AI can make comparable-property research much faster. A feasibility system can search large datasets, identify potentially similar properties, organize transaction information, and use that evidence to build pricing assumptions in a fraction of the time required by a traditional manual process. That speed is valuable when developers are screening a large number of potential opportunities.
The problem is that finding comparable properties isn't the same as finding the right comparable properties. An AI feasibility model can select properties that look similar based on location, size, price, or asset type while missing the characteristics that actually determine whether those properties compete with the proposed development. The financial model may calculate everything correctly while still producing a misleading feasibility result.
Why Comparable Properties Matter in Feasibility Modeling
Comparable properties often provide evidence for some of the most important assumptions in a development model, including achievable sales prices, rents, absorption, product positioning, and competitive supply. Those assumptions then flow into gross development value, development profit, residual land value, financing requirements, and projected returns.
That creates a chain between market research and financial feasibility. If the comparable evidence is weak, the assumptions derived from it can also be weak. Once those assumptions enter the model, the resulting financial metrics can look precise even though the underlying evidence isn't strong enough to support them.
Current real estate feasibility guidance continues to emphasize the importance of relevant comparable projects, recent market evidence, pricing, project positioning, and demand when assessing development viability.
The Closest Property Isn't Always the Best Comparable
Location is one of the easiest variables for an AI system to process. It can search within a defined radius, identify nearby properties, and rank them according to distance, transaction date, size, or price. That creates an efficient starting point, but geographical proximity doesn't automatically make two developments economically comparable.
Consider a proposed luxury apartment development located close to an older mid-market project. The existing project may have abundant transaction data and appear highly relevant because it is nearby, but the two developments could target completely different buyers. Differences in unit sizes, construction quality, amenities, parking, views, branding, and overall positioning could make the older property a poor basis for estimating the proposed project's achievable sales price.
The important question is therefore not simply whether a property is nearby. The more useful question is whether it represents a realistic alternative for the same buyer or tenant and provides evidence relevant to the specific assumption being modeled.
Development Feasibility Makes Comparable Selection More Difficult
Development feasibility creates another challenge because the proposed project may not exist yet. An analyst could be evaluating a 20-storey residential development while the available market evidence consists of completed buildings with different designs, ages, unit mixes, and positioning.
This means the analyst has to interpret comparable evidence rather than simply copy it into the model. A recently completed project may provide useful evidence for pricing, while a project under construction may provide better information about future competition. A rental property may help establish achievable rents but tell you very little about the sales price of a proposed owner-occupied development.
The purpose of the comparable matters as much as the comparable itself. Recent feasibility research similarly identifies comparable project analysis, pricing, demand, absorption, and product positioning as important inputs into development feasibility.
A Wrong Comparable Can Distort the Entire Financial Model
The consequences become more serious when comparable properties feed directly into the financial model. Imagine an AI system selects several premium developments as the primary evidence for a proposed mid-market project. The resulting sales-price assumption is higher than what the proposed development is realistically likely to achieve. That higher assumption increases gross development value, which increases projected profit and can subsequently increase the residual land value supported by the feasibility model.
The developer may then conclude that the site can support a higher acquisition price than it actually can. A problem that started with comparable selection has now affected the acquisition decision.
This is why comparable analysis shouldn't be treated as a separate market-research exercise. When comparable-derived assumptions feed a development model, the quality of those properties can influence the entire investment case.
AI Can Find More Comparables Without Finding Better Comparables
AI is very good at searching, filtering, classifying, and organizing large amounts of information. That can remove a significant amount of manual work from comparable research and make early-stage screening considerably faster. But data retrieval isn't the same as professional judgment. A system might identify properties with similar floor areas and locations while overlooking differences in buyer profile, quality, development scale, amenities, transaction conditions, or market positioning.
That distinction is becoming increasingly relevant as AI tools move into real estate underwriting. Current platforms are being designed to combine comparable evidence with development assumptions, financial analysis, and scenario testing rather than treating comparable selection as an isolated search task.
The important question for an AI feasibility model isn't simply "Which properties look similar?" It is "Which properties provide credible evidence for this specific assumption?"
The Source of the Comparable Matters Too
Even when an AI selects a genuinely similar property, the underlying transaction information needs context. A recorded sales price may have been affected by unusual financing, incentives, concessions, market conditions, or other circumstances that make it unsuitable as a direct benchmark.
This becomes particularly important when AI pulls information from multiple sources and converts it into structured model inputs. A clean spreadsheet can make inconsistent evidence appear more comparable than it actually is.
The analyst therefore needs to understand not only the property selected, but also the quality and timing of the underlying evidence. A recent transaction from a genuinely competing project may be more useful than a larger dataset of older transactions from properties that only superficially resemble the subject development.
The AI Should Explain Why a Comparable Was Selected
This is where explainability becomes critical. If an AI feasibility model selects five properties to support a sales-price assumption, the analyst should be able to investigate why those properties were selected. The relevant questions include whether they serve the same buyer segment, have similar unit sizes and quality, compete within the same submarket, reflect comparable market conditions, and require significant adjustments.
The system doesn't necessarily need to produce a long explanation for every property it finds. What matters is that the reasoning can be inspected when the comparable has a material influence on the financial model.
Feasibilitypro.AI currently combines market intelligence, feasibility-model generation, scenario analysis, and Excel-based analysis. Its public product information also emphasizes sourced market answers, editable Excel models, and the ability to modify assumptions and run sensitivity tests.
That type of transparency is useful because the analyst can move from the initial evidence into the model rather than treating the AI's output as a final answer.
Don't Let an Average Hide a Weak Comparable Set
A common problem is allowing the AI to calculate an average from a large group of properties and treating the resulting number as objective.
Suppose the system identifies ten properties and produces an average sales price of $450 per square foot. The number may appear more reliable because it is based on multiple observations, but the average doesn't tell you whether all ten properties actually compete with the proposed development.
If only four are genuinely comparable while the remaining six differ materially in quality, age, location, or positioning, the average can create false confidence. Reviewing the comparable set itself is therefore more important than simply reviewing the final average.
A smaller set of highly relevant evidence can be more useful than a large collection of superficially similar properties.
Scenario Testing Can Reveal Comparable Risk
One effective way to test comparable-property risk is to run the feasibility model against different reasonable evidence sets.
The base case might use the analyst's preferred comparables, while a downside scenario could use more conservative pricing evidence from competing developments. If changing the comparable set causes the residual land value or projected return to move significantly, the project is highly dependent on the quality of the market evidence.
That doesn't automatically mean the original comparable set is wrong. It means comparable selection is an important source of uncertainty and should receive additional attention before the project moves into deeper underwriting.
AI makes this process more practical because scenario analysis can be performed much faster. Feasibilitypro.AI, for example, currently describes conversational what-if analysis that allows users to change assumptions and compare outcomes without rebuilding the spreadsheet manually.
The Right Workflow Is Find, Review, Challenge, Then Model
A useful AI-assisted workflow should treat comparable selection as an iterative process rather than a single automated step. The system can search the market, organize potential comparables, identify similarities, and present the evidence to the analyst. The analyst can then review the properties, remove weak evidence, adjust assumptions, and test how those changes affect the feasibility model.
This approach is becoming more relevant as development platforms increasingly connect market evidence, development assumptions, scenario analysis, and financial underwriting in the same workflow. Deepblocks, for example, positions its platform around combining site and development analysis with feasibility and scenario testing.
The goal isn't to prevent AI from selecting comparable properties. The goal is to make sure those selections remain reviewable, explainable, and challengeable before they become financial assumptions.
How to Test an AI Feasibility Model's Comparables
The best evaluation is to use a real development opportunity rather than a clean vendor demonstration. Start by reviewing the comparable properties before looking at the final IRR. Compare their location, product type, scale, quality, unit mix, age, amenities, buyer or tenant profile, and market positioning against the proposed development. Then examine transaction timing and any circumstances that could make the observed pricing unusual.
After that, remove the weakest comparables and rerun the model. If the project's economics change materially, you've identified an important dependency that should be investigated rather than hidden inside the model. This test tells you much more than simply asking whether an AI tool can find comparable properties.
Practical Takeaways
The biggest risk from using the wrong comparable isn't an obviously incorrect number. It's a credible-looking assumption that flows through an otherwise accurate financial model.
Before relying on an AI-generated feasibility model, inspect the comparable properties behind its revenue assumptions. Look beyond distance and headline price, and consider whether the properties compete for the same buyers or tenants, have similar physical and commercial characteristics, and reflect reasonably comparable market conditions.
Then test what happens when weaker comparables are removed or more conservative evidence is introduced. If the project's economics change significantly, that dependency should become part of the underwriting discussion.
AI can make comparable research considerably faster, but speed shouldn't be confused with evidence quality. The real value comes when the technology helps analysts find relevant properties, understand why they were selected, challenge the evidence, and see how those comparable assumptions affect the development decision.
Top comments (0)