Peter Lunio explains why the safest first step is not a dramatic transformation, but practical, bounded uses that reduce pressure, support staff, and strengthen tenant trust

1.AI does not have to start with transformation

The first useful AI in housing may not arrive as a transformation programme. It may arrive as a clearer repairs update, a better complaint summary, or 10 minutes saved before a difficult tenant call.

That matters because the most useful opportunities are often quieter than the headlines suggest. They sit in the ordinary work that absorbs staff time every day: summarising complaints, drafting clearer repairs updates, finding the right policy, preparing case notes, or spotting where a tenant may need earlier support.

Housing teams are already stretched. Repairs, damp and mould, complaints, arrears, regulation and financial pressure are all competing for attention. If AI is going to help, it needs to reduce friction in that real operating environment, not create another abstract project.

The question is not whether AI can sound impressive. It is whether it can help people do important work more safely, more consistently and with better information. For boards and executives, that means starting with use cases that are practical, bounded, visible and safe to test.

2.Why quiet use cases matter

The temptation with AI is to start with the biggest promise: transformation, automation, prediction and scale. In housing, that can be the wrong place to begin. The safer and more useful starting point is often a quiet use case that helps a team with work they already understand.

These use cases matter because they are easier to explain, easier to test and easier to govern. A complaints team can judge whether an AI summary has captured the right themes. A repairs manager can see whether a drafted tenant update is clearer. A housing officer can tell whether a case note summary saves time without losing important context.

That makes quiet use cases valuable for confidence. They give staff permission to learn by doing, while giving leaders evidence about what works, what needs control and what should not be scaled. They also help avoid the false choice between doing nothing and launching a major AI programme.

For stretched organisations, the best first use cases are not the most dramatic. They are the ones where benefit, risk and accountability can be seen clearly.

3.The evidence points to a practical middle ground

Recent sector evidence points in the same direction. The NHF/Phoenix survey found that 47% of responding housing associations are already using AI in day-to-day operations, while a further 23% are not using it yet but plan to. But readiness is uneven: 87% reported low knowledge of AI, and 44% had no AI policy in place.

SFHA’s 2026 readiness work showed an average readiness score of around 58%, with appetite and practical use running ahead of strategy, governance and staff confidence. Quiet use cases create a bridge between informal experimentation and full-scale adoption. They allow leaders to learn where AI genuinely helps, where the data is not good enough, and where the risk is too close to tenant outcomes to proceed casually.

4.Five quiet use cases worth testing first

The first quiet use case is complaint theme summarisation. Complaints contain some of the most valuable learning in a housing association, but they are time-consuming to review across teams, stages and service areas. AI can help identify repeated themes, tone issues, emerging risks and patterns that might otherwise stay buried in individual cases. The judgement, however, must remain human. AI can support the learning loop, but it should not decide whether a complaint is fair, whether redress is appropriate, or how an organisation should respond.

The second is repairs communication drafting. Many repair complaints are not only about the repair itself, but about poor updates, unclear explanations or a lack of follow-through. AI can help staff draft clearer appointment messages, follow-up notes, delay explanations and plain-English summaries. Used well, it can improve consistency and tone. Used badly, it can sound impersonal or miss the human context. That is why every message still needs staff review before it reaches a tenant.

The third is policy and knowledge search. Housing staff often work across complex policies, procedures, service standards and regulatory requirements. Time is lost trying to find the right answer quickly, especially in busy contact centres or operational teams. AI-assisted search can help staff locate approved guidance faster and produce a clearer starting point. The control is important: the tool should search trusted, current documents, not invent answers from the open internet or outdated material.

The fourth is case note summarisation. Housing officers, income teams, complaint handlers and managers all spend time reading long histories before making decisions or preparing for conversations. AI can help summarise previous contact, highlight open actions and structure information before a meeting or case review. This can reduce admin pressure and help staff prepare better. But sensitive tenant information should not be copied into uncontrolled tools, and summaries should always be checked against the source record.

The fifth is early arrears or tenancy-support triage. This is more sensitive, but potentially valuable if handled carefully. AI can help teams identify patterns that suggest a tenant may need earlier support, such as repeated missed payments, contact difficulties or vulnerability indicators. The point should not be automated enforcement. It should be earlier, better-targeted human support. Because this can influence tenant outcomes, it needs stronger governance: clear ownership, tested data, human review, fairness checks and a route for challenge.

These are not glamorous use cases. That is their strength. They are practical enough to test, close enough to real service pressure to matter, and bounded enough to govern properly.

5.Proof in practice: what the case studies show

The strongest examples of AI in housing are not abstract. They show AI being used to support earlier action, better communication and more focused staff time.

Wolverhampton Homes’ damp and mould work is a useful reminder that AI is not magic dust sprinkled over poor records. Its value depends on the quality of the underlying data and the speed with which people act on the insight. For repairs and asset teams, prediction only matters if it leads to earlier, better human intervention.

The same principle applies in income and tenancy support. Examples such as Together Housing show how AI can help teams identify where earlier support may be needed. In a cost-of-living environment, arrears are often connected to wider vulnerability. The best use of AI here is not automated enforcement. It is better targeting of human support, with safeguards around fairness, data use and review.

Tenant-facing tools, such as Livv’s AI-powered tenant assistant, show another route. They can help residents find information, navigate services and get quicker answers. But this is also where trust becomes most visible. If the tool cannot answer properly, or if the tenant needs help, there must be a clear route back to a person.

Taken together, these examples point to the same conclusion. AI works best when it is connected to a real service pressure, bounded by human oversight, and measured against outcomes that matter to tenants and staff.

6.The safe-to-test filter

A quiet use case is not safe simply because it sounds small. It is safe because it is visible, bounded, owned and reviewed.

Before testing any AI use case, leaders should ask seven practical questions. What data will the tool use? Could the output influence a tenant outcome? Is the source information accurate enough? Who will review the result before action is taken? Who owns the use case? What evidence will show whether it helped? What would make us pause, change or stop?

Those questions do not need to create a heavy process. They create discipline. A repairs update drafted by AI may be low risk if reviewed by a member of staff. A triage model that shapes priority, vulnerability or arrears action is different. It needs stronger assurance.

If a use case touches complaints, arrears, vulnerability, repairs prioritisation or tenant communication, it should leave an evidence trail that a board, executive team or regulator could understand later. The aim is not to slow useful experimentation. It is to make it safe enough to continue.

7.What this means for boards and executives

For boards, the issue is not whether every quiet experiment needs approval. It is whether the organisation has a clear way to decide what is low risk, what needs review and what should not proceed without stronger assurance. Board members need line of sight: which use cases are being tested, what they touch, who owns them and what evidence is being gathered.

For executives, the challenge is different. They need to create permission for staff to use AI confidently, without allowing informal use to become invisible or inconsistent. The practical question is simple: where could AI help teams this quarter, and what guardrails would make that safe in practice now?

8.Avoiding the wrong first use cases

Not every idea should become an AI use case. Avoid tools that make or strongly influence tenant decisions, rely on weak data, replace human judgement in complaints or vulnerability, or cannot be explained to staff, tenants and boards. Ambition matters, but the first tests should be clear, bounded and accountable.

9. Start small, but not casually

Quiet AI use cases can help housing teams now, but only if they are tested with discipline. Start small, keep the work visible, leave humans in the loop and measure what changes.

The practical next step is to choose two low-risk use cases, name an owner, run them for 30 to 60 days, and measure whether they save time, improve clarity or strengthen service grip.

The opportunity is not to use AI everywhere. It is to use it carefully, where it helps staff, improves services and strengthens trust.

The quiet AI use cases that could help housing teams now