There’s a simple question boards must ask before they sign off on AI, writes Liz Henderson: how strong is the data beneath it? Operational confidence depends upon the answer
“The issue is no longer capability in the abstract. It is control in practice.”
Housing associations do not need another abstract conversation about AI. They need to know whether the tools already shaping live work can be understood, governed and defended. That is a different standard entirely.
In social housing, AI is no longer a distant innovation topic. It is beginning to appear in repairs triage, complaint handling, tenant communications, risk flags and productivity tools used by stretched teams.
At the same time, leaders are working in a tighter, faster and more public environment, where service failures, poor records and weak governance are more likely to be detected and more likely to escalate quickly.
The pressure has changed
That is why the most important question for boards and executive teams is not, “What can AI do?” It is, “How strong is the data beneath it, and how confidently could we stand behind the decisions it informs?”
In my work with boards, that is the point where the conversation becomes more useful. The issue is no longer capability in the abstract. It is controlled in practice.
For housing, this matters now because the pressures are immediate. Repairs, damp and mould, complaint performance and inspection readiness have all moved closer to the centre of strategic leadership.
Data weakness is no longer viewed as an inconvenient back-office problem. It is increasingly seen as a root cause of service failure, a blocker to assurance and, in some cases, a contributor to tenant harm. If AI is introduced on top of fragmented systems, incomplete records or weak accountability, leaders are not adding intelligence. They may simply be speeding up uncertainty.
The middle ground between hype and hesitation
Boards do not need technical mastery to govern this well. But they do need enough competence to challenge assumptions, ask for evidence and set clear expectations for executive accountability.
The same is true for executive teams. They do not need to wait for perfect conditions before making progress, but they do need a practical way to move without creating a new layer of fragility. In housing, that means resisting two unhelpful extremes: blind enthusiasm on one side and blanket caution on the other. Mature oversight sits in the middle. It is structured, proportionate and clear-eyed.
One story, two pressures
This is where the distinction between boards and execs becomes useful.
The board or Audit & Risk voice is not trying to block innovation. They are trying not to be blindsided. They want to know who owns the risk, where the audit trail sits, what the failure mode looks like and whether a decision could be explained if challenged by a regulator, Ombudsman, tenant or journalist.
Executive teams are not anti-AI either. They are trying to avoid fragility. They want to know what problem a tool fixes first, how quickly teams will feel the benefit, whether trust will be preserved and whether implementation adds protection rather than overhead.
The story is not different for each of them. The weighting is. Boards need evidence that can be minuted and defended. Execs need progress that does not create a fresh layer of operational risk.
Why data sits at the centre
Both concerns lead back to the same hinge point: data foundations. Weak data turns promising AI into fragile AI. Fragile AI produces weak assurance. And weak assurance quickly becomes a board-le and executive problem.
We can see this clearly in social housing’s own experience. The sector has already seen encouraging examples of predictive use cases in repairs and damp and mould, but it has also learned that performance falls away when the underlying data is incomplete, inconsistent or poorly governed.
That is not a reason to stop. It is a reason to be more disciplined about what leaders ask for before a use case is allowed to scale.
Too often, AI is still discussed as though the model itself is the main object of governance. In practice, what boards and executives are usually trying to assess is something more basic: can we trust the inputs, can we explain the output, and do we know who is accountable if confidence starts to weaken?
When those questions are answered clearly, AI becomes easier to govern. When they are not, even a modest use case can become a source of anxiety.
Five questions before AI becomes routine
In practical terms, leaders should focus on five things:
- First, purpose: what decision or process is this tool actually influencing?
- Second, data: what information sits beneath it, where are the weaknesses, and how visible are those limitations in plain English?
- Third, ownership: who is accountable at the executive level, who runs it day to day and which committee has sight of it?
- Fourth, human-in-the-loop (human oversight): where does meaningful review sit, especially if the use could shape a tenant outcome, a priority decision or the tone of a communication?
- Fifth, evidence: if the organisation were challenged tomorrow, what documentation could it produce within thirty minutes to show that the use is visible, controlled and understood?
Those are not technical questions. They are leadership questions. They also help resolve one of the sector’s biggest current tensions. Boards need assurance before they support wider use.
Executive teams need something workable, not theoretical, if they are to make progress under pressure. A sensible answer is not another long policy that disappears into a folder. It is a short, operationally useful AI Data Governance Checklist for every live or proposed use case. Done properly, that checklist becomes the bridge between governance and delivery.
From governance theory to operational grip
A good checklist should be simple enough to include in a board pack and practical enough to guide operational teams.
It should show the service context, the use case, the key data sources, the confidence level in that data, the executive owner, the operational lead, the human review point, the escalation trigger, the main risks and the evidence available.
If a user case touches complaints, repairs prioritisation, tenant communication, vulnerability or any route that could result in a “no” for a resident, that should be visible immediately. This is what good governance looks like when translated into action: line of sight, named accountability, meaningful oversight and clarity on the point at which confidence starts to weaken.
The housing sector does not need to slow AI down for the sake of caution, but it does need to become more disciplined about what it allows to become routine. That is particularly true where tools are embedded inside supplier platforms or used informally by staff under pressure.
One of the fastest ways to lose grip is to assume that convenience equals control. It does not. If leaders cannot see where AI is in use, what data it draws on, who owns it and how it would be explained if something goes wrong, then governance is already lagging behind reality.
The discipline that matters now
The organisations most likely to succeed over the next two years are unlikely to be those with the loudest AI strategies. They will be the ones able to demonstrate, calmly and clearly, what is already live, what remains experimental, where material risks sit, who is accountable, where human judgement still applies and how tenant interests are being protected.
Success will depend less on how confidently organisations talk about AI and more on how effectively they maintain control, oversight and accountability when it matters.
For boards, this means asking sharper questions earlier. For executive teams, it means presenting more transparent evidence, particularly around data limitations, operational risks and governance gaps. For the sector as a whole, it means recognising that data quality, explainability and accountability are not secondary considerations once procurement is complete. They are the foundation on which useful and trustworthy AI depends.
In social housing, trust in AI will not be built through ambition statements or technology messaging. It will be built through disciplined stewardship. And the first test of that stewardship remains a simple but critical one: how reliable is the data underpinning the decision?
Strong committee discipline also matters. Audit & Risk may lead on assurance, controls and reporting, but oversight cannot become siloed.
Nomination committees should consider whether board capability and experience remain sufficient, while executive teams should ensure AI is treated as an operational and governance matter with clear ownership, rather than appearing in papers as a broad digital update disconnected from accountability.
Liz Henderson is data and AI technical collaborator at Demystifying AI for Social Housing (DASH)
Image credit: Shutterstock Gen AI
Read next: Smart monitoring: Retrofit’s missing link
Are you a social housing professional? Sign up for a FREE MEMBERSHIP to upload news stories, post job vacancies, and connect with colleagues on our secure social feed.
