Could AI help assess housing need and deliver more social homes?

What is XAI modelling?

Dr Guerrero’s role involves researching different models to analyse public policy, and he has been developing XAI models to evaluate in greater detail whether housing policies work in the real world and understand how people react to new policies.

He uses XAI models because they are more transparent in explaining how and why a model has arrived at its decision, giving insight into how the inputted data relates to the outcome.

This modelling differs to other types such as predictive AI, which is known as a ‘black box’ because it does not explain how the inputted data relates to the predicted outcomes.

For example, US-based online real estate marketplace Zillow used predictive AI unsuccessfully as it overestimated the growth of house prices they purchased and sold many of them at a loss because their algorithm could not provide an insight into explaining the growth of a house price using factors determining its value. 

Mr Guerrero’s XAI modelling can evaluate the behaviour of a household and how this might change depending on how much income they receive, what their financial assets are, any budget constraints they have, or how much inheritance tax and stamp duty they pay. “If you implement the policy, people […] will shift their behaviour as a response to the policy,” the economist explains.

He calls it “bottom-up” modelling where you input these specific details about every household (the micro level), then analyse the outcomes on a larger scale that explain the distribution of wealth, geography and demographics of an area (the macro level) where the housing policy has been tested.

For example, the AI modelling could analyse the impact of changing who pays council tax – either those living in the home or the landlord – in a certain geographical area. “Those are the kind of things that you cannot experiment in the real world, [as] you might generate undesired outcomes harming other populations,” Dr Guerrero adds. “So that’s why these tools are very important to scrutinise [policies] with a high level of detail.

“Bottom-up AI performs experiments on these policies and gives policymakers the capability to complement their decision-making process with evidence on the potential outcomes that different policy designs could entail.”

John Guest, national head of social housing at RSM UK, commented on the impact XAI could have on the housing sector: “There is no doubt that AI could be an excellent tool that registered providers and policymakers can use to complement their existing strategies to determine where housing need is greatest. 

“While we know that the demand for affordable housing remains high, [and] with finite levels of available land and funding, we can see how using these tools could have a positive impact on the sector.”

Will building more homes reduce housing wealth inequality?

The holistic approach that an XAI model offers could be more effective in reducing housing wealth inequality, according to the economist.

Inequality between higher-income and lower-income households has exacerbated over time, and the latest figures show that lower-income households spend an average of 21% of their income on housing costs compared to just 6% for higher-income households.

In a research paper from 2020, Dr Guerrero found that simply increasing the housing stock in the UK would not ease housing wealth inequality but see higher-income individuals buy more housing and increase the rent.

“[It] doesn’t mean the people who need them the most will grab them, it is still the higher-income households who snatch them,” he adds.

The assumption is that increasing the housing stock to ease demand pressures and lower prices will not “generate richer outcomes such as reducing homelessness [or] improving living standards”, as the problems require more detailed modelling.

“It’s not a question of how much you push the demand and the price dropping, it’s a question of accessibility and who is able to buy first. 

“You cannot get that level of detail of who is able to buy first because those details are about interactions [and] behaviour – exactly the type of things that I tried to put in my models.”

Dr Guerrero believes XAI can reduce housing wealth inequality by targeting a demographic of people in a specific location who need access to social housing: “It allows you to design a policy [where you say]: ‘Let’s place housing here and see if we can attract the people we expect,’ [so you] attend to the population that needs it the most.”

Ms Boorman believes this modelling can help meet the demands of the current market: “Our current planning system and housing targets set many years ago do not take stock of what the UK population demands are currently, never mind predict what future needs are. 

“Data and AI could map out the future housing needs of the UK’s population, co-ordinated with a new planning reform and regime to focus on key areas and working in partnership with private companies.”

Using XAI to test a private sector rent cap

The government raised its social housing rent caps up to 7.7% in England in April for the 2024-25 year, but the private rented sector does not have caps in place for tenants.

Dr Guerrero believes introducing private rent caps could become an option for policymakers because XAI models can analyse potential policies far more comprehensively. 

“Politicians say we cannot implement those measures because they would disincentivise investment in the housing market, [and] from the side of the landlords, it would imply less income,” he says. “But that’s just speech. We would want to test that with data and science to see to what degree that makes sense.”

Modelling this policy could see what the impact would be on landlords’ behaviour when trying to reduce tenants’ cost of living. “It could harm homeowners by depreciating their house values,” says Dr Guerrero.

“It could also incentivise landlords to invest less in their properties if no appropriate regulation is enforced. Which of these outcomes would dominate such a policy? The answer is not straightforward.”

The use of an XAI model would allow a rent cap policy to be virtually tested to a specific area where they may want to introduce it and see what the outcome could be: “This is why this type of AI can be extremely helpful,” emphasises Dr Guerrero.

Addressing homelessness

Last month, Dr Guerrero published a paper that explored how public infrastructure influences housing affordability, focusing specifically on the impact of Transport for London’s Elizabeth Line rail service.

The research saw him compare the outputs of his XAI model to black box AI models. While Mr Guerrero found it surprising to see the black box AI models overestimate price changes in the housing market compared to the XAI model - similar to what Zillow’s predictive AI model had done – the results demonstrated how house affordability would decrease away from the central London area and exacerbate housing wealth inequality.

But Dr Guerrero believes the AI modelling in this study is a great step in understanding how to improve social housing near key infrastructure, and can be adapted to reduce homelessness, too, as housing charity Shelter shared figures estimating that by the end of 2023, homelessness in England has increased by 14% to 309,000 since the previous year.

“We are currently extending this model [for] the rental property market to get homelessness rates right,” says Dr Guerrero. “If we achieve this, we would be able to study a set of policy interventions, [with] social housing being one of them. 

“We are very close to getting there so I am confident we will soon have a new generation of housing policy models ready to be adopted by policymakers.”

Current AI use in the housing sector

Dr Guerrero views the interest from local authorities like WMCA as “very positive”, but with numerous pathways XAI modelling can go into within the housing sector, and he is certain it will revolutionise the process of policymaking.

“It’s a no-brainer,” he says. “It’s not about having a tool that replaces the decisions of the experts or replaces their analysis, [it] is more of enriching the toolkit that decision-makers have.”

Other AI models have been piloted by the Department for Levelling Up, Housing and Communities (DLUHC) to start innovating the housing sector in other ways.

Partnering local planning authorities with the property technology (proptech) sector has been successful, as trials at Chesterfield Borough Council in speeding up the process of responses to planning consultations has delivered “280 hours” of time savings for planning officers using generative AI and a digital consultation portal to help develop a local plan.

The department is also exploring alternative AI models to the generative one used, as well as considering the results from a digital planning programme that uses another AI-based solution to allow staff to “pull-out information and policies from local plan documents [to] support the querying of local plans data”.

For DLUHC, the development of these business operations through its PropTech Innovation Fund highlights how to “boost productivity and efficiency across the housing and planning industry”.