Travis is named for his travel department’s motto, “Travel Is Simple.” He was built by Michael McSperrin, the Krakow, Poland-based global head of facilities and support services for talent acquisition and management consultancy Alexander Mann Solutions.
The new chatbot provides automated responses to written questions posed by Alexander Mann Solutions employees about travel policy and process. McSperrin built the bot using a Microsoft platform called QnA Maker. “You can create a database of questions along with corresponding answers, and you can interact with it on different platforms,” said McSperrin. Alexander Mann Solutions is deploying Travis mainly on Skype for Business, which offers messaging in addition to the video-call service for which it is better known. Travis also will deploy on a Web platform.
Fifty travelers at Alexander Mann Solutions have participated in pilots of Travis since January, and the bot is scheduled for companywide launch in July.
AMS has 2,750 employees in EMEA, and 600 each in the U.S. and Asia/Pacific. It spends an annual seven-figure euro sum on travel. About half the company’s employees travel each year. Global online adoption through travel management company Egencia is north of 90 percent. There is in-country support from Egencia in Poland and the U.K., which account for 50 percent of the global workforce, and a regional reservations center in Belgium for those based elsewhere in EMEA. English is spoken throughout the company.
Changing Communications
McSperrin does not consider himself an expert technologist, but he used QnA Maker without any training. He did bring in the company's IT team, however, to connect Travis to Skype for Business.
McSperrin introduced Travis because “it’s just the way the world is working these days,” he said. “We find people want information instantly and quickly just like they have on their mobile phones. Our top 30 travelers generally know what to do, what the policy is and how the system works, but employees who travel only once a year or even once a quarter are not necessarily familiar with them. We were getting a lot of questions from people who, for example, didn’t know how to log into their profile or who was going to make the approval for their booking—very simple, basic questions.”
The company has a SharePoint intranet site with a static document answering all the most common policy questions, but “we were still getting a lot of these questions coming through,” McSperrin said. “People tend to rely on someone knowing what they need to know rather than going looking for it. Instead of logging on to a portal, going to the right page, finding the right document and looking through it, they would much rather ping a message to someone and rely on them to give the answer.”
He added, “We did direct people to the website, but people either wouldn’t look or wouldn’t find the answers they were looking for because they only skim-read it.”
What Goes Into a Chatbot Database
Alexander Mann Solutions was investigating artificial intelligence and robotics to improve interaction with customers when McSperrin discovered QnA Maker and saw an opportunity to use similar technology for in-house communications.
Initially, McSperrin uploaded the SharePoint Q&A to create the Travis database, but early testing with travelers taught him how much more was required, especially the need to recognize variations in vocabulary and phrasing: the different ways to ask the same question.
Inputting variations has proved time consuming. McSperrin has also programmed small talk—such as responses to “How are you?”—into the tool to make interaction natural and, therefore, warm employee attitudes towards the tool.
But is it really worth all the effort? It’s too early to answer that question with statistical evidence, but McSperrin said, “Ultimately, yes. We recognize some time needs to be invested into it, but in the long term, we believe it will pay off. The time saved on answering routine questions can be redirected to answering complex questions.”
If Travis cannot find an answer, it advises the employee to contact the travel team, and QnA Maker logs the questions it is unable to answer. McSperrin and colleagues write responses and upload them to the bot in real time if the questions are relevant for a wider audience.
Where to Go from Here
Travis has its limitations. It is not a fully fledged chatbot. “It can’t validate,” said McSperrin. “For example, if a traveler asks, ‘What are our preferred hotels?’ It can’t come back and say, ‘In which location?’ Farther down the line, we want to build Travis out to be a more intelligent bot so that it can ask those verifying questions.” McSperrin said a Travis 2.0 of this kind would require a switch from QnA Maker to other, more sophisticated Microsoft chatbot tools.
Before going deeper with its chatbot development, however, Alexander Mann Solutions is likely to go wider. The company plans to use the experience of creating Travis to build similar tools for answering frequent staff questions about topics such as finance and HR.