The Boiler Upgrade Scheme (BUS) offers a government grant of up to £7,500 towards the cost of a heat pump for those wishing to replace an existing fossil fuel heating system, such as a gas boiler. While this incentive aims to support the transition to low-carbon heating, uptake has been slower than expected. Although uptake is on the rise now, it is still not at the level needed for the government to reach its targets for net zero. As part of Nesta's sustainable future mission and our continued exploration of challenges and barriers to uptake of BUS, we set out to understand public awareness of the scheme and the barriers preventing greater adoption.

To investigate this, we conducted a series of AI-facilitated interviews exploring perceptions of the BUS. These interviews aimed to uncover levels of awareness, concerns and misconceptions about eligibility and application processes. Our goal was to generate insights that could inform policymakers in shaping more effective outreach and support strategies.

A key innovation in this project was the use of artificial intelligence (AI) tools to support qualitative analysis. We tested whether natural language processing (NLP) methods and large language models (LLMs) could help speed up the early stages of thematic analysis, making it easier to identify key themes in the interview data.

We have been working on this since October 2024.

We used a combination of NLP techniques and qualitative thematic and sentiment analysis to identify broad themes in the interviews. By combining computational methods with human oversight, we were able to extract meaningful findings without the extensive manual effort typically required in qualitative research. To support further exploration, we built an interactive dashboard where researchers can browse interview responses by theme and explore the interviews in greater depth.

Our approach identified key themes such as cost considerations and the environmental benefits of heat pumps (see table below). It also distinguished some nuances in discussions of the eligibility criteria, for example identifying one cluster, “eligibility and fairness”, which contains responses related to whether the eligibility criteria are too stringent, and another, “eligibility and awareness”, comprising responses related to users’ knowledge about the eligibility criteria and how to apply.

We note that these themes are named and described primarily through automated analysis using LLMs – highlighting the potential of generative AI for scalable, automated insight generation with a human-in-the-loop.

Below, you can explore the interview data interactively: hover over a point to see the text of the response. This landscape of points is generated using an automated topic modelling approach (a more traditional machine learning method compared to generative AI). Each point is a response by an interviewee, with similar responses being situated closer in the visualisation.

While the quotes are shown here in isolation to protect participants’ anonymity, each one remains linked to the full anonymised transcript in our underlying data that we hold at Nesta. The colour indicates the topic that was assigned automatically by the topic modelling approach.

Overall, our analysis of these interviews at the time confirmed our initial perception that uptake of BUS may be hampered by a lack of awareness of the scheme, so we now welcome the recent government campaign to increase awareness. Beyond this, many respondents expressed confusion about eligibility criteria, indicating that members of the public are unaware of the requirements and expect them to be complicated or restrictive, potentially discouraging applications.

The clusters of responses around home improvements such as insulation further indicates misconceptions about the scheme, and an area where improved messaging could aid uptake. While insulation is an advantage, it is no longer a requirement of the scheme, with the restriction of no outstanding recommendation for loft and cavity wall insulation being removed in May 2024. 

Notably, a sizable portion of users were positive about heat pumps and other home improvements that could reduce their carbon footprint. Taken together with the other themes that emerged, this suggests that members of the public are keen to reduce the environmental impact of their home heating, but misconceptions around eligibility for support, coupled with underlying low levels of awareness are reducing the uptake of government support.

As Nesta continues to work on making heat pumps more appealing and affordable to consumers, we will also keep exploring opportunities for using AI to support our qualitative research work. For example, while here we used LLMs for generating summary descriptions of the topics, they could also be used to answer more specific research questions about each topic.

Acknowledgements

We thank Camille Stengel for her feedback on the automated topic modelling approach.