USDA Chatbot Prototype: Using Artificial Intelligence to Connect Customers to Knowledge at Scale
By The USDA Customer Experience Center of Excellence Team
During the Customer Experience Center of Excellence (CX CoE)’s research on the USDA Farm Loans experience, we observed that both producers and USDA staff tended to not use the search functionality within USDA.gov when looking for information on USDA websites. This was further corroborated by our web analytics work during our Web Modernization project, where we found many instances of ineffective content tagging and unnatural site navigation.
In an effort to improve USDA’s search experiences and make it easier for customers to find information, we explored how a chatbot could improve results when searching for information. In order to do this, we needed to find a good use case with a substantial amount of data.
Developing the Bot
An area that provided a large amount of relevant information for this opportunity came from the Farm Service Agency’s AskFSA knowledge base, choosing questions that represented the breadth of content on the site, and ones whose question and answer content would fit well with the chatbot interface. We focused on about 75 “intents,” which is the industry term for a users intentions when asking questions. (For example, “What is the PLC payment formula?” or “How do I calculate PLC payments?” are two separate questions but just a single intent.)
While the team’s data scientists were creating the infrastructure and training the chatbot, our CX researchers interviewed USDA customers to understand what they did and didn’t find useful in a chatbot. Working from these user insights, our designers created the chatbot’s appearance, personality, dialog flows, and other interface and experience elements (including, designing reporting dashboards).
Furthermore, we adapted the knowledge articles to the chatbot format, trimming them down and fitting them to a multi-stage question format.
Caption: Examples of the chatbot icons being developed.
Caption: Mockups of the chatbot reporting dashboards.
Not yet deployed, the USDA Chatbot currently exists in a sandbox/test environment that mimics the USDA development environment. However, recent testing shows that using this functionality shows great promise for enhancing USDA’s search and knowledge base features.
As chatbot work and development continues, we anticipate that the public will be seeing more chatbots at USDA and throughout the federal government. As this happens, we’ll continue to assess how we can deploy chatbots using artificial intelligence and natural language processing to meet the following end goals: better answering customers’ questions; being able to scale and lower costs; and freeing up staff to focus on issues truly requiring human judgment and knowledge.
Caption: Screenshot of the USDA Chatbot prototype on a mockup of the USDA website.
Return to Updates