Of everything I have done in property, registered children's homes are the part I am most careful talking about. Not because the numbers do not work. They do. Because it is the one corner where getting it wrong does not mean a void or a thin yield. It means a vulnerable child in a setting that should never have existed. So when people ask me whether AI is “transforming” this sector, my honest answer is that it is transforming the easy half and leaving the hard half exactly where it was. The gap between those two halves is the whole sector.
Start with the demand, because that part is real
The shortage of registered places is structural, not cyclical.
The need here is not a market trend. It is a shortage the state is legally obliged to fix and currently cannot. The number of children reported to Ofsted as placed in unregistered homes rose from 147 in 2020-21 to 982 in 2023-24. As at September 2024, around 775 children were living in unregistered accommodation, staying an average of about six months, in settings meant for emergencies. Running an unregistered children's home is illegal. It happens anyway, because there are not enough registered places, and councils are left choosing between bad options.
Around 84 per cent of residential care settings on Ofsted's register are privately owned. Local authorities pay premium rates to independent providers to secure beds because they have no alternative. The Children's Wellbeing and Schools Act 2026 has handed Ofsted new powers to fine illegal provision, which is welcome, but a fine does not create a bed. The structural answer is more good, registered provision, and that is a property and development problem before it is anything else.
There is a wrinkle worth knowing. Those new fining powers apply to children's homes but not to supported accommodation, which is the fastest-growing and least-regulated part of the same system. For anyone working here, the line between a registered children's home and a supported living setting is not a technicality. It changes the regulator, the standards, the commissioning route and the risk profile of the whole deal. AI will not draw that line for you. You have to know which side of it you are building on before you commit a penny.
Where AI genuinely earns its place
Sourcing sites for this kind of use is brutal, manual work, and this is where AI has changed my week.
Finding the building is the start. The right property for a small children's home, typically one to six children, is specific. Domestic in feel rather than institutional, the right size and layout, the right area, away from the wrong neighbours and near the right services. Screening stock at that level of nuance used to be legwork. Now I can filter at scale and only walk the ones that fit.
Tracing ownership is the part that used to eat days. Land in this country threads itself through estates, family trusts and nominee companies. AI maps the title, surfaces the corporate entity behind it, tells me how long it has been held and flags the ownership changes that hint at a seller. Off-market is where these deals live, and research that took a week now takes minutes.
Planning and environmental context comes back in the same pull. Is a change of use realistic. What is the area actually like. Any ground stability, flood or contamination flags worth a closer look. The end-value side, the comparable sales that tell me whether a conversion stacks up, lands in the same query rather than three separate ones. And a pipeline is never static, so re-screening every site as new data arrives, rather than rediscovering a problem in a Monday meeting, is exactly the unglamorous work AI is good at.
There is also the question of where to build at all. Need is not spread evenly across the country. Some authorities are desperately short of beds and paying accordingly, others much less so. Layering local demand and commissioning patterns over a sourcing search means I am not just finding a suitable building, I am finding a suitable building in a place where the placements actually exist. That used to be tribal knowledge held by a handful of operators who had been in their patch for decades. It is now something you can ask a question of.
And then you hit the gate
Everything above is information work, and information work is what AI does well. The moment you cross into actually creating a registered children's home, the machine stops being useful and the human becomes the entire job.
A children's home must be registered with and inspected by Ofsted under the Care Standards Act 2000 and the Children's Homes Regulations 2015. The building has to meet the standards. The care has to meet the standards. In the model most investors use, you own the property and lease it to a regulated operator who runs it under their own registration, and you are not involved in delivering the care. That separation is sensible, but it means the deal lives or dies on the quality of the operator, and judging an operator is not something you hand to software.
It is also slower than people expect. Ofsted has been hit with an unprecedented rise in registration applications and has had to revise how it prioritises them, so registrations are taking longer. An AI tool will happily tell you a building is perfect for the use. It has no idea whether the operator will pass inspection, whether registration lands this year or next, or whether the home will be any good once it opens. That is all judgement, and the stakes on that judgement are a child's safety.
Say the ethics out loud
The sector is healthier when people are blunt about this. There is real public scrutiny of private profit in children's social care, and Parliament has openly worried about over-reliance on private providers. That scrutiny is not unfair. When you are housing looked-after children, “the numbers work” is not a sufficient answer on its own.
My own line is simple. The income in this sector is stable precisely because the demand is non-discretionary and the leases are long, but that stability is a consequence of providing something genuinely needed, not a trick to be gamed. If your plan only works by cutting the standard of the building or backing a weak operator, it is not a plan, it is a future enforcement notice. Build to a standard you would be content to have inspected, back operators you would trust with your own family, and the returns look after themselves. Get that the wrong way round and AI will simply help you find the wrong building faster.
What I actually use the tools for
So here is the honest division of labour. AI does the sourcing, the ownership research, the planning and market context, the appraisal maths and the pipeline monitoring. It compresses what used to be weeks into an afternoon, which means I can look at far more sites and put real attention on the ones that matter.
The data underneath those tools is the bit I am fussy about, because a children's home appraisal built on stale ownership or thin comparables is worse than no appraisal at all. It is a confident mistake in a sector that cannot afford them. It is the reason we built Propalt to pull from the registers daily rather than scrape and hope, though the principle stands whatever you use.
But the home itself, the registration, the operator, the standard, the safeguarding, the decision to proceed at all, that stays human, and it should. AI can find the building. It still cannot register the home, and in this sector that is not a limitation to engineer around. It is the point.
Sources and further reading
— Financial sustainability of children's care homes — Public Accounts Committee, Parliament (2026)
— Children's Wellbeing and Schools Act 2026, key reforms — Together Trust
— Children's homes registration and Ofsted prioritisation — Ofsted social care blog
— How children's home property investment works, and the rules that govern it — Prem Property
Co-founder of Propalt and a property investor and developer for 20 years
