The decision gets made in a conference room eighteen months before the first truck rolls, and it is one of the most expensive decisions on the whole job: do we build this 220-unit residential project stick-built on site, ship the units as volumetric modules from a factory three states away, panelize the walls and floors and assemble them on site, or component-prefab just the bathroom pods and the MEP racks. Get it right and you shave four months off the schedule and lock the labor before the local trades tighten. Get it wrong and you have eighty modules sitting in a laydown yard accruing storage charges because the crane sequence was never coordinated, or you discover at month nine that the factory's tolerance stack does not meet the structural engineer's connection detail and now the EOR will not approve the as-built condition. AI changes the front of this decision: it runs the trade-off analysis across the prefab options fast, surfacing the cost, schedule, and quality implications of each path in hours instead of the weeks a manual study takes. But the constructability judgment and the decision itself stay human, and the memo the AI drafts is design-assist input, advisory to the engineer of record and architect of record who own their stamped scope, never a substitute for it. This lesson builds the workflow that produces the modular-versus-site decision memo for a 220-unit project, and it teaches you to verify the cost and schedule assumptions before the decision rides on them.

The Expensive Decision at the Front of the Job

The prefab-versus-site decision is expensive because it is early, irreversible, and touches every downstream system at once. It is early: you commit to a delivery method before the design is fully developed, betting on a schedule and a cost the design has not yet confirmed. It is irreversible: once the factory line is booked and the first modules are in production, you cannot quietly switch back to site-built without eating the deposit, the tooling, and the schedule you bought the method to protect. And it touches everything: the structural system, MEP coordination, transportation logistics, crane plan, site sequence, financing draw schedule, inspection regime, and the EOR and AOR scope all change depending on which path you take. A decision that early, irreversible, and connected punishes a rushed or under-analyzed answer.

Think of the four delivery methods as four different vehicles for the same trip, the way a builder already thinks about moving a heavy load: you can carry it piece by piece in a pickup (site-built, maximum flexibility, slowest), load it whole onto a flatbed (volumetric modular, fastest delivery but you are committed to the load fitting the road), break it into a few large crated assemblies (panelized, a middle path), or ship only the awkward heavy parts pre-assembled and finish the rest on site (component prefab, surgical). Each vehicle has a different cost, a different speed, a different risk if the road has a low bridge, and a different amount of finishing work waiting at the destination. Nobody picks the vehicle by reading a brochure; they pick it by running the route against the load, and that is the trade-off analysis this decision needs. The expensive part is not the analysis; it is making the call without one, or with one so slow it arrives after the design has hardened around an assumption nobody tested.

The Four Delivery Options and What Distinguishes Them

The decision framework runs across four delivery methods, and naming them precisely matters because they carry different cost, schedule, and quality profiles. Site-built (stick-built) is the baseline: everything is fabricated and assembled in place by field crews, with maximum design flexibility and the lowest tooling commitment, but the slowest schedule and the most exposure to field labor availability, weather, and site congestion. Volumetric modular ships fully finished three-dimensional room modules from a factory, with the fastest field schedule because the units arrive nearly complete and site work runs parallel to factory production, but the highest commitment to early design freeze, the tightest transportation and crane constraints, and a tolerance regime that has to reconcile factory and field. Panelized ships flat wall, floor, and roof panels that are assembled into the structure on site, a middle path that captures some factory efficiency and quality control while keeping more design flexibility and lighter transportation than volumetric. Component prefab ships only discrete assemblies, bathroom pods, prefabricated MEP racks, exterior wall cassettes, while the rest is site-built, a surgical approach that targets the highest-value, most repeatable, or most labor-intensive scopes for the factory and leaves the rest flexible.

The framework names all four rather than treating it as a binary modular-or-site question because the right answer is frequently a hybrid: panelize the structure, component-prefab the bathrooms and MEP risers, site-build the amenity spaces and the unique ground-floor retail. A 220-unit residential project is a strong prefab candidate precisely because it has high repetition, the same unit type stamped out hundreds of times, which factory production rewards, but the repetition is rarely uniform across the whole building, so the framework evaluates each scope against each method rather than forcing one method on the whole job. The AI's value here is that it can run the trade-off across all four methods, and across the mixed cases, fast enough that the team can actually look at the hybrid options instead of defaulting to the two everyone already has an opinion about.

The Cost, Schedule, and Quality Trade-Off

The trade-off is the heart of the decision, and it runs on three axes that pull against each other. On cost, the methods trade field labor for factory labor, tooling, and transportation: volumetric modular reduces field labor and its supervision, general conditions, and weather risk, but adds factory cost, module transportation, crane mobilization, and the carrying cost of modules that arrive before the site is ready, so the cost comparison is not a single number but a structure of costs that shifts between site and factory, and the answer depends on local field labor rates, haul distance, and repetition. On schedule, prefab's signature advantage is parallelism: factory production runs concurrently with site work (foundations, podium, utilities), so the field schedule compresses, which is why modular can shave months, but that advantage is real only if the early design freeze, the factory lead time, and the crane and delivery sequence are coordinated, because a module that arrives before its place is ready does not save schedule, it accrues storage.

On quality, factory production offers controlled conditions, repeatable assembly, and inspection at the line rather than at height in the weather, a real advantage for repetitive scopes, but it introduces the tolerance-stack problem: factory and field tolerances have to reconcile at the connection, and a module built to a tight factory tolerance still has to land on a field-poured foundation built to a looser field tolerance, so the quality question is not just "is the factory better" but "do the factory and field tolerances meet at the joint the EOR detailed." These three axes pull against each other: the fastest schedule (volumetric) demands the earliest design freeze and tightest logistics, raising coordination cost and tolerance risk; the most flexible method (site-built) gives up the schedule and the factory quality control. The trade-off analysis surfaces all three axes for each method so the team can see what each path costs on the other two, and that is the analytic and generative work the AI does fast, drafting the comparison into a readable structure so the team is looking at a populated trade-off matrix in hours, not assembling one over weeks.

The AI runs the trade-off across site-built, volumetric, panelized, and component prefab fast, surfacing the cost, schedule, and quality implications of each path. But the constructability judgment and the decision are the human's: the AI drafts the analysis and the memo, the team and the EOR and AOR own the decision, and the cost and schedule assumptions are verified before the decision rides on them.

The Analysis Is the AI's, the Decision Is the Human's

This is the decide-then-draft discipline applied to a delivery-method decision, the load-bearing distinction of the lesson. The AI is good at the analytic and generative work: it can pull the unit repetition from the design, estimate the field-labor displacement, structure the cost comparison across the methods, model the schedule parallelism, and draft the whole thing into a memo, fast. What it cannot do is own the constructability judgment, because that rests on things the AI does not carry: whether this factory can hold the tolerance this connection detail needs, whether the haul route has a clearance the modules will fit through, whether the local crane market can support the lift sequence in the delivery window, whether the early design freeze the modular path demands is realistic given where the design actually is, and whether the owner's financing and phasing can absorb the front-loaded factory payments. Those are judgments a builder makes by knowing the project, the market, and the methods, and they are the difference between a trade-off matrix and a decision.

So the workflow is decide-then-draft, not draft-then-rubber-stamp: the AI assembles the trade-off so the team can decide, then drafts the team's reasoning into a memo the EOR and AOR can act on. The inversion to avoid is treating the AI's recommendation as the decision, because that recommendation rests on assumptions (the cost basis, the schedule durations, the tolerance feasibility) the AI has not verified against this project's reality, and a decision this expensive and irreversible cannot ride on unverified assumptions. The AI accelerates the analysis to the point of decision; the human makes the decision and owns it.

The Memo Is Design-Assist Input, Advisory to the EOR and AOR

The output is a design-assist constructability memo, and "design-assist" defines exactly what it is and is not. Design-assist is the relationship where the contractor (or a specialty trade) brings constructability input to the design team during design, contributing the builder's knowledge of how the work goes together so the design is more buildable. The memo is that input in document form: it tells the EOR and AOR what the contractor's analysis says about the buildability and trade-offs of the delivery methods, the connection details a modular approach would require, the tolerance reconciliation the factory-to-field joint needs, the sequence the crane plan implies. But it is input, advisory to the design professionals, and does not change who owns the stamped design. The EOR owns the structural design and the connection details that make a modular or panelized system work; the AOR owns the architectural design and the assemblies. The memo informs their judgment; it does not substitute for it and does not carry their stamp.

This is the same boundary the program has held throughout: the licensed professional's stamped scope is the professional's, and AI-generated or contractor-contributed input is advisory to it, never a replacement. A memo that recommends a modular approach is recommending that the EOR detail the connections and tolerance regime that approach requires, and the EOR has to evaluate whether that is achievable and stamp it or not. If the memo asserts a tolerance the factory cannot hold and the EOR cannot detail around, the EOR says no, and that no is the design professional exercising the judgment the memo cannot exercise for them. So the memo is written and read as advisory design-assist input: it gives the EOR and AOR the contractor's constructability analysis to inform their stamped decisions, explicit that delivery-method feasibility ultimately depends on the design professionals' judgment about whether the details, tolerances, and assemblies can be designed and stamped.

Verify the Cost and Schedule Assumptions Before the Decision Rides on Them

The verification gate for this workflow is the assumptions, because the trade-off analysis is only as good as the numbers underneath it, and the AI's numbers are estimates until someone confirms them against this project's reality. The cost assumptions come first: the AI's field-labor displacement rests on a labor rate and productivity assumption, the factory cost on a vendor's pricing that may be a list price rather than a quote, the transportation cost on a haul distance and permit assumption, the crane cost on a lift count and mobilization assumption. Each is an input the team has to confirm, because the cost comparison between methods can flip on a labor rate or a haul cost, and a decision driven by an unverified cost basis is driven by the AI's guess. The dollars consequence is direct: if the modular path looks cheaper because the AI underpriced the transportation or the crane, the project commits to a method that is actually more expensive and discovers it after the line is booked.

The schedule assumptions are the second gate, where prefab's headline advantage lives or dies. The AI's schedule compression rests on the factory lead time, the parallelism between factory production and site work, and the delivery and crane sequence, each an assumption the team verifies against the factory's actual capacity, the site's actual readiness sequence, and the real delivery logistics. The classic failure is the module that arrives before its place is ready: the AI's schedule assumed just-in-time landing, but if the foundation or podium runs late, the modules sit in a laydown yard accruing storage and the parallelism that justified the method evaporates. So the team verifies the factory lead time against a real production slot, the parallelism against the real site sequence, and the delivery sequence against the real crane and route before the schedule advantage is treated as real. The AI's cost and schedule assumptions are surfaced explicitly and verified before the decision rides on them, because this decision is too expensive and too irreversible to ride on numbers nobody confirmed, and the verification is proportionate to the stakes: the higher the commitment a method demands, the more rigorously its assumptions are confirmed before the project commits.

The Applied Problem: The Modular-vs-Site Decision Memo for a 220-Unit Residential Project

Here is the exercise. Produce the modular-versus-site decision memo for a 220-unit residential project, evaluating the four delivery methods (site-built, volumetric modular, panelized, component prefab) against the named cost, schedule, and quality criteria, in the form of a design-assist constructability memo addressed to the EOR and AOR. Use the AI to run the trade-off analysis fast: have it pull the unit repetition from the design, structure the cost comparison across the methods (field labor versus factory cost, transportation, crane, and carrying cost), model the schedule parallelism and compression for each method, and surface the quality and tolerance implications, then draft the comparison into a readable trade-off matrix and a memo. The AI's job is to get the team to a populated, structured analysis fast; the team's job is to bring the constructability judgment to it.

Then do the human work the AI cannot. Verify the assumptions: confirm the labor rates, factory pricing, haul distance and transportation cost, and crane count against this project's reality, and the factory lead time, site-readiness sequence, and delivery logistics against the real schedule, flagging any place where the cost or schedule comparison would flip if an assumption is wrong. Bring the constructability judgment: whether the factory can hold the tolerance the connection details need, whether the haul route fits the modules, whether the local crane market supports the lift sequence in the window, and whether the early design freeze the modular path demands is realistic given where the design is. Decide, as the team, which method (or hybrid) to pursue, and draft the memo as the team's reasoning, explicit that it is design-assist input advisory to the EOR and AOR who own their stamped scope, with the cost and schedule assumptions and their verification stated so the design professionals can see what the decision rests on.

The deliverable is the modular-versus-site decision memo, and the lasting product is a workflow that uses AI to compress the trade-off analysis from weeks to hours while the team owns the constructability judgment and the decision, the design professionals own their stamped scope, and the cost and schedule assumptions are verified before the decision rides on them. The professional who masters this gets a populated four-method trade-off in front of the team early enough to consider the hybrid options, brings the builder's judgment about tolerance, logistics, and design maturity the AI cannot carry, and produces a memo that is honest design-assist input to the EOR and AOR rather than an AI recommendation dressed as a decision, the only way a delivery-method decision this expensive and irreversible can responsibly use the AI's speed.

Key Takeaways

• The prefab-versus-site decision is expensive because it is early (committed on incomplete design), irreversible (the factory line, deposits, and tooling are sunk once production starts), and connected (it touches structure, MEP, logistics, crane, financing, inspection, and the EOR and AOR scope at once), so it punishes a rushed or under-analyzed answer.
• The framework names four delivery methods: site-built (maximum flexibility, slowest, most field-labor exposure), volumetric modular (fastest field schedule via parallelism, earliest design freeze, tightest logistics and tolerance regime), panelized (a middle path with some factory efficiency and more flexibility), and component prefab (surgical, factory-building only the highest-value or most repetitive scopes).
• The right answer for a 220-unit project is frequently a hybrid, because high unit repetition rewards factory production but the repetition is rarely uniform across the whole building, so the framework evaluates each scope against each method rather than forcing one method on the whole job.
• The trade-off runs on three axes that pull against each other: cost (field labor traded for factory cost, transportation, crane, and carrying cost), schedule (prefab's parallelism compresses the field schedule only if design freeze, lead time, and logistics are coordinated), and quality (factory control versus the factory-to-field tolerance-stack reconciliation at the EOR's connection detail).
• The AI runs the trade-off across all four methods fast (the analytic and generative engines), surfacing the cost, schedule, and quality implications and drafting them into a readable matrix and memo in hours, so the team can actually consider the hybrid options instead of defaulting to a modular-or-site binary.
• The constructability judgment and the decision are the human's: this is decide-then-draft, where the AI drafts the analysis and the team brings the judgment (tolerance feasibility, haul route, crane market, design maturity, owner financing) and owns the decision, because the AI's recommendation rests on assumptions it has not verified.
• The memo is design-assist constructability input, advisory to the EOR and AOR who own their stamped scope: it informs the design professionals' judgment about whether the connections, tolerances, and assemblies can be designed and stamped, but it does not carry their stamp and does not substitute for their judgment.
• The verification gate is the assumptions: the team verifies the cost basis (labor rates, factory pricing, transportation, crane) and the schedule basis (factory lead time, parallelism against site readiness, delivery and crane sequence) against the project's reality before the decision rides on them, because the comparison can flip on a labor rate or a haul cost and the modules that arrive before their place is ready accrue storage instead of saving schedule.