AI Is Only As Good As The Data You Feed It
Or, why even the smartest AI can become social housing’s most confident guesser
Over the last year, it feels like every conversation about technology has eventually landed in the same place.
Artificial Intelligence.
Whether you’re in housing management, asset management, repairs, finance, customer services, or sitting on an executive team, you’ve probably been told that AI is going to transform the way we work.
And to be fair, it probably will.
But before we talk about AI itself, I think we need to talk about something far less exciting.
Data.
Because despite all the headlines, powerful demonstrations, and promises from software providers, the truth is surprisingly simple:
AI is only ever as good as the data it learns from.
And that’s particularly relevant in social housing.
The Smartest Person In The Room… With Half The Information
Imagine you’ve just hired a new employee.
They’re incredibly intelligent.
They can write reports in seconds.
They can summarise vast amounts of information.
They can explain complicated subjects in plain English.
They never get tired, and they’re available 24 hours a day.
There is just one problem.
They’ve only been given access to half the information they need.
At that point, something interesting starts to happen.
Instead of saying “I don’t know”, they begin making educated guesses.
Some of those guesses are surprisingly accurate.
Some are completely wrong.
And the dangerous part is that they deliver both answers with exactly the same level of confidence.
That’s essentially what happens when AI doesn’t have access to good-quality data.
In the AI world, these mistakes are commonly referred to as “hallucinations”. It’s a dramatic term, but what it typically means is far less mysterious:
The system is trying to fill gaps in its knowledge with prediction rather than fact.
Why Data Matters More Than The AI
A lot of organisations are currently focused on choosing an AI tool.
ChatGPT.
Gemini.
Copilot.
Claude.
The reality is that these tools are becoming increasingly capable and increasingly similar.
The bigger question is:
What data are they working from?
Because an AI model trained on poor-quality, inconsistent, incomplete, or unvalidated information will produce poor-quality outputs no matter how sophisticated the technology behind it is.
It’s a bit like repairing a boiler with the wrong service history or planning an asset investment programme using incomplete stock condition data.
The sophistication of the process doesn’t matter if the information going in is flawed.
The old saying applies:
Garbage in, garbage out.
Unfortunately, AI can make that garbage sound incredibly convincing.
Labelled Data vs Unlabelled Data: A Housing Association Analogy
Let’s take a practical example.
Imagine someone gives you 10,000 repairs records.
One version contains thousands of lines of text.
Another version has been carefully categorised and validated.
Responsive repair.
Void repair.
Emergency callout.
Electrical.
Heating.
Communal.
Property type.
Completion time.
Cost.
Contractor.
Suddenly the information means something.
This is the difference between what data scientists often refer to as unlabelled and labelled data.
Unlabelled data is simply information.
Labelled data is information with context.
And context is what allows AI to understand what it is looking at.
Without context, AI has to make assumptions.
With context, AI can identify patterns.
The fewer assumptions required, the lower the risk of hallucinations and incorrect conclusions.
Social Housing Has A Data Advantage
This is where social housing is actually better positioned than many industries realise.
Housing providers have spent decades collecting operational information.
Repairs.
Assets.
Resident satisfaction.
Tenancies.
Arrears.
Complaints.
Services.
Compliance.
The challenge has never been a lack of data.
The challenge has been making sense of it.
The average housing provider has information spread across housing management systems, CRM platforms, finance systems, asset management solutions, spreadsheets, contractor portals, surveys, emails and documents.
The data exists.
The question is whether it is structured, consistent, complete and trusted.
Because those qualities matter far more than the AI model sitting on top.
Why Housemark Has Been Preparing For AI For 25 Years
What’s interesting is that Housemark’s role in this space started long before anybody was talking about Large Language Models.
For more than 25 years, Housemark has been helping housing providers benchmark performance through validated and comparable data.
That process has always involved asking important questions:
Is the data complete?
Is it consistent?
Is it comparable?
Does it mean the same thing across different organisations?
Can it be trusted?
Those questions were important for benchmarking.
They’re even more important for AI.
Because while AI technology has arrived incredibly quickly, trust isn’t built quickly at all.
Trust comes from confidence in the underlying information.
In many ways, Housemark’s greatest AI asset isn’t AI.
It’s decades of validated housing data and sector intelligence.
The Future Belongs To Organisations That Trust Their Data
When people talk about AI, they often imagine a future where technology replaces work.
I think the reality is much more practical.
The organisations that will benefit most from AI won’t necessarily be the ones with the biggest technology budgets.
They’ll be the ones with the best understanding of their data.
The ones that know where it comes from.
The ones that understand its quality.
The ones that have invested in governance, validation and consistency.
Because ultimately, AI isn’t replacing expertise.
It’s amplifying whatever information it has available.
If that information is reliable, AI becomes a powerful assistant.
If it isn’t, AI becomes a very confident guesser.
And when you’re responsible for homes, communities, safety, compliance and resident outcomes, confident guessing isn’t good enough.
That’s why the conversation shouldn’t start with AI.
It should start with data.
Because before we ask whether AI can transform social housing, we first need to ask a much simpler question:
Can we trust the information we’re giving it?