The use of AI in housing has reached a tipping point during the last 12 months. From a niche tool used primarily by the sector for machine learning and predictive analytics, it’s now become a part of everyday life for thousands of housing professionals. But how can we ensure the technology is rolled out in a way that’s safe, accurate and ethical at a time when regulation of AI in the UK is virtually non-existent? Neil Merrick investigates.
In some ways, housing associations stumbled across AI by accident. For the past few years, a small number have used predictive analytics to flag up tenants on the brink of serious rent arrears and prioritise where they offer support.
The results are impressive, with landlords that successfully help tenants overcome financial problems reporting fewer tenancy failures and less money lost through properties sitting vacant.
But when it comes to AI, predictive analytics is only scratching the surface and, in some people’s eyes, no more than advanced machine learning.
Those more serious about artificial intelligence are inviting bots or AI tools to check policy documents, compose letters and even analyse complaints. This can include assessing how angry tenants are when they contact a customer services centre.
For the past two years, Thirteen Group has used an AI bot developed by its own employees to verify rent rises for tenants on the national universal credit landlord portal. Its use of Robotic Process Automation (RPA), a Microsoft tool, was backed by the Department for Work and Pensions, which invited other landlords to follow suit.
The bot saves thousands of working hours while allowing staff to focus on supporting customers, says Hassan Bahrani, Thirteen’s director of IT, cyber and data security.
More recently, the association began using Voicescape, an AI-empowered caseload manager, to contact tenants about rent issues. Since 2023, Thirteen’s year-end debt has fallen from 3.24% to 2.55%. “We’re shifting away from tactical adoption of AI to becoming more strategic,” says Bahrani.
‘Guardrails’
Nearly one third of Magenta Living’s office-based staff are trained to use Copilot, Microsoft’s best known AI tool or large language model (LLM). This includes 55 employees who studied at the Wirral AI Academy, set up by Wirral Chamber of Commerce to upskill local businesses and help ensure AI is used ethically.
The important thing, says Ian Cresswell, Magenta’s director of technology, is that housing providers understand and grasp the potential of AI while recognising its limitations and risks. “AI is here and all around us,” says Cresswell. “We’re embracing it while acknowledging the need for guardrails.”
“AI has it uses but, whatever the tech bros may say, there are things that I don’t see it doing in the future”
Mark Shephard, head of data, Yorkshire Housing
Nowadays, most excitement surrounds the use of generative AI, which can check policy documents comply with government legislation, and ‘agentic AI’, which is able to draft letters and produce other materials.
Among those embracing AI is Hayley Ward, Magenta Living’s marketing and brand director. The association’s Words Open Doors Agent (WODA) helps staff, she says, to create messages that reflect Magenta’s tone of voice and values, saving time and reducing complexity in the process.
“AI doesn’t replace people. It supports them,” says Ward. “WODA helps us to write letters that feel clear, warm and human, but the real magic comes from colleagues. It means we see the person, not the property.”
Across the sector, most social landlords have adopted a more cautious approach, dipping their toes in the water and seeing what others are doing as the clamour to become more AI-savvy becomes difficult to ignore.
According to Guy Marshall, an AI consultant and board member at One Manchester Housing, AI has only appeared on most landlords’ radar during the past year.
The idea of automating decisions is superficially appealing, says Marshall, especially if it cuts costs. But if you’re going to throw data at an LLM, you must be sure all the data is accurate and fit for purpose.
The danger is seeing AI as a ‘magic bullet’ without preparing properly, making the sector vulnerable to pressure from suppliers and vendors of AI. “A lot of providers would love a quick win,” says Marshall, director at data and technology consultants Fuza. “If you automate decisions based on rubbish data, you will simply get bad outcomes faster.”
‘Wild west’
Another problem is that, just as people are getting to grips with AI, the technology changes and throws up new challenges. “It’s like the wild west,” says Ian Wright, founder and chief executive of the Disruptive Innovators Network.
Wright doubts whether, at present, most of the sector knows how to make the most of AI without handing power over to chatbots and other automated tools. “People fundamentally misunderstand what AI is there to do,” he says. “Leaders don’t know the questions to ask,” he adds.
There’s also the problem of AI fakes, exemplified by a barrister who last year relied on fictitious case histories generated by AI to make his clients’ case during a court hearing. “Most of your role is going to be that of an auditor,” says Wright. “You’re going to have to check AI because it lies.”
Regulation of AI in the UK is almost non-existent, increasing pressure on housing association boards and local authority landlords while at the same time giving them more freedom to explore and test the potential of AI.
The regulator view
In November, as part of its latest risk profile of the sector, the Regulator of Social Housing said: “It’s important that landlords manage their data in accordance with all relevant laws and regulations and understand the implications for data protection of adoption of new technologies such as AI.”
This isn’t the case in wider Europe. From next month [Feb], housing providers in the EU must comply with the AI Act. Staff will need at least a basic level of AI literacy if they use AI for allocations, risk assessment (including rent arrears and antisocial behaviour), fraud detection or customer contact (such as chatbots).
“Employees who work with AI must understand what the system does, what the risks are, and how human oversight is exercised,” says Henk Korevaar, ambassador at CorpoNet, a Netherlands-based network for housing associations.
“Having a minimum level of AI competency doesn’t mean that everyone needs to become a data scientist or programmer,” he adds. “It means that organisations must ensure that employees who work with AI understand what they are doing and where their responsibilities lie.”
“People fundamentally misunderstand what AI is there to do. Leaders don’t know the questions to ask”
Ian Wright, chief executive, Disruptive Innovators Networks
While the UK largely relies on self-regulation, social landlords have it within their powers to ensure things don’t go badly awry. AI may be crying out for data, but it can also help organisations to improve its quality.
At Magenta Living, Ian Cresswell points to their data platform Microsoft Fabric as an effective way to rationalise data. The association also uses Microsoft Purview, a governance and compliance platform, to monitor exactly how far AI reaches into the organisation and ensure AI is kept in check. “There’s always a human in the loop,” he says. “AI should support decisions, not make them alone.”
Working together
There are also signs of social landlords working in tandem. Thirteen was keen not to commercialise its use of the RPA tool for verifying UC claims and recommended it to Platform Housing Group. “It’s important that we’re sharing as a sector,” says Hassan Bahrani. “Ultimately, the customer is going to win.”
Bahrani stresses that staff continuously check the accuracy and validity of decisions made by the bot, while an AI ethics committee is under consideration. “If you use AI with colleagues or customers then you should be transparent about how AI is part of that process,” he says.
In addition to using RPA for universal credit claims, Platform introduced the ‘silent tenant’, a tool for checking on the welfare of elderly and more vulnerable tenants who haven’t been in contact with the association for long periods.
Each tenant receives an automated call and, if they fail to respond via their phone keypad, an officer visits their home. “The machine is thinking for you to provide a better outcome,” says Jon Cocker, Platform’s chief information officer.
“If you automate decisions based on rubbish data, you will simply get bad outcomes faster”
Guy Marshall, director, Fuza
Since early December, residents calling Bromford Housing Association’s customer service centre in Wolverhampton have initially been welcomed by an AI-supported interactive voice response (IVR) system. The call is then put through to a member of the customer services team, who’s aware of the caller and why they contacted the association.
Colin Goodbody, head of customer operations at Bromford, says the association isn’t only providing a better service but can collect data revealing typical problems, as well as ‘customer sentiment’. This is used to support training and improvements to services.
Enabling colleagues
Later this year the service will be expanded so, rather than callers being put on hold, AI provides staff with articles and other information that can be quickly relayed to callers. “It’s not about creating agents that are robots and taking away personality,” says Goodbody. “It’s about enabling our colleagues to help customers more efficiently.”
Yorkshire Housing is in the throes of testing how AI can analyse notes made by call centre staff during discussions with tenants who make complaints, including the sentiment expressed by complainants.
Mark Shephard, the association’s head of data, performance and information security, stresses that it’s early days, but believes AI has the potential to help staff understand the root causes of complaints faster and prioritise improvements.
AI, explains Shephard, can examine large volumes of text in ways people find tricky, if only due to the time involved. But there’s also a level of bias in judgements made by an AI tool or app that staff and organisations must take into account.
Ultimately, AI is here to stay. As LLMs build up case histories, they should become more knowledgeable and hopefully proficient. AI also has the potential to take the drudgery out of day-to-day tasks, including checking guidance or reports.
All this doesn’t mean social landlords need fear AI or should allow it to call the tune. “AI has its uses but, whatever the tech bros may say, there are things that I don’t see it doing in the future,” adds Shephard. “While it may help us identify efficiencies in our business processes, I don’t think AI is going to be removing people’s day jobs.”
Case study: Predicting tenancy failure
For the past six years, a predicting tenancy failure model created by Together Housing has been 84% successful in forecasting when tenants are in danger of losing their home due to rent arrears.
Since 2023/24, the housing group calculates that it’s saved more than £3m by correctly identifying tenancies that are at risk, making targeted interventions, and providing residents with wraparound support rather than needing to relet homes.
Together owns more than 38,000 properties, mostly in northern England, and employs 1,600 staff. Eight years ago, Stephen Batley, its assistant director of business development, was appointed to improve the way the group compiles and uses data, and subsequently ensure AI did not “run amok”.
Eight years on, Batley says the success of the tenancy failure model shows how a combination of effective data science and personalised skills shown by Together’s staff can help tenants to remain in their homes.
In addition to achieving 84% accuracy in forecasting tenancy failure, the model is 81% successful in identifying preventable terminations and classifying reasons for a tenancy being terminated.
Data scientists experimented with various models and techniques to come up with the model, while consulting frontline staff who understand residents as individuals, adds Batley. “This project is the perfect example of marrying the technical with the human to make a positive difference to people’s lives,” he says.
The savings made since 2023/24 coincide with turnover of properties falling from 15% to 6%. Among the factors taken into account are the age of tenants and the type of property they live in, with staff focusing on the most vulnerable households.
According to Batley, it’s important not to become too dependent on AI and recognise how predictive models are complemented by human endeavour. “We’re trying to take a balanced approach and use AI in a pragmatic way while ensuring there are checks and guard rails,” he adds.
AI dos and don’ts
DO
- Ensure staff using AI are fully trained and understand what they’re trying to achieve
- Require staff to check decisions made by AI, including negative judgements that affect tenants
- Check data used to generate decisions via AI is accurate and fit for purpose.
DON’T
- Be afraid to explore the full potential of AI for housing
- Use AI for communication or making major decisions without informing tenants and other customers
- Allow AI to run wild and take over operations or decision-making.
