The Nava Labs team started by conducting extensive research with caseworkers and community partners to inform the project’s direction and chatbot details. This helped the team build trust with the caseworkers while ensuring they built a product that meets people’s needs. Then, the team designed the chatbot to write its responses based only on the specific documentation that caseworkers are trained on — like state-specific policy manuals — to reduce the likelihood of hallucinated answers. 

The chatbot also provides direct-quote source citations for its summarized answers. This enables caseworkers to see the source material in context and ensure the answer’s accuracy — another safeguard against misleading summaries. If there’s an error, the Nava Labs chatbot allows caseworkers to give feedback on the quality of the answer. This enables the team to quickly address any issues and avoid similar problems in the future. 

During the pilot, the caseworkers at partner organizations were not required to use the chatbot; instead they could try it on their own volition. If caseworkers continue using the chatbot, it’s because they think it’s trustworthy and helpful. This can also help the chatbot spread via word of mouth, with caseworkers who find it trustworthy recommending it to colleagues.

To learn more, read our case study.

When a Bay Stater gives birth, they’re eligible for medical leave and family leave to bond with their child. However, some eligible applicants aren’t opting into both medical and family leave when they apply, which can create administrative burden for applicants and DFML staff.  

To help applicants choose the correct type of leave when they apply, DFML is working with Nava to conduct usability testing on changes to the PFML application. With each round of testing, DFML is getting more targeted in ways to improve the application and ensure applicants get the leave they’re eligible for. 

We’re also helping DFML update how employers request exemptions from the state PFML program. Previously, this experience was external to DFML’s systems. We’re helping DFML build and enhance the experience in their employer portal, which will improve experiences for employers and DFML staff.  

DFML worked with Nava to run several rounds of usability testing and a pilot with employers and DFML staff. This process enabled DFML to gather early feedback, test data integrations, and de-risk the exemptions process before they launch it to all employers. 

To learn more, read our case study.

Claimants previously shared that it was challenging to understand their claim status and next steps after they applied. So, we consulted UI experts at NJDOL on how to give claimants more visibility into their claims and context when benefits are delayed. Together, we identified portal features to design and test with claimants, and then iterated based on the feedback.

These features alert claimants of key next steps, like verifying their identity, starting to certify for benefits, or responding to email requests from NJDOL. The portal also provides information like claimants’ weekly benefit rate and maximum benefit amount, and claim status updates such as temporary disqualification information, eligibility issues, and payment history. We helped NJDOL populate the portal by integrating data from multiple backend systems, and we worked with NJDOL to present information in a user-friendly format.

We also helped NJDOL open up the portal’s new intake application to 100% of claimants. Because NJDOL works iteratively, they initially launched the application to standard claimants, which excludes less common “fringe claims.” Launching this way delivered the most value to the most claimants the fastest. To launch the application to 100% of claimants, we helped NJDOL update screener logic, the application’s business logic, and application questions to account for fringe claim requirements.

The updates to the claimant portal make it easier for people to apply for benefits, understand next steps, and keep up with their claims status. In turn, NJDOL helped reduce burden for call center staff and claims processors.