# AI Site Feasibility and Massing: What Generative Massing Tools Solve **Generative massing software and AI concept design are both sold as "AI for site feasibility," and they answer completely different questions.** One tells you how many units fit inside the zoning envelope. The other tells you what the building should be. Developers who buy the wrong one for the question in front of them either get a beautiful render that fails underwriting, or a solved envelope that nobody wants to fund. This is a guide to the split — what generative massing tools like TestFit are genuinely good at, where their output stops being the answer, and how the handoff between the two actually works on a real deal. --- ## What do people actually mean when they say "massing"? The word covers two jobs, and most of the confusion in this category comes from not separating them. **Massing as a solved envelope.** How many units fit on this parcel at this setback and this height limit. How much parking that triggers. What the resulting leasable area does to the pro-forma. This is arithmetic under constraints — there is a correct answer, and it can be optimised. **Massing as visual bulk.** What the volume reads like from the street, in daylight, next to its neighbours. Whether it looks like the kind of building that belongs there. What it is made of. This has no correct answer, only better and worse ones, and it cannot be optimised because the objective function is judgement. Both are legitimate uses of the word. They are not the same activity, they happen at different moments, and they are served by different software. Once you name which one you need, the tool choice stops being a debate. --- ## What does a generative massing tool actually solve? The category — TestFit is the name most developers know, and there are others — takes a site boundary and a rule set and solves for envelopes that satisfy it. The inputs are structured: parcel geometry, setbacks, height and storey limits, parking ratios, unit mix targets, and the practical geometry of corridors, cores and bays that determines what is actually buildable at that footprint. The output is a set of schemes with numbers attached. What that gets you, concretely: **A yield answer in minutes rather than a week.** The question "does this parcel support 84 units or 61" used to require an architect to model the options. Generative massing collapses that to a parameter sweep. **Sensitivity you can actually run.** What happens to yield if the height limit goes up one storey. What a variance on the rear setback is worth. What the parking ratio costs you in units. These are the questions that decide a bid, and they are only answerable when re-running the study is cheap. **A defensible number.** When the yield came out of a rule-based solve rather than a sketch, it survives contact with an underwriting committee. This is real, and it is not something concept design tools do. If your question is quantitative, this category is the answer and nothing in this article should talk you out of it. --- ## Where does the massing answer stop? The moment the question changes from *how much* to *what*. A solved envelope is an extruded volume with a unit schedule attached. It has no material. It has no fenestration logic. It has no relationship to daylight, to the neighbouring roofline, to how someone approaches the entrance. It cannot tell you whether the scheme reads as a decent piece of housing or as a parking podium with apartments on top — and those two can have identical unit counts. That gap matters commercially, not just aesthetically, because of who you have to convince next: **Capital partners** are underwriting a business plan, but they are also making a judgement about whether this sponsor builds things that lease and sell. A block diagram with a yield table does not answer that. **Planning officers and design review panels** in most jurisdictions respond to character, massing in the visual sense, materials and street presence. A compliant envelope is the floor, not the argument. **Brokers and early buyers** cannot sell from an envelope at all. Pre-sales and pre-leasing need imagery that shows what living or working in the building is like. So the yield study is necessary and it is not sufficient. It closes the underwriting question and opens the positioning one. --- ## Which tool for which job? The honest mapping. Note how much of this table points away from concept design tools — that is the point of the table. | Job to be done | Right tool | Why | |---|---|---| | Unit yield on a specific parcel | Generative massing software | Rule-based solve with a correct answer; not a visual problem | | Parking count and ratio testing | Generative massing software | Numeric constraint tied to unit count | | Setback and height envelope | Generative massing software, plus a planning consultant | Depends on local code, which imagery cannot encode | | Pro-forma sensitivity across schemes | Generative massing software + your financial model | Requires re-solving with numbers attached | | Aesthetic direction for a site | AI concept design (Nuit and similar) | Qualitative judgement; no optimal answer to solve for | | Material and façade options | AI concept design | Comparative, visual, decided by taste and market | | Investor and committee imagery | AI concept design | Needs to be legible to non-technical readers | | Broker and pre-sale marketing visuals | AI concept design | Buyers respond to atmosphere, not envelopes | | Schematic interior character | AI concept design | Sits downstream of both yield and exterior direction | | Code compliance sign-off | Licensed architect | Not an AI capability in either category | | Structural and MEP feasibility | Engineers | Not an AI capability in either category | | Permit and construction documents | Licensed architect, CAD/BIM | Regulatory, dimensioned, stamped | Three of those rows go to neither category. That is worth keeping in view when a tool in either camp starts describing itself as end-to-end. --- ## What does the handoff between them look like? The useful framing is that **massing output is an input to concept work, not a replacement for it.** Four things are worth carrying across the handoff: **The envelope.** Footprint, storey count, rough bulk. This is the constraint the concept visuals have to respect. If the solve said five storeys, generating an eight-storey concept produces a picture of a building you cannot build. **The program.** The mix the yield test assumed — unit types, ground-floor use, amenity. This determines what the building has to look like it does. **The site logic.** Where the entry landed, where parking went, where the open space ended up. These are the decisions that most affect how the building reads at street level, and they are usually already fixed by the time the yield study closes. **The binding constraints.** Which rule actually shaped the outcome — the height limit, the rear setback, the parking ratio. Knowing this tells you which parts of the massing are negotiable and which are not. Written down, those four are a paragraph. That paragraph is the brief for the concept phase. In Nuit the equivalent is the project brief that every generation reads from, so the visuals stay anchored to a site that has already been tested numerically rather than to whatever the model would have invented on its own. The result is imagery that is honest about the scheme: the right bulk, the right storey count, the right program — with material, light and character added on top. That is a much stronger thing to put in front of a capital partner than either half alone. Our guide to [pitching a design concept to investors](/blog/pitch-design-concept-to-investors/) covers what that package needs to contain. --- ## How does this fit a real deal timeline? A small residential development, three parcels under consideration: **Screening — massing software.** Run each parcel against the rule set. Two of the three clear the yield threshold; one does not and is dropped before any design spend. Days, not weeks. **Underwriting — massing software plus the financial model.** Sensitivity on the two survivors. One parcel is viable only with a setback variance; the other clears as-of-right. The bid strategy comes out of this. **Positioning — concept design.** On the parcel you are actually pursuing, generate exterior directions against the tested envelope. Two or three material and character options rather than one guess. This is where the earlier piece on [reducing feasibility study costs](/blog/reducing-architectural-feasibility-study-costs-with-ai/) does the cost math on why running several visual options is now affordable. **Capital and planning conversations — concept design output.** The deck gets both: yield table from the massing solve, imagery from the concept work. The two agree with each other because the concept respected the envelope. **Schematic design onward — licensed architect.** Both artefacts become the brief. The architect gets a project where the numbers are already tested and the direction is already agreed, which usually makes the engagement shorter and cheaper than starting from a blank site. The wider stack around this phase is covered in [AI tools for property developers](/blog/ai-tools-for-property-developers/). Note where the AI stops in this sequence. It stops before anything is stamped, priced by a contractor, or filed. --- ## What about multi-building sites? Sites with several buildings shift the balance again. The yield question gets harder — building-to-building relationships, shared parking, phasing — and stays firmly in the massing software's territory. But the qualitative question gets harder too, and in a way that specifically needs visual work: a masterplan has to read as one place. Consistency of character across buildings, the treatment of the space between them, how phase one looks when phase two is still a hoarding. These are not solvable by extrusion, and they are the questions a planning committee asks first. Concept tools handle the site-scale visual layer here — see [AI masterplan generator](/blog/ai-masterplan-generator/) for what that looks like in practice — while the density and unit numbers behind it come from the massing solve. Again: two tools, one handoff, and neither substitutes for the other. --- ## Where do AI floor plans sit in this? Between the two, and this is a common place to get burned. A schematic plan generated from a brief communicates program, adjacencies, circulation logic and rough proportion. It is genuinely useful for showing an investor how the building works, and for giving an architect a starting point that reflects your intent rather than their assumption. It is not a measured plan. It does not carry dimensions you can build from, it does not resolve structure, and it does not encode the corridor-and-core efficiency that a yield solve depends on. If you need a plan that supports a unit count, that plan comes from the massing tool or the architect — not from an image model. We have written separately on [how accurate AI floor plans actually are](/blog/ai-floor-plans-accuracy/), and the summary is: schematic-accurate, not construction-accurate. The practical rule: use generated plans to communicate intent, use solved plans to support numbers. --- ## How do Nuit's models fit into this? Nuit runs three models, and they are three different jobs rather than three quality tiers. **Light** is the faster and cheaper of the three — it costs half what a standard generation costs. It is the one to use when you are exploring: sweeping through directions on a site, testing whether a material family is worth pursuing at all, generating enough variants that individual misses do not matter. **Nuit N** and **Nuit F** behave differently from each other and produce a different character of result. Neither one is the upgrade. Which one suits a given project is a judgement the user makes by trying both. What none of the three do: compute yield, check zoning, verify a setback, or size a structure. That is not a roadmap gap we are being coy about — it is a different category of software, and the honest answer for those questions is the one at the top of this article. --- ## So what should a developer actually buy? Match the tool to the question you are being asked most often. **If your bottleneck is deciding what to bid** — you are screening parcels, and the thing slowing you down is not knowing what fits — buy generative massing software. That is its job and it does it well. A concept design tool will not help you here and you should not let anyone sell you one for it. **If your bottleneck is what happens after the numbers clear** — you have a viable parcel and now need something a capital partner, a planning officer or a broker can respond to — buy a concept design tool. Massing software will not produce that; it was never trying to. **If you are running deals end to end**, you need both, and the total is still a small line item against a single week of architect time at the concept stage. The mistake is not buying both. The mistake is expecting either one to do the other's job and discovering it at the point where the deck is due. The clean version of the split: **massing software tells you the building is worth doing. Concept design tells you what it should be.** Neither is the whole feasibility answer, and any tool claiming to be both is describing something it does not do. --- ## Related reading - [AI Tools for Property Developers](/blog/ai-tools-for-property-developers/) — the wider AI stack across the three phases a developer controls… - [Reducing Feasibility Study Costs with AI](/blog/reducing-architectural-feasibility-study-costs-with-ai/) — where the cost savings in the visual side of feasibility actually come from… - [AI Masterplan Generator for Multi-Building Sites](/blog/ai-masterplan-generator/) — site-scale concept layouts and consistency across buildings… - [Pitch a Design Concept to Investors with AI](/blog/pitch-design-concept-to-investors/) — what an investor-ready concept package has to contain… - [How Accurate Are AI Floor Plans?](/blog/ai-floor-plans-accuracy/) — schematic-accurate versus construction-accurate, and where the line sits… ## Frequently Asked Questions ### What does a generative massing tool actually do? It takes a site boundary and a set of rules — setbacks, height limits, parking ratios, unit mix, corridor and core standards — and solves for building envelopes that satisfy those rules while maximising a chosen variable, usually unit count or leasable area. The output is a quantitative site test: how many units fit, how much parking is required, what the resulting yield looks like. TestFit is the best-known product in this category. The output is a solved envelope, not a designed building. ### Should I use a massing tool or an AI concept design tool? Use whichever matches the question you are being asked. If the question is how many units fit, what the parking count is, or whether the pro-forma clears, use a generative massing tool — that is a numeric problem and concept design tools do not solve it. If the question is what the building should look like, what it is made of, and how it reads to a buyer or an investment committee, use a concept design tool. Most development projects need both, in that order. ### Can AI image tools calculate unit yield or check zoning compliance? No. Image-generation tools including Nuit produce visual concepts, not measured or rule-checked output. They do not compute floor area ratios, verify setbacks, count parking stalls, or confirm code compliance. Treating a rendered massing image as a yield study is how developers end up with a scheme that cannot be built at the density the pro-forma assumed. Yield and compliance come from massing software, a planning consultant, and eventually a licensed architect. ### Why can't the yield-optimal scheme go straight into an investor deck? Because a yield-optimal envelope is a correct answer to a numeric question and says nothing about how the building reads. It has no material, no light, no street-level presence and no character. Investment committees, planning officers and buyers respond to what the building will feel like, and an extruded block does not carry that. The envelope is the constraint the concept work has to respect, not the thing you present. ### What does a massing study hand over to concept design? Four things worth carrying across: the envelope itself — footprint, storey count, overall bulk; the program mix that the yield test assumed; the site logic, meaning where the entry, the parking and the open space landed; and the constraints that shaped the outcome, such as a height limit or a required setback. Written as a short brief, those four become the context that keeps the concept visuals honest against the numbers. ### Does Nuit replace TestFit or similar massing software? No, and it is not trying to. Nuit produces concept-stage visuals — exteriors, schematic floor plans, interiors, and site-scale concept layouts — from a written brief. It does not solve unit yield, parking ratios or setback compliance. For a unit-yield study on a specific parcel, a developer should use a generative massing tool. Nuit is the step that follows, when the question moves from how much fits to what it should be. ### Where in a development timeline does each tool belong? Massing software belongs at screening and underwriting, when you are deciding whether a parcel works at all and what to bid. Concept design belongs immediately after, when you need something to show a capital partner, a planning pre-application, or a broker before pre-sales. The licensed architect belongs after that, taking both the numbers and the concept direction into schematic design and everything downstream. --- **Try Nuit free — 10 generations, no card required.** Once the numbers clear, turn a tested envelope into concept visuals a capital partner can respond to. [Start your project →](https://nuit.archi)