Commentary: Bots and Live Agents Both Have a Place in Government

Implementing a chatbot or virtual assistant bot is almost required today, no matter the type of business or public agency.

In a recent review of a county website in Texas, it took me some time to find information on tax payments. I also had to navigate to additional sites (URLs). This is where a bot can help reduce searching. However, it appears that it takes time to make it right. As I have mentioned in a past article, the California Department of Motor Vehicles (DMV) has had much success with their chatbot implementation, using the QnABoton AWS Solutions.

According to the California DMV, their chatbot, Miles, has responded to 1.5 million interactions in the first half of 2022. Miles can now also manage website interactions as well as engagements through the customer service phone line. The DMV had this to say: “The DMV is continuously working to improve Miles the Chatbot so it can serve more customers effectively and save them time, whether during or after normal business hours.”

The process will always be evolutionary. This is not a new concept; things change, which has always been the case in how tasks have been accomplished. Before, technology was all about 0s and 1s; now, with artificial intelligence and machine learning, the technology builds in the flexibility of change. It makes the interactions dynamic!

Most county and city jurisdictions are informational bots, which are more useful and efficient for local governments than live agents. The primary goal of an informational bot is to provide information and resources quickly – to serve as a rapid navigation tool. For example, asking a question and quickly getting a response is essential in an emergency situation. Government sites typically make all that information available, but searching and navigating is tedious and can take more time. Asking a question and getting a response, even if only to navigate to another page, is much quicker. This type of bot requires the most effort in populating and sourcing data.

At Placer County government, for Bob the Chatbot, we use a tool recommended by our content management system (powered by CivicPlus), Frase Answers by Frase.io. There are other bot sourcing tools, such as Amazon Kendra, which can scan not only pages but documents to generate a repository of information that can then be provided as a source to an endpoint, such as a website chatbot or telephone system. Amazon Kendra can be implemented with many solutions and has a seamless integration with QnABot.

For the California DMV for example, a more targeted approach is required, and both chatbot and telephone endpoints are leveraged. The agency can build its own question-and-answer database for the chatbot, which is a more specific approach in combination with the already scanned information in the repository.

OK, here comes the next layer: the live transfer to an agent, or “agent assistance.” This layer falls beyond and usually cohabitates with the information bot. This is the most difficult to implement and has several different use cases. Making sure the user is directed to the most resilient effort is beneficial to both the user and the provider, reducing efforts on both sides of the conversation, not to mention cost, time and customer satisfaction.

One might ask whether the public sector should downplay the level of customer satisfaction. I do not think we in government intend to; it just requires agencies to elevate the expectation of a better user experience than before. I do not believe there is great value in live bot services for government or public agencies, outside of a handful of specific use cases. The abundance of information and providing a bridge between what the public knows or understands versus the technical terminologies can be handled solely with an informational bot.

At Placer County, the product by Frase.io scans all content on our website, via site mapping. A similar approach can be accomplished using Amazon Kendra. It also has the benefit of scanning additional URLs (additional websites). This ability allows us to pull in data from published GIS sites on ArcGIS as well as other domains and provide tailored responses for more detailed information, some dynamic in nature such as emergency information, road closures, weather conditions, etc. Once this scanning methodology is configured, we can begin to map or train customer questions to the proper responses, i.e. the “technical terminology.” This bridges the gap between day-to-day phrases and legal/technical government vocabulary.