A developer’s first question on any site is whether it can support what they want to build. That sounds simple. It isn’t. Answering it requires synthesizing data from a half-dozen independent sources: the municipal zoning code, the utility district records, the county permit history, the state legislature, regional construction cost benchmarks, and the active sales and rental market. None of those sources are in the same place, and none of them are designed to speak to each other.
I run a civil engineering firm, Calichi. We do predevelopment feasibility work for developer clients and public agencies across four offices. I’ve watched this study get done enough times to know exactly what it takes and exactly where the hours go.
What a predevelopment site feasibility study actually involves
Start with zoning. You pull the parcel data and the municipal code, identify the applicable zoning district, and read the development standards: density limits, height limits, front and rear and side setbacks, lot coverage, floor-area ratio, parking ratios, and whether short-term rental use is permitted in that zone. That tells you what the site can yield under existing entitlements.
Then you determine whether the project program fits that baseline. Often it doesn’t, which means a rezone is on the table. At that point the analysis forks. You’re now modeling two scenarios: what the site produces under current zoning and what it yields if the rezone succeeds. Each scenario needs its own development standards, its own yield calculation, and its own financial picture. You need to understand why a planning commission would approve the change, what the procedural steps are (zone change application, traffic impact analysis, public hearing, noticing, findings, site plan review), and what happens if the rezone is denied or slips six months. The rezoning analysis also requires comparable rezones in the jurisdiction, a read on community sentiment, and an explicit condition under which you’d recommend against pursuing it at all.
Then utility infrastructure. Water: is there a confirmed service territory? What does the water purveyor say about available pressure and capacity at this location? Is a new main extension required? Sewer: same questions, plus whether the collection system has capacity or whether an upsizing will be needed. Storm drainage: what’s the downstream constraint, the detention or infiltration requirement, and the applicable design standard? These answers don’t come from the same source. They come from the water district, the sewer authority, the city public works department, and sometimes the regional flood management authority, depending on the jurisdiction.
Then development impact exactions. System development charges in Oregon can be substantial on a multifamily project. California layers school impact charges, public art obligations, and affordable housing in-lieu requirements depending on the jurisdiction and product type. Washington adds GMA transportation concurrency charges. Illinois and Texas have their own structures. The applicable list depends on state and municipality, and leaving it out of the cost model is a meaningful error.
Then state housing legislation. In most of the jurisdictions where we practice, there’s a body of recently-enacted law worth checking: density bonus programs, height bonuses for affordable set-asides, by-right approval pathways for projects meeting objective standards, and limitations on local government’s ability to deny a conforming application. Oregon SB 1537 allows an additional story of height for qualifying housing projects. California’s Density Bonus Law provides density increases in exchange for affordable set-asides. Washington HB 1110 mandates middle housing in residential zones. Which of these applies to the specific project, at the specific location, and what it does to the unit count or approval timeline, is part of a complete analysis.
Then the construction cost picture. A developer needs to know what it costs to build the proposed program on this site. That means a cost model built by CSI division: site work, concrete, structural framing, mechanical and electrical and plumbing, finishes, specialties, general conditions. Not a single blended cost per square foot. A line-by-line estimate calibrated to the specific IBC construction type (type V-A wood frame carries different costs than type III-A or type I-A) and the quality tier. When below-grade parking is in the program, that adds a structural decision: concrete podium or glulam. Both carry different cost and schedule profiles, and both need to be modeled when the developer is choosing between them.
Then market data. Comparable sales per square foot. Long-term rents by bedroom type. Cap rates for this product and submarket. Land values from recent comparable transactions. These set the revenue side of the model and determine whether the cost picture supports the project.
All of that, gathered, synthesized, and organized into a workbook a developer or lender can actually use, is a predevelopment site feasibility study.
What it used to take us
At Calichi, a full site feasibility report covering constraints, utility infrastructure, and the entitlement path ran 20 to 30 hours of engineer time, and that study now comes back in 3 to 4 hours. That’s our own production-measured result.
The time didn’t go to the hard part. The judgment, the read on whether a specific rezone was realistic given the planning commission’s recent record, the interpretation of a non-standard zoning provision, the call on a utility capacity question where the records were ambiguous: those decisions deserve an engineer’s attention. The rest of it, the code lookups, the SDC schedule hunts, the permit history pulls, the comp research, the multi-source synthesis, the workbook construction, follows the same pattern on every project. That’s what consumed the majority of those hours.
How the agent does it now
The request comes in with the site address and the product type. From there:
- It parses the address, product type (eleven are supported: multifamily condo, multifamily apartment, multifamily rental, single-family residential, townhome, mixed use, office and industrial, heavy industrial, self-storage, retail, and hospitality), IBC construction type, and unit mix.
- It runs an eleven-category research pipeline: long-term rental comps, short-term rental income and occupancy, comparable sales, construction costs, permit history and development impact exactions, consultant and contractor availability, zoning and development standards, rezoning analysis, land comps, cap rates, and state housing legislation.
- For each category it fetches from available APIs first (government data, HUD, FRED), then runs targeted web searches, then validates and confidence-scores the merged result.
- It assigns a numeric confidence score from 0.0 to 1.0 to every data point. Government and primary official sources score at 0.90 to 0.95. Major aggregators score at 0.70. Single-source or estimated values score at 0.40 or below.
- It generates an eight-tab workbook with live formulas: Project Summary, Zoning and Entitlements, Unit Mix, Construction Budget, two revenue and hold tabs that adapt by product strategy (residential gets sale and rental tabs; office and industrial gets lease revenue and hold analysis; self-storage gets units and occupancy ramp), Deal Summary, and Sources and Market Data.
- Every projected figure carries its confidence rating inline, not on a separate lookup tab. Any figure with confidence below 0.70 is visually flagged with a warning background so whoever opens the file knows immediately what requires field verification before a commitment is made.
- Derived metrics, NOI, implied value, yield on cost, IRR, sale profit, and development margin, carry the confidence of their lowest-confidence input. A high-confidence cap rate applied to a low-confidence NOI estimate produces a low-confidence result, and the workbook shows that explicitly rather than hiding it.
- When a rezone is in scope, the agent models two scenarios with a financial impact comparison, the required planning actions, a numbered risk factor table, comparable rezones in the jurisdiction, community sentiment, and an explicit condition under which the rezone should not be pursued.
- When below-grade parking is in the program, it models concrete podium and glulam structural options in parallel, with side-by-side cost comparisons and schedule implications for each option.
- The Deal Summary recommendation carries a gating prefix based on research confidence: standard if eight or more research categories reached medium-plus confidence (0.60 or above), "PRELIMINARY" at five to seven categories, "INSUFFICIENT DATA" below five. The model doesn’t present a confident recommendation when the underlying research didn’t support one.
The QC gate
The output doesn’t go directly to the engineer. It runs through a validation pass first.
The gate checks that the workbook was built with live formulas across at least two tabs, not a static table. It confirms that confidence columns are present throughout. It scans for unfilled placeholder strings, TBD cells, and broken formula references. It verifies that the Deal Summary carries the required structured lists: project strengths, key risks, and next steps as numbered tables, with the risk table carrying severity and confidence columns, not narrative paragraphs alone.
There are pre-generation hard failures that block the workbook from being built. If every source came back at confidence below 0.40, the gate treats that as evidence that no real research ran and stops the job before building an empty-looking analysis. If the construction cost model is more than half zeros, the agent re-runs research before proceeding. An incomplete workbook handed to a developer is worse than no workbook: it reads like analysis and isn’t.
Every externally-sourced value, its source URL, its access date, and its confidence score gets logged in an audit trail that ships as a PDF appendix with the deliverable. The developer and the reviewing engineer can see exactly what the model is built on and which inputs need field verification before a decision gets made.
What stays human
The go/no-go decision. The interpretation of a non-standard zoning provision. The judgment on whether a specific rezone is realistic given what I know about that planning commission. The PE review on the civil and utility engineering components. The relationship with the developer, the lender, and the planning department.
The agent produces the research synthesis and the financial model. The engineer reviews the analysis, owns the judgment calls, and applies the stamp on any engineering component. That isn’t a caveat to the capability: it’s the design. A predevelopment study often sits at the front of a significant capital commitment. It needs to be an evidentiary package that can withstand scrutiny from a lender or a partner, and the PE review is the mechanism that makes it that.
The agent does the production pattern-work. The engineer does the engineering.
Where it applies
The agent supports all eleven product types: multifamily condo, multifamily apartment, multifamily rental, single-family residential, townhome, mixed use, office and industrial, heavy industrial, self-storage, retail, and hospitality. On applicable types it models short-term rental scenarios alongside long-term hold projections.
It runs wherever we do predevelopment work, across developer-client engagements, public agency planning studies, and internal site evaluations before design work starts. The jurisdictional variation is the part that’s slowest to do manually and most consistent to automate: different code editions, different SDC structures, different state housing statutes. The research pipeline resolves the jurisdiction from the address and applies the relevant statutory framework.
What this means for your firm
If you do predevelopment services, the shape of the problem is the same regardless of firm size or practice area. Licensed staff spending most of a study on research and assembly that follows the same pattern every time. The judgment at the end doesn’t consume those hours. The production work to get there does.
I’m not going to tell you the number at your firm. The 20-to-30-hour figure is Calichi’s own, measured on our own studies. Your number depends on your staff mix, your standards, and how your studies are structured today. The honest way to find it is to measure it on your actual work, not estimate it from someone else’s.
That’s what the first two steps of how I deploy are for. An AI Readiness Audit identifies where the highest-value automation sits across your practice. If predevelopment is one of those workflows, a Strategic AI Discovery puts my engineers alongside yours on the actual studies your firm produces, maps what’s patterned versus what requires judgment, and delivers a workflow-by-workflow picture of where the time goes.
I built this inside Calichi before I offered it to anyone. Predevelopment feasibility was one of the clearer cases: well-defined inputs, a structured output that has to be complete to be useful, research-heavy, with judgment confined to specific decision points that a licensed engineer can’t and shouldn’t hand off.