Digital agent Ava guides San Antonio transit riders

Consultants from IBM® Expert Labs collaborated with a team of VIA developers who were working on their first AI project. “There were weekly planning calls about things the IBM team was working on and things our team handled,” says Young. “We had assigned some junior developers and we wanted them to grow. The IBM experts were really great about sharing, saying, ‘Hey, if you get stuck or need help, please reach out to us.’”

After agreeing on the requirements and architecture, the team analyzed call center data to categorize it and pinpoint common queries the assistant would need to answer. The intelligence was provided by IBM watsonx Assistant software, a multilingual conversational agent from IBM Cloud® that can understand questions texted in natural language and answer in kind.

The team next turned to the most common question, “When is the next bus arriving at my stop?” VIA had already created a service that assigns each stop a five-digit code, which riders can text to Swiftly Systems Inc., a VIA partner. Then, Swiftly uses AI to analyze real-time bus sensor and traffic data to predict the next arrival. The developers integrated the Swiftly service into the assistant. They also integrated the API of VIA partner HERE.com, a mapping service that offers location and point-to-point directions.

In addition, there was a need for operational intelligence. VIA managers would profit from understanding the digital customer interactions, including the number of users and conversations and the content of questions. The developers created a dashboard report that is automatically sent to stakeholders, without them having to query the system.

The analytics are powered by the IBM Watson Studio development environment, IBM Cognos® Dashboard Embedded analytics and the IBM Db2® database, all from IBM Cloud. These technologies extract the conversation history from IBM watsonx Assistant and analyze it to determine how well the agent has responded to users. They then present results in a dashboard display that shows business KPIs—demonstrating to executive stakeholders the assistant’s value.

The VIA team named the tool “Ava,” for Automated Virtual Assistant, and after a period of testing and review deployed it in late 2020 on VIA’s website. Ava also is available from VIA’s mobile app, goMobile+. It can answer more than 150 common questions 24x7 in English and Spanish and predicts next-bus arrivals in real time. Ava has become quite popular, conducting thousands of conversations each month.

After releasing the first version of Ava, the developers continue to add to and refine its features. As an example, initially Ava referred users back to the call center when it was unsure of an answer. But this might lose the customer if the call center were closed. Now, through a digital handshake, Ava turns such engagements over to SPS DGTL, a social media management company, which queries users for contact information and communication preferences.

“The assistant kind of knows when it didn’t answer a question correctly,” says Young. “SPS collects user information, including the conversation itself, and turns it over to the SPS DGTL team, so we have a sure way to reach these people.”

Another tweak came on the analytical side. After realizing the power of IBM tools, a VIA developer on her own created a second analytical dashboard, this one for the Customer Care team. It provides a daily accounting of the questions Ava missed or was unsure of. As Young says, “It’s really helpful that the Customer Care team can see what’s being asked and where the gaps are, and then go back and train Ava to become more intelligent.”

Since going live, Ava has conducted more than 41,000 conversations with almost 28,000 unique users, and currently averages nearly 3,000 conversations per month. And Ava provides a response to 96% of users. “I’m thrilled that we’re seeing so many conversations and such strong performance,” says Young.

VIA personnel are pleased with the agent’s impact. “Ava has made it more convenient for people to get information, and ultimately, that improves the customer experience,” says Young. “And the people that use the assistant instead of the call center free up phone lines, lessen queue times and relieve our live agents, which helps us provide better service.”

Young also notes that the project was cost effective, in part because it relied on cloud computing. “We didn’t have to put in any dedicated infrastructure,” he says. “We just capitalized on really smart cloud technology.”

Bigger picture, a key benefit is the knowledge IBM transferred to VIA during the project—knowledge that will serve the organization well over time. In fact, developers are already working through a long list of enhancements.

Richer location and directional services are a priority, using the APIs to answer questions like “How can I get from my home to Walmart?” Or perhaps predicting bus arrivals when users aren’t at a stop, or graphically showing a user’s location and nearby stops. For those who prefer to call, developers are adding a voice interface. The solution will likely be an option in the future for customers calling VIA’s call center. And the system may switch users to a live agent if Ava is stumped during call center hours.

“AI and digital agents were new to our team and to our developers,” says Young. “We wanted to create something that our whole team could support, sustain and grow, and have a partner that would do it along with us. That’s exactly what we gained with IBM.”

Founded in 1978, VIA (link resides outside of ibm.com) provides multimodal transportation to San Antonio, Texas, and 13 member cities throughout Bexar County. VIA makes around 36 million passenger trips each year through bus, paratransit, and VIA Link ride-share services. It also operates vanpool, park and ride, and special event services. Employing around 2,200 people, VIA’s operating region covers seven transit centers, nearly 7,000 bus stops and eight park-and-ride centers over 1,208 square miles.