The data layer: are your AI tools looking at the right information?
AI will only ever be as good as the information it can reach.
In the early 2000s, I spent a decade developing bespoke asset management systems to support capital delivery and reviewing and rewriting the operational processes around them. What stayed with me wasn’t the technology. It was watching people work.
To complete a routine task, users would flick from screen to screen and from source to source, gathering the information they needed piece by piece. The task was usually captured in a Standard Operating Procedure (SOP), often written by a competent colleague. How well the process performed depended as much on the people following it as on anything else.
My job as a developer was to work out which information a process needed, then build those data points into the application so that it, rather than the user, did the heavy lifting.
Today there is a great deal of talk about how AI will shape social housing delivery models. For those of us already using it, the potential is real. But one risk links the past to the present: the data layer.
What is the data layer?
The data layer is the collection of information an organisation holds that drives its day-to-day operations. It takes many forms and is widely dispersed: policies and procedures, emails, documents, spreadsheets, databases, customer contact records, project plans, knowledge bases, websites, budgets, organisational charts and more. A person’s role and responsibilities usually determine what they can access.
That dispersal is the problem. It is the same fragmentation I saw twenty years ago, and AI doesn’t make it disappear. It may simply hide it.
There is a useful parallel in data science; an article I read said, “Your CEO reports you have 50 petabytes of data. Your data scientists say 80% of it is unusable for AI”. This finding mirrored my own experience. To frame it differently – if the data is messy, will AI need AI to make sense of it?
To realise the benefits of AI, the unglamorous work of improving the data layer comes first, and it determines the success of everything else.
Why the data layer matters for AI
AI uses the knowledge it can access. That single idea leads to three practical points.
1. It provides organisational context
Housing providers share a broadly identical core purpose, but operational differences matter. Policies, procedures, structures and culture vary from one organisation to the next, and many have their own “insert name here” way of doing things: how repairs are triaged, how a vulnerable tenant is flagged, how an exception is approved. Embedding that way of working in systems helps it gain traction. An AI tool that can’t see it will fall back on generic answers.
2. Completeness and accuracy determine reliability
How much you can trust an output depends on the quality of the input data. A recurring theme in regulatory judgements is a failure to understand asset performance. Poor stock condition data weakens compliance with the Decent Homes Standard and undermines planned investment programmes. With Awaab’s Law now placing firm expectations on landlords to deal with damp, mould and other hazards, the cost of not knowing your homes has risen.
Imagine asking an AI tool which homes are most at risk of damp and mould. If condition data is several years old, held in three different places, or recorded inconsistently, the answer will sound confident and may be wrong. That is arguably worse than no answer at all.
3. Mapping organisational knowledge is the critical first step
Before choosing an AI solution, an organisation needs to know what information exists, where it is stored, in what format, who owns it and who uses it. Without that map, AI relies on general concepts rather than drawing on organisational experience.
What this looks like in practice
The risks aren’t confined to tidy, structured databases. Some of the most revealing examples involve the messier material.
- Tenant-facing chatbots. A chatbot answering from an outdated policy page or an old SOP will give wrong advice with total confidence. Which version of the policy is the right one? If you can’t answer that, neither can the AI. Above all, people prefer to communicate with humans, rather than feel they are simply being processed. So tread carefully when considering using AI here.
- Repairs and responsive maintenance. An AI triage assistant is only as good as the job histories, property attributes and contractor data behind it.
- Duplicate and conflicting records. When two systems hold different tenancy statuses, or different addresses for the same home, AI will inherit the conflict and may quietly pick a side.
- Access and sensitivity. Because access is usually role-based, AI that can see everything a user can see may surface material that person technically had access to but would never have found, such as safeguarding notes or vulnerability flags. Permissions, GDPR and sensitive tenant data need to be considered before any rollout.
“Surely AI can cope with messy data?”
It’s a fair challenge. Modern AI can read unstructured documents, summarise long email threads and spot patterns that people would miss, and that is a genuine opportunity. But it can only work with what it is given. It might not tell you which of three conflicting documents is current, or that a stock condition record is five years out of date, unless the organisation has made that clear. AI can cope with messy data, but it can’t make it trustworthy.
Where to start
You don’t need a perfect data estate before you begin. A practical approach looks like this:
- Map what data exists, where it lives, who owns it and what format it is in.
- Prioritise the data that matters most to your first AI use case.
- Assess its quality: is it complete, accurate and up to date?
- Assign ownership, so someone is accountable for keeping it right.
- Start small, with one contained use case, and learn from it before scaling.
The foundation comes first
Twenty years ago, my task was to bring the right information to the person doing the work. The task is similar now, except the “person” may be an AI. The technology will keep changing, but the foundation won’t. Organisations that understand and look after their data layer will get far more from AI than those that don’t.
AI can only use the knowledge it can access. The question for every provider is: what have we given it?

About the Author
Adam Borrie is an experienced asset management consultant with a business systems and change management background, who has developed a holistic approach to business improvement.
News & Insights
Read the latest housing sector news, blogs, and commentary from ARK.
Beyond Decent Homes Compliance: improving repairs performance and resident experience
By Luke Beard ·Decent Homes tells you what a property looked like on the day someone inspected it. It says nothing about how …
The data layer: are your AI tools looking at the right information?
By Adam Borrie ·AI will only ever be as good as the information it can reach. In the early 2000s, I spent a …
Risk resolution in housing repairs: why job completion KPIs can be misleading
By Luke Beard ·Decent Homes tells you what a property looked like on the day someone inspected it. It says nothing about how …
Subscribe to our newsletters for the latest industry insights
Our newsletters and reports will keep you updated on topical issues from the sector as well as what’s happening at ARK.