Born of pandemic, Georgia’s chatbot opened another avenue to answers
The onset of the Covid-19 pandemic created a sudden and unprecedented need for timely and accurate information. With cases rising exponentially, everyone from medical professionals to average Georgians looked to their public health departments for answers.
In the beginning, it was questions about symptoms, how the virus spread, treatment, and local case numbers. Phone lines and email inboxes were jammed. There was no way to keep up with the surge in those early weeks. Our team got together to discuss whether there might be a technology solution that could help improve constituent experience and provide in-demand information.
We dusted off a dormant project plan for a chatbot and rushed to implement it on the Department of Public Health’s website and on Georgia.gov. The chatbot would allow users to type questions in their own words as if they were talking to a live person and get up-to-date answers with links to reputable sources. By providing a way to get answers to the most frequently asked questions, the chatbot could free up agency staff to handle more complex issues. At least that’s how it was supposed to work.
By answering simple questions online, the chatbot would make it easier for call center agents to focus on more complex issues.
Now that some time has passed, we thought it was a good time to reflect on our experiences from the project – how we used Microsoft QnA Maker to develop and deploy a chatbot, as well as some of the challenges we faced, lessons learned along the way, and ideas for the future.
In an effort to provide quick, efficient service to constituents, we needed a chatbot that would answer commonly asked questions. It would be available 24/7 to provide information on a variety of topics. In addition to providing another avenue for digital users to get information, we believed the chatbot would also help to relieve some of the pressure on call centers. By answering simple questions online, the chatbot would make it easier for call center agents to focus on more complex issues. Ultimately, the goal was to make the most efficient use of agency resources and help constituents get the information they needed as quickly as possible.
Here’s What We Did
The first step was to collaborate with the agency call center and web teams to determine the most commonly asked questions suitable for an automated chatbot. We used a spreadsheet to organize and format the QnA (Question and Answer) pairs and keywords. Next, we developed a style guide to chatbot responses. The goal was to put informative, helpful, and concise responses in the chatbot to respond to users in a quick and friendly manner.
We launched with up to 30 QnA pairs, which addressed the vast majority of constituent questions. The responses answered frequently asked questions and/or directed users to a webpage with more information. The chatbot read keywords to match user queries with the most relevant QnA pair in the Knowledge Base.
As we were developing and formatting the content, we kept mobile-friendliness, reliability, and accessibility top of mind. We also partnered with state agencies to provide them with technology and analytics support. Agencies received a monthly analytics report that detailed chatbot use, daily volume, top webpages, and devices. This allowed agencies to update their chatbot content as necessary.
Agencies Who Needed It the Most
The chatbot was used by agencies most affected by service changes due to Covid-19. The first agency we served was the Department of Public Health (DPH) to answer questions about Covid-19. We also enabled this bot on the Georgia.gov website. However, after a few weeks, virus-related questions began to die down as information became more widely available.
The next immediate need was for the Department of Labor (DOL), as people began filing for unemployment in record-breaking droves. DOL reached out to us for help on a Friday afternoon, and by the following Monday we delivered an operational chatbot. It wasn’t perfect, of course, but it provided a much needed resource in a time of crisis. Additionally, we launched a bot for the Department of Driver Services (DDS), which had seen its in-person operations upended by pandemic protocols.
Benefits of Chatbots
Chatbots are easy to implement and can provide a quick response to user queries. They are also easy to update with new information as situations evolve, making them ideal for providing timely support during a crisis. In addition, chatbots can answer an unlimited volume of the same questions, making them a valuable resource for users who need reliable information. Finally, chatbots are quick and easy for constituents to use, making them an accessible and convenient tool for those seeking help and support.
We've learned a lot of lessons about chatbots. Right away, we committed to launching an MVP (minimum viable product) rather than spending precious time during a health crisis working and reworking a customized product until it was “perfect.” There were no drawbacks to doing this. So desperate were Georgians for information, anything that gave constituents another avenue into state government was seen as a win.
Another lesson, which may seem obvious, is that the bot is only as smart as those tending to it behind the scenes. Daily analysis of what users were asking and how they were asking drove continuous updates to the bot’s knowledge base. As we programmed the bot’s “brain,” we gained a better understanding of how it needed to complement the website’s overall content strategy. Writing for a chatbot is different from writing for a webpage, which is different from writing for print. We began to see that users relied on the chatbot to direct them to deeper sources of information elsewhere.
Writing for a chatbot is different from writing for a webpage, which is different from writing for print.
We also learned that many users are more comfortable interacting with a chatbot than they are calling a phone hotline or sending an email. In many cases, their questions could be more immediately answered without being stuck on hold for a long time.
The Covid-19 chatbot was a great example of how technology can be used to help people during times of crisis and confusion. By providing accurate and up-to-date information, as well as a quick and easy way to get that information, chatbots helped reach more people faster, making a difference for many people throughout the state.
At first, we didn't really know how well a chatbot would work or how people would respond to using it. We simply needed to create something that would provide answers to people in crisis during the pandemic. But as we have started exploring the possibilities, we realize that a chatbot can be much more than just a simple assistant. A chatbot could be an essential part of any agency’s customer service strategy, as we spoke about at Spring 2022’s GOVTalks.
To that end, we are currently testing a new Ivy chatbot on the DSGa site to help our users self-serve and for us to evaluate it as a proof-of-concept before offering it to agencies. Ivy actually uses artificial intelligence, automatically extracting knowledge directly from web pages and other sources of information such as PDFs, whereas the Microsoft QnA Maker required everything to be entered by a content manager. We are excited to explore the potentials of this new tool and will keep you apprised of our findings.
The truth is, people are no longer satisfied with mediocre experiences. They expect more from brands and businesses and yes, even government agencies. So, there is a huge opportunity for agencies who are willing to explore this new frontier, and we are happy to help them understand how to create exceptional chatbot experiences and to embrace this technology so that they can improve their ability to reach constituents.