United StatesPublic sector2026 – ongoingProsper Global
Translating, tagging and grading community feedback reports
Prosper Global cut community feedback analysis from up to a week to about an hour in a Claude pilot
Problem
Prosper Global runs more than 200 programs in over 30 countries. Its feedback mechanism receives reports through hotlines, forms, help desks and in-person talks, often in local languages. Staff had to transcribe, translate, theme, grade and route each report by hand, and grading differed between staff.
Approach
The organization used Claude to translate reports into English, tag them by theme and propose a severity grade on a six-level matrix, with a short rationale. Direct identifiers and sensitive content were redacted first. A trained staff member reviewed each output and made the final decision. The work was still in testing, and the organization planned a wider rollout and a link to its Zendesk feedback system.
Results
Takeaway
Keeping a trained reviewer on the final grade let the organization speed up sensitive casework while measuring how closely the AI matched people.
Categories
- Industry
- Public sector
- Business function
- Production and quality
- Innovation type
- Process innovation
- AI technology
- Generative AI
- Adoption
- Off-the-shelf tool or platform
- Evidence
- Vendor case study
Sources
Source summaries
Anthropic's customer story describes how Prosper Global tested Claude to translate, theme and grade community feedback reports, with staff making the final call. It reports feedback analysis falling from up to a week to about an hour, a task going from 8 hours to 30 minutes, and about 75% agreement with human graders on 20 cases. It also gives pilot survey results, including an 88% average time reduction and more than 540 hours saved in one month. The page carries no date; the Internet Archive first saved it on October 16, 2026.
Full details are in the original sources.
