Streamlining asylum decisions with AI-powered transcript summarisation
Publicly sourced
The challenge
The Home Office faced significant time burden on asylum decision-makers who must read and analyse lengthy interview transcripts, often extending to 50+ pages per case, creating high cognitive load and time constraints that slow processing of asylum cases.
The solution
The Asylum Case Summarisation system uses OpenAI's GPT-4 to automatically convert lengthy asylum interview transcripts into concise summary documents. Built internally by the Home Office, the system operates as a human-in-the-loop tool that assists decision-makers with faster case familiarisation while maintaining human control over all asylum decisions.
The results
Two 8-week pilot phases (May-June and September-October 2024) across two Decision Making Units evaluated the system with test and comparison groups. Phase 1 involved 60 decision-makers using AI summaries and 15 using traditional methods; Phase 2 involved 45 using AI and 15 using traditional methods. The test group processed 334 cases with AI summarisation whilst the comparison group processed 95 cases using traditional methods, demonstrating efficiency gains alongside accuracy challenges:
23 minutes saved per case on average (32% time reduction)
65% of respondents reported they would continue using the tool
Asylum Case Summarisation (ACS)
Streamlining asylum decisions with AI-powered transcript summarisation
The challenge
The Home Office faced significant time burden on asylum decision-makers who must read and analyse lengthy interview transcripts, often extending to 50+ pages per case, creating high cognitive load and time constraints that slow processing of asylum cases.
The solution
The Asylum Case Summarisation system uses OpenAI's GPT-4 to automatically convert lengthy asylum interview transcripts into concise summary documents. Built internally by the Home Office, the system operates as a human-in-the-loop tool that assists decision-makers with faster case familiarisation while maintaining human control over all asylum decisions.
The results
Two 8-week pilot phases (May-June and September-October 2024) across two Decision Making Units evaluated the system with test and comparison groups. Phase 1 involved 60 decision-makers using AI summaries and 15 using traditional methods; Phase 2 involved 45 using AI and 15 using traditional methods. The test group processed 334 cases with AI summarisation whilst the comparison group processed 95 cases using traditional methods, demonstrating efficiency gains alongside accuracy challenges:
Details
Links