Following our recent success building the UK Government's first ever public-facing AI-enabled self-service for the Foreign, Commonwealth and Development Office (FCDO), Caution Your Blast Ltd (CYB) kicked off this year by partnering with Homes England to prototype another AI-assisted government service. 

The goal? To empower customers of the Help to Buy Equity Loan scheme to self-serve information on GOV.UK, reducing the burden on their busy customer service team.

Why are customers struggling to self-serve?  

The Help to Buy Equity Loan scheme - a government scheme for first-time buyers in England - is pretty complex. It involves a lot of intricate information regarding eligibility, loan amounts, interest rates, repayment terms, fees, and the impact of property value fluctuations on the loan. As a result of this complexity, the information on GOV.UK is necessarily detailed and rooted in policy. 

While GOV.UK is a treasure trove of essential information, its very nature (detailed and comprehensive) can sometimes become a barrier. Users looking for specific answers within policy-heavy content often struggle to see their own situation reflected in long-format pages. On top of this, the language and structure may not always align with how a customer frames their specific questions or understands their situation. The result of this is a natural inclination to pick up the phone, even when the answer is readily available online.

How can AI help?

We prototyped an AI-assisted triage tool which helps users find the information they need in GOV.UK by breaking down long-format pages into short ‘articles’, which link out to the more detailed guidance on GOV.UK. When a user asks the AI a question, it doesn't generate any answers. Instead it selects the most relevant pre-written articles by analysing the user's query and identifying the key topics and the underlying context. 

To make the response feel more personalised, each article is associated with a variety of pre-written headings that the AI can choose from. These headings are designed to cover a range of ways that users might phrase their questions. This nuanced approach makes the self-service experience feel more tailored and relevant to the user’s personal circumstances, helping to reassure them that the information they’re reading is applicable to them. 

Designing for a user’s ‘whole problem’ 

Part of my role as a service designer is to understand that users don't experience services as a collection of isolated transactions. People usually transact with government services as just one part of a broader holistic goal. In the context of Help to Buy Equity Loans, users are rarely focussed on the act of repaying their loan. Instead they are focussed on life goals: moving to a new home, moving in with their partner or clearing their debt. This frames how they search for and understand information on GOV.UK. 

With this in mind, I built a holistic understanding of the user’s wider journey by mapping all of the interactions within the Help to Buy lifecycle - from navigating subletting applications and processing admin payments to retrieving forgotten account numbers. This allowed me to pinpoint the scenarios where users are likely to have straightforward queries that our AI-assisted triage tool could effectively address, and crucially, to differentiate those scenarios from those where the user really does need to speak to someone.

Users reach out to the customer service team for a spectrum of needs, from straightforward tasks to more sensitive personal circumstances. For instance, a user looking for guidance on how to make a payment is very different from a user who is worried that they may not be able to make a payment. The first is information that can readily be self-served, while the latter requires urgent assistance. Recognising this critical difference in urgency and need is crucial for an effective triage.

A more personalised user experience 

Having a ‘whole problem’ perspective also helped us shape the AI enquiry tool and the articles to our best advantage. Where Large Language Models (LLMs) really shine is its ability to interpret a user’s language and respond in a way that feels personalised to their situation.

 It does this by:

• Capturing nuances in language: The LLM is able to consider subtleties in content and the broader context of a user’s question. This allows them to return relevant articles with headings that fit users' individual circumstances or direct them to contact details if their enquiry sounds urgent

• Allowing a user to ask multiple questions simultaneously: A user may have multiple different questions at the same time. Where this would ordinarily mean searching for information on different GOV.UK pages, the AI is able to answer multiple questions in a single enquiry

• Joining information up: By connecting multiple related pieces of information in response to a single query, we are designing for the user's whole problem, not just fragmented parts of it. Users can now get a comprehensive overview of their enquiry without having to conduct multiple searches or make a phone call

The impact

The initial results of our prototype have been promising. We conducted comparative testing, in which we asked users to find specific information using both the current GOV.UK pages and our AI-assisted triage. Not only were users more able to do so using the triage, they were more confident that the information they found was what they needed, greatly reducing their need to speak to somebody. This indicates the potential to significantly reduce the volume of calls to the customer service team, freeing them up to handle more complex and nuanced enquiries.

Conclusion 

This project offers valuable insight into the evolving role of AI within public services, demonstrating how it can be seamlessly integrated into the existing GOV.UK user experience, helping users to navigate information which they might otherwise struggle to understand. By removing the risks associated with generative AI and instead leveraging it to connect users to existing information via smart design decisions, we can safely enable more efficient and user-centric experiences.

If you have any thoughts or questions about this article, feel free to get in touch! info@cautionyourblast.com