It is nearly four years since OpenAI launched ChatGPT and artificial intelligence (AI) took on a huge significance across the globe. Generic AI is now becoming embedded across every sector, including property, making it widely accessible to businesses and professionals alike.
But the limitations of generic AI are now equally well understood, including the potential for hallucinations and bias. Limiting these risks is central to using AI effectively.
Paul Davis, chief executive of Nimbus, tells Property Week how the company has developed an approach to AI that delivers more reliable, transparent outputs for professional property work.
AI adoption is accelerating across property, but why isn’t generic AI enough for professional property work?
AI offers significant benefits, but real value comes from being able to use it while avoiding its limitations. What you often find with generic AI is that you can ask the same question in two slightly different ways and receive two different answers. That’s not acceptable in a professional property context.
Generic AI is very helpful for some things: it can help refine written content and summarise information. However, when it comes to professional work, where you’re advising clients or making investment decisions based on its outputs, that level of inconsistency becomes a real concern.
Why did Nimbus take a vertical approach with Agentic AI Professional rather than building a general-purpose AI solution?
We’re solving that first problem around generic AI and how you stop it from hallucinating by adopting a vertical approach. We have specialist AI assistants that we train on the key information for each specific problem.
An example is our legal assistant. In a property environment, I want that legal assistant to have a comprehensive understanding of the 1954 Landlord and Tenant Act. It needs to understand what Section 25 and Section 26 notices are, and the difference between them. We’ve built that assistant to understand the 1954 act and how valuation is carried out, then constrain it to very specific tasks. When you ask a sensible question, it knows which principles to rely on, producing much more consistent outputs.
Natural language questions often contain implied assumptions. For example, if you’re talking about a tenant, a lawyer would typically assume the 1954 act applies and that security of tenure exists, but that might not always be the case.
The AI understands those assumptions, while recognising where the act has been excluded. What that means is the system understands those underlying assumptions, so it avoids confusing commercial legislation with something like the Renters’ Rights Act. When we’re talking about rent reviews, the system understands the type of asset we’re discussing.
Responsible AI use is a growing concern for the property industry. How is Nimbus solving this?
The starting point is ensuring the information that goes into the prompt is accurate, structured and reliable. One of the key things is sourcing. It’s about understanding where the information has come from and how the AI reached its conclusion. Microsoft and RICS are two of the organisations we’ve worked with to help solve this challenge.
The system’s security is also critically important. If it’s referencing growth rates for the office market, but those figures have actually come from a residential Savills report, that’s not an appropriate source to rely upon. The important thing is that you can trace those references back to their original source and quickly identify when the wrong information has been used.
How does agentic AI overcome the inconsistency arising from different prompts and user inputs?
Number one is helping it interpret the intent behind the question. As the prompt enters the system, it’s enriched with trusted context, giving the AI a much clearer understanding of the background information. If you were asking your lawyer a question about a lease, you’d give them the lease first, and then they’d provide their advice. It’s a similar principle here.
The second thing is that we limit what the AI can do and ensure every recommendation it produces is grounded in real information. There are additional controls within the underlying system that we can tighten because we’re providing so much context from the outset.
Equally, we get the AI to check its own work. Most questions are assessed in two or three different ways, with the system validating that the response remains consistent across each approach before returning an answer.
What are the biggest research and workflow challenges that AI assistants can help property professionals overcome today?
There are many different disciplines across the property profession. Most begin with an initial research phase. Using the legal example, you have to read the lease first. Leases are often 50 pages long, and professionals become very efficient at working through them. However, that initial research phase can still be time-consuming. What is this asset? What’s the lease? Who are the owners? Bringing all of that information together takes time.
That’s where AI is particularly valuable, provided it’s looking in the right places and summarising trusted sources. Professionals still need to verify that the AI has used the right sources and that the output is accurate.
How can firms move from simply using AI tools to embedding AI into day-to-day workflows?
Microsoft talks about several stages of AI maturity, and I think most businesses are still at the first stage, using ChatGPT to answer individual questions. The next step is automating day-to-day tasks, perhaps summarising emails or drafting responses.
The real value comes when AI becomes part of the work you’re delivering. That’s where vertical AI tools can really accelerate delivery by supporting professional workflows.
Firms can also consider attempting to build these capabilities internally. I think that carries a degree of risk, because it’s difficult to know whether the system is producing reliable and consistent answers. That ultimately comes back to prompt engineering and understanding how to ask questions in a way that consistently produces reliable outputs. Being highly proficient at prompt engineering isn’t necessarily a core skill within the property profession.
Organisations should build confidence in these tools by running them alongside existing workflows, understanding how they operate and validating that the outputs are accurate before relying on them more widely.
Looking ahead, how do you see specialist agentic AI changing the way property professionals work?
I think the research burden will reduce significantly. That creates more capacity to concentrate on the areas where professionals add the greatest value.
I think you’ll see the focus of the property professional move towards much higher-value work. That means spending more time advising clients, negotiating transactions and making strategic decisions that drive stronger commercial outcomes, better lease terms and increased capital values.
Those are the activities that ultimately create value, but they’re often constrained by the amount of time available once the research has been completed.
This article originally appeared on Property Week on 22nd July 2026.
