Here is a fact that should bother anyone who develops housing: planning a multifamily project takes longer than building it. Research from the Federal Reserve Bank of Atlanta found that projects spend, on average, three to four months longer in the planning stage than in construction. For scale, Census Survey of Construction averages for 2001 to 2016 put the build itself on a 10-or-more unit project at roughly 13 months.
So the slow part of development is not the part with cranes in it. It is the part with PDFs.
That is where the time-saving pitch for AI usually starts, and where it usually goes wrong. Cutting pre-development time is not one problem. Pre-development is two clocks running in sequence, owned by different people, and software only touches one of them.
What are the two clocks in pre-development?
Your clock covers everything before you file: finding candidate sites, pulling ownership, interpreting zoning, checking hazards, testing what the parcel can hold, deciding whether to pursue it. You set the pace. However long it takes, that is a decision you made about how you work.
Their clock starts when you submit: completeness review, staff report, notices, hearing calendars, environmental review, appeals. You do not set this pace. You wait.
Every honest conversation about cutting pre-development time starts by separating these, because the techniques that compress one do nothing to the other. A tool that promises to shorten your entitlement is selling you something that does not exist.
Which parts of pre-development cannot be compressed?
The institutional clock is slow for structural reasons, and the numbers are not encouraging.
Neighborhood opposition alone adds an average of 7.4 months to a development timeline when it appears, according to the 2022 NMHC and NAHB cost-of-regulations study. That same study found regulation accounts for 40.6 percent of multifamily development cost.
In California the environmental review layer adds its own exposure. A Holland & Knight study of 498 appellate decisions from 2013 to 2023 found housing was the single largest target of CEQA appeals at 25.5 percent of cases, that nearly 90 percent of challenged housing projects were infill, and that CEQA litigation typically takes four to five years to resolve. In related trial-court figures, lawsuits filed in 2020 challenged roughly 48,000 approved housing units, just under half that year’s housing production. The study’s authors back Proposition 45, so read the framing with that in mind, but the decision counts are court records.
No amount of data tooling moves any of this. A council vote is a council vote.
Which parts can be compressed?
Your clock is almost entirely an information problem, which is exactly the kind of problem that collapses when the data is already assembled.
Consider what screening one parcel traditionally requires: the assessor portal for ownership, the zoning map and then the code text behind it, FEMA’s flood viewer, a utility map, sales comps from a third source, and an incentive check against QCT or DDA boundaries. Six systems, none of which talk to each other, each with its own search interface, and no two of which name the same parcel the same way.
Now consider repeating that across every parcel in a submarket to find the handful worth a site visit.
That arithmetic is the whole argument. Manual screening does not scale, so developers cope by screening fewer sites – which is not a time saving, it is a narrower funnel. We broke the comparison down parcel by parcel in AI Site Selection vs Manual Parcel Screening.
What does “AI-native” actually mean?
The term is doing a lot of unearned work in this market, so it is worth being specific.
AI bolted on is a chat window sitting on top of a system that still stores zoning as a PDF link. You ask it a question, it tells you which document to go read. The lookup still happens in your head.
AI-native means the parcel is the unit of data, every layer is already joined to it, and the question is the interface. You do not search six systems and reconcile the answers. You describe the site you want and the screen returns the parcels that match, with the reasoning attached.
The difference shows up in what you can ask. A bolted-on tool answers “what is the zoning here.” A native one answers “which parcels in this corridor are over 7,000 square feet, outside a flood zone, zoned for more than four units, and not owned by the same party as their neighbor.”
Why is selection worth more than speed?
This is the part most articles about AI and development miss, and it is the part that actually earns the headline.
The biggest time saving in pre-development is not doing diligence faster. It is not entering entitlement on a site you should have rejected in week one.
A site with a methane zone, a historic overlay or a flood designation you discovered late in diligence has already consumed the expensive clock – the one you do not control. Given that planning alone runs months longer than a build that itself averaged over a year, a dead deal that reaches entitlement costs you well over a year of institutional calendar spent on nothing. You cannot get that back, and the cost compounds, because it crowded out the deal that would have worked.
So the honest version of cutting pre-development time in half is this: you halve the clock you own, and you stop spending the clock you do not own on sites that were never going to survive it. The first is a direct saving. The second is larger, and it only comes from screening deeply enough, early enough, to say no while saying no is still free. That dynamic is the subject of How Developers Fall Behind in Pre-Development.
Where does ArchiWise fit?
Everything above is about the first clock, and that is the only clock we claim to affect.
Filters run across a drawn search area or the current map view – zoning, lot size, vacancy, ownership type, hazards, incentives – and return a live parcel count you can export or push into a project. Open any single parcel and it carries thirteen tabs covering property, zoning, ownership, sales, financing, taxes, permits, listings, comps, hazards, signals, area and incentives. The agents answer questions against that data in plain language, so the zoning interpretation happens where the parcel is rather than in a separate tab.
What that replaces is the six-system shuffle, not the hearing calendar. For how the hazard layer fits into early screening, see our hazard risk assessment guide.
What can this not do?
It will not shorten a staff review, predict a council vote, survive an appeal, or tell you what a planning director will say about your massing. It will not make a jurisdiction move faster, and it does not know what is coming that has not been recorded yet.
What it does is make sure that when you enter the slow clock, you enter it with the right site, and that the months you spend waiting are months spent on a deal worth waiting for.
If you want to see where that screening sits in the wider workflow, our AI site selection page walks through it from the first search onward.
Frequently asked questions
How long does pre-development actually take?
Longer than construction. The Federal Reserve Bank of Atlanta found projects average three to four months longer in planning than in building. The exact figure varies by region – the Northeast and West run longer than the Midwest and South – and mixed-use projects take longer than multifamily-only.
Can software shorten the entitlement process?
No. Entitlement runs on the jurisdiction’s calendar: completeness review, staff reports, noticing periods, hearings and appeals. Software can prepare you for it and can keep you out of it on sites that will not survive, but it cannot make a city move faster.
What is the difference between AI-native and AI-enabled software?
AI-enabled usually means a chat interface added to an existing system, which still points you to a document to read. AI-native means the underlying data is already joined at the parcel level, so the answer comes back resolved rather than as a link.
What slows pre-development the most?
Opposition and review, not research. Neighborhood opposition adds an average of 7.4 months on its own, and in California CEQA litigation typically takes four to five years to resolve. Research time is the part you control, which is why it is the part worth optimizing first.
Is faster screening the main benefit?
No. The larger benefit is rejecting unsuitable sites before they consume entitlement time. Screening speed is worth weeks; avoiding a dead entitlement is worth more than a year.