Chatbot Smooths Internal Policy Queries for Community Services District

The Cosumnes Community Services District has adopted a plug-and-play chatbot solution that’s helping its staff parse through a broad range of policy and procurement documents more efficiently.

CCSD manages the parks and recreation and fire departments, as well as emergency services for the cities of Elk Grove and Galt, IT Director Farnam Shadmani explained.

“We have over 100 policies, and finding information in the policies for the staff was very challenging,” Shadmani said. “They had to open multiple policies to find the information they were looking for.”

The chatbot platform assisting with this effort comes from DenserAI, a California-based company specializing in quickly embeddable technology that can pull information from an organization’s webpage or a curated list of documents, such as an internal policy document.

In the three months that CCSD has been leveraging the product, Shadmani said the results have been positive, adding that the system is easy to set up.

“We uploaded all of our policies into the system and created an iframe and posted it into our intranet,” he said.

DenserAI CEO Zhiheng Huang said the platform works in one of two ways — either by scraping information from a specified website, or by sifting through documents the client provides.

“The goal is just to make sure when the chatbot is built, it's not like ChatGPT that can openly answer any question,” Huang said. “This chatbot is dedicated to answering questions based on those files. … If you ask some question unrelated to those files, the chatbot [can be configured] to refuse to give you the answer.”

While models like ChatGPT are often criticized for creating false information or failing to provide adequate sourcing pulled from the broader Internet, the Denser chatbot provides clear citations of where the information it presents came from, Huang said.

“I think that's actually, it's the major advantage of our service compared to ChatGPT,” he said. “If they're asking anything related to their policy, the internal staff will get the answer as well as the sources used to generate this answer. This is actually the most useful feature for the internal use case.”

If the tool can’t locate a piece of information based on the provided prompt, it won’t guess or create an answer and will instead alert the user that the prompt is outside of its search parameters.

The tools also provide a lead generation feature for public-facing chatbots, Huang explained, adding that it can catalog name and contact information inputs from users for future outreach depending on how the tools are structured on the back end.

Similarly, Huang noted that query logs are available for staff to review within the back-end dashboard, offering deeper insights into the types of requests that internal or external users are looking for. From a security perspective, Huang says his team is not allowed to access the information provided to the system without express permission from the client.