The real estate industry has a complicated relationship with AI. Half the conversations you'll find online are either breathless "AI will replace analysts!" takes or dismissive "Excel is all you need" counterarguments. Neither is particularly useful if you're actually trying to build or assess projects.
So let's cut through it. Here's what AI is genuinely doing well in feasibility right now, where it's still falling short, and what that means for how you work.
The Hype First (So We Can Get Past It)
The most oversold idea is that AI can autonomously run a development feasibility from end to end. Feed it a site address, and out comes a viable business case. That's not where we are not even close.
Feasibility isn't just a calculation problem. It's a judgment problem. You're making calls on absorption velocity, achievable sales rates, construction cost contingencies, and what a planning authority is actually going to approve versus what the zoning technically allows. AI doesn't have the local market intuition for that, and a lot of it frankly can't be sourced from training data.
The other overhyped thing is "AI-generated reports." What most tools produce is AI-assisted formatting and structuring of outputs you could have made yourself. Calling that intelligence is generous.
What AI Is Actually Good At Right Now
Automating the Repetitive Structure of Models
One thing that genuinely saves time is using AI to set up model frameworks cost category structures, cash flow waterfall templates, sensitivity table layouts. Tools like Northspyre have leaned into this on the project cost management side, helping owners track committed versus projected costs with a layer of automation that reduces the manual entry burden. That's real. It's not glamorous, but it compounds over a 3-year development program.
Scenario Modelling at Speed
Running 50 sensitivity scenarios manually is tedious. AI-assisted tooling can accelerate that loop significantly. EstateMaster has been around long enough to know what practitioners actually need from a feasibility engine, and their approach has always been grounded in the numbers rather than the marketing. Where AI adds value there is in how quickly you can stress-test assumptions across multiple variables simultaneously yield on cost, construction cost escalation, sales rate changes without rebuilding the model each time.
Zoning and Site Analysis as a Starting Point
This is probably the most genuinely useful application. Deepblocks does interesting work here using AI to interpret zoning codes and quickly assess what's developable on a given parcel. It's not a planning approval, and it's not a substitute for a planning consultant, but as a first-pass filter when you're looking at multiple sites? It saves meaningful time. The key word is "starting point." Anyone using it as a final answer is going to get burned.
Soft Cost Estimation and Benchmarking
Getting to a rough feasibility number quickly before you've spent money on consultants is something AI-enhanced tools are getting better at. Aprao has built a product aimed at exactly this use case: early-stage feasibility with enough structure to be useful without requiring you to already know your construction cost breakdown in detail. The question isn't whether it replaces a detailed QS it doesn't but whether it helps you decide if a site is worth pursuing in the first place. For that, it's genuinely useful.
Making Assumptions Transparent and Auditable
This is underrated. One of the real problems with Excel-based feasibilities is that assumptions get buried and nobody can trace where a number came from six months later. feasibilitypro.ai has thought about this — building assumption transparency and auditability into the model structure rather than treating it as an afterthought. That's less about AI and more about discipline, but AI can help surface which assumptions are driving your returns versus which ones barely move the needle.
Where AI Is Still Genuinely Struggling
Market Data Quality
AI tools are only as good as the data they're working with. In established markets with good transactional transparency, you're fine. In GCC markets, secondary cities, or anything off-plan dominated, the data gaps are significant. Garbage in, garbage out — and the AI doesn't know the difference.
Construction Cost Accuracy
Generic construction cost benchmarks are almost useless for feasibility. Costs vary massively by submarket, contractor availability, supply chain conditions, and specification level. AI tools that spit out a per-square-metre number with confidence are the ones you should be most skeptical of.
Planning and Regulatory Nuance
Zoning codes are a starting point, but planning decisions involve discretion, political context, design quality assessments, and community responses that no AI currently handles well. A tool that tells you a site can support 200 units doesn't know that the local council has been rejecting anything over 6 floors for the last two years.
Replacing Experienced Judgment on Assumptions
The biggest risk is using AI-assisted tools to bypass the hard thinking on key assumptions. What's the right sales velocity for this product type in this market right now? What contingency is appropriate given where the contractor market is? Those are questions where experience matters, and AI doesn't have it.
The Practical Takeaway
AI in real estate feasibility is most useful as a productivity layer — not a replacement for the analytical work. The developers getting genuine value from it are using it to do more analysis faster, not to skip the analysis entirely.
The tools that are worth paying attention to are the ones that have been built by people who understand development finance — not generic AI platforms that have been dressed up with real estate language. That's a meaningful distinction, and it matters when you're making decisions that involve tens of millions of dollars.
The honest position: if you're not using any of these tools yet, you're probably leaving efficiency on the table. If you think they can replace experienced underwriting, you're going to make expensive mistakes.
Top comments (0)