Commentary: Speaking of Chatbots, How Is DMV’s Miles Doing?

Let me just start with a “Wow!”

About two years ago, I reviewed Miles, the California Department of Motor Vehicles’ chatbot — and here we are still talking about it. This is great news! In late 2018 and early 2019, chatbot technologies including natural language processing and machine learning were visionary, just beginning development within many government organizations. Chatbots were viewed as a question: “Why?” The return on investment (ROI) was not completely understood, and there were no analytics to justify the cost for government agencies. Fast forward to today and we are now able to review analytics that can provide us answers to the why and show us the ROI.

According to the California DMV, of the 3.5 million sessions Miles handled between August and October 2021, 67 percent interacted with Miles during regular business hours. This leaves an amazing 33 percent of interactions that Miles was able to support after regular operations. Miles provided answers although no live agents are available after hours, when people might have more time to ask them. Of the 67 percent chatting with Miles during business hours, an amazing 94 percent were handled by the chatbot, and only six percent escalated the session to speak directly with an agent.

I must stop here and give credit where credit is due: The DMV has really turned things around. I remember waiting in lines at the DMV that wrapped around buildings 25+ years ago. Now, people are getting answers to questions immediately with Miles, at any time of day. DMV kiosks are going up around U.S. post offices to handle registration renewals. These are great examples of how the DMV is providing residents flexibility and working around their busy lifestyles.

Now back to how Miles is speeding down the highway. Recently I reached out to the DMV Office of Public Affairs and asked for some feedback. Outside of the analytics I mentioned, they provided some more great details:

What happens when Miles can’t answer a question?

“If Miles is unable to provide a proper response to the customer, the chatbot will bring up a confused state (saying that it’s still learning and spinning its wheels) and proceeds to offer the customer the 1-800 number for further assistance,” the DMV Office of Public Affairs responded.

This is extremely important to understand, like the old saying, “With age comes wisdom.” Chatbot AI is eternally learning ¯ especially in government, where rules and regulations are ever changing. Read on to see how this is resolved by Miles, with the same technology used by Placer County.

What advancements have been made with Miles? 

“The team regularly adds new dialogs based on reviewing and analyzing customer intents. If the team notices certain issues/topics trending in the customer intents, new dialogs will be created to address those. Additionally, 13 languages were added to translate the current dialogs,” the DMV Office of Public Affairs replied.

As you read in the first question/comment about what happens when there is no answer, the bot logs the question as unresolved. There is no other way to help improve results without this important feedback loop. Again, at Placer County we follow this same methodology. It is essential, as mentioned in a prior article, to provide more dynamic results during an active emergency. Placer County was able to achieve this during recent fire events.

How has Miles been incorporated with the contact center?

“The live chat teams reside within the contact centers, so when a customer needs to speak with an agent after talking with Miles, they are directed to a contact center live chat team member,” the DMV Office of Public Affairs said.

This live agent concept is no different from the private sector — shopping sites, for example, where you need more specific assistance than the virtual bots can provide. Live agent escalation, seamless in the bot, is one theory that many bot solutions are starting to offer.

To summarize, now that we see the effectiveness and safety at this level of AI — and remember, there are four major categories of AI — expect to see more chatbot solutions popping up at all levels in government. The bottom line is do you want to wait on hold or search for answers for hours, or ask a chatbot and get a response in seconds? Even if the bot spins its wheels the first time around, you will likely still find the answer faster than researching on the web, calling or waiting in the office!