• AI for Good
• Digital Inclusion
• HealthTech

Written by Andy Bell, Founder & CEO, HelpFirst

Working with NSPCC’s Childline service has opened my eyes to a challenge that is all too common in public services and the charity sector: how to help dedicated staff and volunteers spend more time supporting service users, and less time on admin?

Every year, Childline’s dedicated counsellors have hundreds of thousands of interactions with young people – through a combination of web chat, voice calls, and personal inbox messages. Each interaction must be carefully documented and risk assessed. This is vital for making sure the needs of any child or young person that reaches out for help remains at the heart of what Childline does. Good documentation is core to their holistic approach to safeguarding; it saves the young person having to repeat themselves if they come back again, and, in the rare instances where Childline need to take action to keep a child safe, it means they can share accurate information with external authorities to protect those that need it most.

But here’s the challenge: writing up a typical counselling session can take a long time, especially because staff and volunteers know the importance of getting it right. Tens of thousands of hours are necessarily spent annually on documentation, when there are always more children and young people that could be reached.

Earlier this year, HelpFirst conducted a Proof of Concept project with Childline to explore how AI could help. The goal wasn’t to replace human judgment, connection or empathy – those remain essential – but rather to assist with the time-consuming tasks of summarising the content of a conversation and suggesting relevant coding categories (which are a vital way of how Childline evidences the voice of the child and the impact of its services). We experimented with different AI methods to generate draft write-ups that  could be reviewed and modified.

The results exceeded everyone’s expectations. The AI-generated summaries achieved over 90% accuracy on Childline’s quality assurance metrics. Summaries could be generated in seconds at a cost of pennies, and Childline staff estimate that, if fully operationalised, this has the potential to reduce write-up time by up to 80%.

What does this mean in real terms? The time saved is estimated at enabling Childline to deliver up to 60,000 additional counselling sessions with children and young people per year (a 30% increase in capacity). 

Beyond the core task of session write-ups, we’ve identified other valuable applications of this technology. AI could provide quick summaries of a young person’s contact history at the start of a session, it could support more in-depth reviews for Childline’s regular service users, and generate live summaries of ongoing interactions to support those managing Childline shifts. The NSPCC is also looking at how these same principles could also apply to its NSPCC Helpline, where adults with concerns about children or young people can contact specialist staff for advice, guidance and support.

My three take aways:

1. Start with service design. While AI technology is advancing rapidly, the real challenge lies in understanding user needs and designing solutions that truly serve them. We spent  considerable time working with NSPCC staff to define exactly what makes a good summary and how to structure information most usefully.

1. Custom summaries are much better than off the shelf solutions. Tools that create a general summary from a transcript or set of case notes can miss crucial information, hallucinate (make things up) or be confusing. Some LLMs are more familiar with the kind of language seen in counselling (and similar services) than others. But more importantly, customising the model with prompt engineering and fine tuning can make a huge difference.

1. Enabling staff to speed up, not full automation. Staff add the most value by using their experience to weigh up risk and protective factors and make decisions about what action to take. Our approach focused on using AI to handle the time-consuming tasks of gathering and organising information, freeing up staff to focus on direct interactions, analysis and judgment. It’s essential to retain the human element of checking AI-generated content, with the ability to make changes and improve the system.

Read More AI for good

Comments are closed.