# Help caseworkers get benefit rules right with an AI assistant

In Nava's evaluation, an AI assistant was estimated to improve caseworkers' accuracy on benefit questions by 40%.

Organisation: Nava PBC (Nava Labs); Amplifi; Gates Foundation
Status: Pilot evaluated
Place: United States
Categories: Financial inclusion & benefits, Government & public services
Try it with: Claude, ChatGPT / OpenAI models, Gemini, Grok
Time to try: 20 minutes

## Why it matters

Benefit rules are complex, and wrong answers mean families miss support they are entitled to. Tested tools that help caseworkers get it right extend the safety net.

## How it works

Nava ran a randomised trial with 125 caseworkers and a 14-week pilot with 61 caseworkers across six organisations in Los Angeles County, inside Amplifi's Benefit Navigator. The 40% is an average, with bigger gains on harder questions. About 65% of caseworkers with access used it (an average of 14 prompts each), usage declined over time, and the Net Promoter Score was 11. Nava said in 2024 that a 3.6 million US dollar Gates Foundation grant, received with two partner organisations, would fund piloting the chatbot.

## Who it's for

Benefit agencies, navigators, NGOs and social-service leaders.

## The prompt

```
You support a benefits caseworker. Answer only from the official program rules I paste below.

Client situation: [household size, income, location, current benefits; no names].
Question: [question].

1. Give the answer in plain language.
2. Quote the exact rule sections you relied on.
3. List any facts you still need from the client before the answer is certain.
4. Flag related programs the client may also qualify for, with the rule that suggests it.

If the rules don't answer the question, say 'Not covered by the provided rules'. Never guess eligibility.

[Paste official program rules]
```

## Caveats

Mixed adoption results; Nava reports both gains and declining usage.

## Sources

The evaluation is Nava's own; the Gates page independently confirms the grant only.

1. Nava PBC: RCT with 125 caseworkers; 14-week pilot; 61 caseworkers; 40% accuracy; 65% use; NPS 11. Evaluating a GenAI-powered assistive chatbot for caseworkers (https://www.navapbc.com/case-studies/evaluating-ai-assistive-chatbot-caseworkers)
2. Gates Foundation: 3,632,912 US dollar grant to pilot AI tools. Committed grant: Nava PBC (https://www.gatesfoundation.org/about/committed-grants/2024/10/inv-079363)
3. Nava PBC: 3.6 million US dollars Gates Foundation grant. Nava Labs awarded grant to pilot tools powered by generative AI (https://www.navapbc.com/news/nava-labs-gates-foundation-ai)

Fact-checked: 16 Oct 2026

Page: https://escapevelocitysociety.com/picks/nava-caseworker-benefits-assistant

See Selected Projects: https://escapevelocitysociety.com/picks.md
