Shelter, a charity supporting people facing homelessness and bad housing, wanted to analyze live chat transcript data to better understand user needs and improve their services.
Their Client Data Insight team sought to extract key insights from conversations, including identifying recurring issues, key phrases, and user sentiment. They also aimed to leverage machine learning and cloud-based tools for scalability and automation.
The Curve collaborated with Shelter to develop data science techniques and machine learning models that fit their Python-based workflow and Azure cloud strategy. Using natural language processing, the team extracted key phrases, locations, and organisations from chat transcripts. A machine learning model was trained on prior transcripts to classify user concerns into categories such as Goal, Problem Area, and Tenure, achieving an accuracy of approximately 85%.
By implementing AI-driven analysis, Shelter can now gain deeper insights into service users’ concerns, optimise their live chat services, and automate data analysis for future scalability. This project has enabled them to make data-driven improvements in their mission to defend the right to a safe home.
Software Development
“This project has helped to massively accelerate the team’s capabilities in terms of the tools that we have been able to apply in various new ways and has helped us to support the web chat and other teams within Shelter to get more value from their data and a richer understanding of the help they provide to their clients”
Dean Robinson
Client Data Insight Manager, Shelter