoroughly, according to Fischer. "We have a moderator who tags every question and attributes a level of confidence to the intention," she said. As a result, the bot's understanding of queries has improved, she added.
The bot did need to be trained for "small talk," Fischer noted. It also needed to have a "personality" built into it, as well as deal with hostility from angry customers. SNCF does not intend for the bot to pretend to be a human agent but to identify itself as a bot, she said.
Challenges of SAP Conversational AI with legacy systems
The SAP Conversational AI bot pulls data from a patchwork of legacy systems, including a CRM system, according to Fischer. This has posed challenges and led to an effort to reorganize IT systems, she said. Currently, SNCF uses no other SAP products.
Additionally, CRM is still new for SNCF, particularly since no one is required to provide their identifying information to ride public transportation, Fischer noted. AI and machine learning can help the agency learn more about riding patterns, as well as use data from the SNCF app to understand customers on a more granular level. However, SNCF plans to do more with data provided by users, such as where problems exist, and use that to propose alternate routes, instead of using personal data to market to customers, she said.
SAP Conversational AI bot to ease burden on call centers
As part of the strategy, SNCF plans to use the Conversational AI bot to help customers with self-service activities, like checking train times and mapping routes, and to use the call center for more value-added tasks, according to Fischer. One of the projects for 2019 is to connect the bot and chat with the call center, as well as use it on Twitter, a platform that SNCF uses to update riders on the status of every train line, she said.
Ultimately, SNCF has made headway with programming its Conversational AI bot, but it still has a way to go before the 2024 Summer Olympics and the increase in ridership. Once the bot completely understands natural language queries and is fully integrated into SNCF's legacy systems, it sounds like it can go a long way toward helping riders of the Paris metropolitan area system find their way through the underground maze.