If you’ve ever sat on a sub­way under­ground when your train grinds to a halt for no appar­ent rea­son, you’ll know how crit­i­cal it is to get infor­ma­tion to fig­ure out what to do next. And how frus­trat­ing it is when you don’t. In our world where we’re accus­tomed to hav­ing every answer at our fin­ger­tips, it’s more impor­tant than ever for mass tran­sit sys­tems to deliv­er help­ful infor­ma­tion, in real time, to their rid­ers. Here come the arti­fi­cial-intel­li­gence-pow­ered chat­bots to our rescue. 

AI Takes the Train in Paris

Take this tran­sit sce­nario to a rail sys­tem as com­plex as what’s in place in Paris, where 3.2 mil­lion pas­sen­gers ride 3,200 trains each day on the SNCF lines between the city cen­ter its sprawl­ing sub­urbs. That’s sec­ond only to the num­ber of trains that run dai­ly in Tokyo.

At the 2018 SAP­PHIRE NOW and ASUG Annu­al Con­fer­ence, we had the chance to speak with Olivia Fis­ch­er who’s lead­ing an arti­fi­cial-intel­li­gence-dri­ven chat­bot project with the dig­i­tal media team at the Paris region branch of SNCF. (In case you’re won­der­ing, SNCF stands for the Société nationale des chemins de fer fran­cais, or in Eng­lish, the French Nation­al Rail­way Com­pa­ny, and has been oper­at­ing since the 1930s.)

A Per­son­al­ized Expe­ri­ence for Pas­sen­gers at Scale

SNCF is build­ing a chat­bot pow­ered by SAP Leonar­do Recast Con­ver­sa­tion­al AI to help dai­ly trav­el­ers plan their trips and nav­i­gate any of the inevitable delays that hap­pen on a rail sys­tem. Accord­ing to Fis­ch­er, it’s all about deliv­er­ing a good cus­tomer expe­ri­ence to every­one. ​“Our obses­sion is help­ing peo­ple find the right infor­ma­tion about where to go at the right time, when they need it, what­ev­er the device they’re using, what­ev­er their sit­u­a­tion,” she explained. ​“This may be the sub­ur­ban peo­ple going to work every day, but it’s also tourists. As pub­lic trans­porta­tion, we must take care of every cus­tomer. That’s where our chat­bot is real­ly inter­est­ing for us, because it helps us deliv­er tai­lor-made ser­vice and per­son­al­ized infor­ma­tion at an indus­tri­al­ized scale.” 

Teach­ing the Baby­bot About Transportation

Devel­op­ing the ini­tial chat­bot took only about nine weeks with help from Recast Con­ver­sa­tion­al AI. Fis­ch­er called it the ​“baby­bot, because it’s almost like a baby. You have to raise it and teach it how to speak.” 

They launched this baby­bot on a small scale to the SNCF agents who work in each train sta­tion, as well as to a few cus­tomers who were ear­ly adopters. In that first wave of a few hun­dred peo­ple using the chat­bot dai­ly, SNCF learned a lot about how peo­ple inter­act with these bots. ​“Peo­ple are both sus­pi­cious about using a bot with AI, but also have real­ly high expec­ta­tions,” Fis­ch­er recalled. ​“Because peo­ple are used to a cer­tain lev­el of con­ver­sa­tion and don’t under­stand the lim­its of a bot service.” 

The bot had to be trained using both deep learn­ing and hands-on instruc­tion. ​“It’s not a self-learn­ing robot,” Fis­ch­er explained. ​“We are not to that extent of AI yet. So that means we have a mod­er­a­tor who tags every ques­tion, and then the bot attrib­ut­es a lev­el of con­fi­dence to how well it under­stands the question’s inten­tion.” Above a cer­tain lev­el of con­fi­dence, the bot will answer with­out refor­mu­lat­ing the ques­tion or ask­ing for more information.

The Spaghet­ti Plate of Lega­cy IT

Fis­ch­er recount­ed the biggest chal­lenges for this project as, ​“the nat­ur­al lan­guage pro­cess­ing for trans­porta­tion terms, which are very spe­cif­ic, and con­nect­ing the bot with our spaghet­ti plate of IT lega­cy sys­tems. Because SNCF is an old com­pa­ny, the IT was built in lay­ers so there are bits and pieces every­where. We’ve been doing some work to reor­ga­nize and inte­grate this data, so the com­pa­ny can use it more easily.”

A Mul­ti­tal­ent­ed, Mul­ti­lin­gual Chatbot

The chat­bot is cur­rent­ly capa­ble of speak­ing to its users in French, Eng­lish, and Span­ish. This will come in handy for the legions of tourists who will vis­it Paris for the Sum­mer Olympic Games in 2024, and pri­or to that for the Rug­by World Cup in 2023. ​“We expect a lot of tourists for these,” Fis­ch­er explained. ​“The bot will help us to deal with this pop­u­la­tion and the lan­guages that would be very expen­sive to devel­op with­in a tra­di­tion­al mes­sag­ing platform.” 

The chat­bot also helps pro­vide acces­si­ble ser­vices to rid­ers who are hear­ing impaired, vision impaired, and more. It’s strict­ly text-based at the moment, but Fis­ch­er said they’re explor­ing use cas­es that may involve voice commands.

A Bot With Per­son­al­i­ty

SNCF made the choice to bring some human­i­ty to some­thing that is basi­cal­ly just a robot. Users who inter­act with the bot will find it makes small talk and lit­tle jokes. ​“When you say, ​‘thank you,’ the bot answers with smi­ley emo­jis, ​‘Thank you, I do my best,’” Fis­ch­er recalled. ​“Nat­u­ral­ly, peo­ple, when they deal with the bot, they want to chat.”

But all inter­ac­tions aren’t marked with smiles. There are a lot of prob­lems that arise in pub­lic trans­porta­tion that make peo­ple angry. ​“Espe­cial­ly for pub­lic trans­porta­tion, real-time infor­ma­tion is always becom­ing obso­lete,” said Fis­ch­er. ​“To deal with these lim­its and the hos­til­i­ty of some cus­tomers, the bot has to express its lim­its and admit that it’s a bot.”

No Threat to Agents

Through­out the bot devel­op­ment process, SNCF has remained trans­par­ent about its aims for the bot and how it can help its human agents. ​“The bot can­not be an SNCF agent. It’s a bot, and it stays in its place. It deliv­ers a cer­tain ser­vice very well, but then, for the real rela­tion­ship, you have agents,” Fis­ch­er explained. ​“You can call the call cen­ter. You can go to sta­tion and speak to real peo­ple. It’s two dif­fer­ent expe­ri­ences.” Fis­ch­er has actu­al­ly found that the chat­bot alle­vi­ates some of the frus­tra­tions SNCF agents expe­ri­ence from answer­ing the same ques­tions over and over. 

The Train Trip of the Future

SNCF plans to launch the bot with its full capa­bil­i­ties in ear­ly 2019. As the chat­bot evolves, that’s where it brings real val­ue to SNCF’s cus­tomers. ​“To under­stand the real behav­iors and needs of the pas­sen­gers — if they need traf­fic info or an itin­er­ary. When do they need to be informed? That’s the real work, and that’s real­ly the core of the chat­bot,” Fis­ch­er said. 

Fis­ch­er and her team hope to inte­grate the bot with SNCF’s social media pres­ence. Cur­rent­ly, every train line in the Paris region already has a Twit­ter account that deliv­ers infor­ma­tion as quick­ly as pos­si­ble. For exam­ple, the com­mu­ni­ty man­ag­er can update the Twit­ter account with infor­ma­tion about a slow­down, then the bot could deliv­er per­son­al­ized infor­ma­tion to users who want to reroute their trips in response. 

“What’s real­ly impor­tant for me is how the bot is cre­at­ing a big rev­o­lu­tion in cus­tomer infor­ma­tion in trans­porta­tion,” Fis­ch­er said. ​“It’s going to bring real­ly great ben­e­fits where we can go to them with the infor­ma­tion and speak in nat­ur­al lan­guage. That’s a real revolution.”

Look­ing for inno­v­a­tive ways to enhance your cus­tomer expe­ri­ence with SAP tech­nolo­gies? Join us for the Cus­tomer Expe­ri­ence Inno­va­tion Forum, our newest event for retail­ers, con­sumer prod­ucts com­pa­nies, and wholesalers.