AI workflow

AI Site Selection for Commercial Real Estate: A Land-First Screening Workflow

How AI site selection screens parcels for a specific asset across one or many markets, which data layers it runs on, how scoring works, where retail trade-area tools fit, and a car wash screen in Hillsborough County, Florida.

By Nathan Robinson, Founder, Plotzy Published 14 min read

AI site selection is software that screens land for a specific asset across one or many markets, then ranks what passes for a person to visit. It joins parcels to zoning, future land use, traffic, demographics, flood, and utility data, and uses language models where the rules live in text, mainly the zoning code. Retail analytics platforms judge how a store will perform in a trade area. A land-first workflow answers the earlier question: which parcels can legally and physically hold the asset at all.

This page covers commercial real estate site selection by asset type, the data layers, scoring, and the site visit, then works one real example: express car wash sites in unincorporated Hillsborough County, Florida, screened on November 19, 2026.

What does AI site selection do, and what does a person still decide?

It does the wide, repetitive part: every parcel, every rule, every market, the same way each time. People do the narrow part: the shortlist, the drive, the call to the owner, and the decision.

StageSoftwarePerson
Define the programTurns the buy box into testable rulesSets the buy box and the deal-breakers
ScreenTests every parcel against zoning, size, flood, trafficReviews flagged rows
Read the codeFinds the use, its status, and the sectionSettles ambiguous uses and planned developments
ScoreRanks by weighted criteriaSets the weights and challenges the ranking
VisitPrints a site packet with maps and citationsChecks access, visibility, neighbors, and what the maps miss
ActPushes the shortlist and owners to a CRMCalls the owner, negotiates, and decides

Land-first or trade-area-first: which tool answers which question?

Most products sold as site selection software, including most AI tools for site selection, start from consumers and stores. A land-first AI site selection tool starts from parcels and the code. They answer different questions, and a team may need both.

Vendor (sites checked November 19, 2026)What its own site says it doesStarting point
Placer.aiA location intelligence and foot traffic data platform, with retail analytics for store benchmarking and site selectionVisits and trade areas
Buxton, now part of AudienseLocation intelligence that scores sites and analyzes trade areas; its SCOUT platform covers market planning and whitespace analysisConsumers and trade areas
TangoPredictive Analytics, described as location intelligence software for retailers, alongside lease and portfolio toolsA retailer's store network
KalibrateLocation intelligence for retail, competitor trade-area insights, and fuel and convenience store volume forecastingSales and volume forecasts

These vendors own trade-area analytics, and a land-first workflow does not try to replace them. The difference is the unit of work:

  • Trade-area tools take a candidate location and predict its performance from who lives, works, and travels nearby.
  • A land-first workflow takes a market and returns the parcels where the use is allowed, the site fits, and the owner can be found, before anyone forecasts sales.

For visit-driven assets such as QSR and car washes, run both: screen the land, then forecast the survivors. For assets that do not live on foot traffic, such as industrial outdoor storage or warehouses, zoning and land constraints decide most of the screen.

What criteria matter by asset type?

The program sets the rules. The zoning test comes first, because it is cheap to run and every later step depends on it. The table uses unincorporated Hillsborough County's use table (LDC Sec. 2.02.02) as the example code.

AssetZoning test in unincorporated Hillsborough CountyLayers that decide the restGuide
Self-storage"Mini Warehouses": Conditional (administrative review) in CG, CI, and M, under Sec. 6.11.60Households and existing supply in the trade area, visibility, floodSelf-storage
Express car wash"Car Wash Facilities": permitted in CG and CI; conditional in CN, where Sec. 6.11.121 requires self-service or automated washes under a roof with at least two sides; accessory only in MTraffic counts, competing washes, lot depth for stackingCar wash
Industrial outdoor storage (IOS)"Open Storage" and contractor storage yards: permitted in CI and MTruck access, impervious limits (75% in CI and M), floodIOS
QSR with drive-through"Restaurants with Drive-Up Facilities": conditional in CN, CG, CI, and MTraffic counts, daytime population, access and median openingsGas station and QSR
Multifamily under Live LocalState law requires counties to allow it on land zoned commercial, industrial, or mixed use if at least 40% of units are affordable rentals for 30 years (Fla. Stat. 125.01055(7)(a)1). The county lists BPO, OR, CN, CG, CI, and M as eligibleFuture land use, utilities, schools, and the county's lawsuit (see below)Multifamily, Live Local Act
Data centerThe use table has no row named data center, so classification may need a county interpretationPower and substation distance, water, fiber, floodData center site selection

Hillsborough County has filed a lawsuit challenging the Live Local Act's constitutionality. Its Live Local page says any decision to proceed under the act is at the applicant's risk. A multifamily screen there should carry that flag on every row.

Which data layers does a site selection model need?

Each layer has a source, a date, and a known way of being wrong. Record all three.

LayerSource in the exampleWhat it answersWatch for
ParcelsHillsborough County Property Appraiser parcel file, county-hosted, 531,626 parcelsSize, use code, address, owner of recordAcreage is the appraiser's, not a survey; its zoning field can lag the atlas
ZoningCounty zoning atlas layer (last updated November 14, 2026) and the LDC (Supplement 58)District and rezoning numberThe layer warns it may not reflect current status
Future land usePlan Hillsborough map service and plan Table 2.2Density and FAR caps, PD requirementsThe plan's cap can be lower than the zoning's
TrafficFDOT Annual Average Daily Traffic layer: 2025 counts, source data created November 14, 2026Vehicles per day by road segmentCovers roads in FDOT's Traffic Characteristics Inventory, not every local road; a nearby segment is not proof of frontage
DemographicsCensus American Community Survey 5-year estimates (2020 to 2024 is the latest)Households, income, age by tract or block groupMargins of error grow as areas shrink
FloodFEMA's National Flood Hazard Layer, via the county's copy (latest LOMR effective September 28, 2026)Special Flood Hazard Area (SFHA)Screening only; elevation and a survey decide the site plan
UtilitiesCounty service area layers, then the utilityWhether water, sewer, and power reach the siteService area is not capacity; get a utility letter
CompetitionProperty appraiser use codes (2710 full-service and 2720 self-service car washes)Existing supply nearbyNew and under-construction sites, and miscoded ones, are missing

How does the workflow run, step by step?

  1. Write the program as gates and scores. Gates are pass or fail: the use is allowed, the size fits, no flood zone. Scores rank what passes: traffic, competition, demographics.
  2. Pick the markets and their codes. Each jurisdiction gets its own rule table. In Hillsborough's parcel file, 10,509 parcels carry CG or CI, but only 3,117 are unincorporated. Most of the rest are in Tampa, whose own code governs districts with the same letters; see the Tampa zoning guide.
  3. Run the zoning gate from the code text. List the districts where the use is permitted, conditional, or prohibited, with sections. This is the step AI zoning analysis automates.
  4. Filter by size, shape, and use code. Vacant land, obsolete buildings, or both, depending on the program.
  5. Confirm zoning on the official layer. Drop or flag split-zoned parcels and planned developments.
  6. Intersect the hazard and plan layers. Flood zones, future land use categories, overlays.
  7. Score. Traffic within a set distance, competitors within a radius, demographics for the trade area.
  8. Review and visit. A person clears the flags, then drives the top of the list.
  9. Hand off. Shortlist, owners, and notes go to the CRM for outreach.

How should sites be scored?

Keep gates and scores separate. A parcel that fails a gate should never be rescued by a high score, and a score should never hide why a parcel ranks where it does. An example rubric for an express car wash, for illustration only:

CriterionTypeExample rule
Use permitted in the districtGateP only; C goes to review
Lot sizeGate1 to 3 acres
FloodGateNo part in the SFHA
TrafficGate, then scoreAt least 25,000 vehicles a day within 40 m; more scores higher
CompetitionScoreFewer coded car washes within 2 miles scores higher
DemographicsScoreHouseholds and income in the trade area, from the ACS
Plan and zoning flagsDeductionSplit plan category, lower FAR cap, planned development

The thresholds are the example's inputs, not industry rules. Your operator's own data should set them.

Built for your team

Have Plotzy build this workflow on your criteria

Plotzy builds AI workflows that run on parcel, zoning and owner data: your buy box, your markets, your review steps, with every answer traced to its source. Start with a 30-minute call about the workflow you want automated.

  • Working prototype in the first 2 weeks
  • Answers cite the code section, record or dataset they came from
  • Outputs land in your CRM, spreadsheet or deal pipeline
Book a 30-minute call

30 minutes, no pitch deck

Worked example: express car wash sites in Hillsborough County

The buy box: unincorporated Hillsborough County, 1 to 3 acres, vacant commercial land, zoned where car washes are permitted, no part in the SFHA, and within 40 meters of a road segment carrying at least 25,000 vehicles a day.

The rules

Table 2.02.02 permits Car Wash Facilities in CG (Commercial, General) and CI (Commercial, Intensive). In CN it is a conditional use limited by Sec. 6.11.121 to self-service or automated washes in a roofed structure with at least two sides. In M it is accessory only. Section 6.01.01 then sets the box:

StandardCGCI
Minimum lot10,000 sq ft, 75 ft wide20,000 sq ft, 100 ft wide
Front yard30 ft30 ft
Maximum FAR0.27; 0.25 in eleven agricultural and residential plan categories (endnote 29)0.30
Maximum height50 ft, with extra setback above 20 ft (endnote 8)50 ft, same endnote
Building coverage27%30%
Impervious surface70%75%

The funnel

StepParcels leftWhat was removed
Unincorporated parcels coded CG or CI in the appraiser's file3,1177,392 inside Tampa, Temple Terrace, and Plant City
1 to 3 acres882Smaller and larger parcels
Coded vacant commercial (1000)59Built parcels
CG or CI on at least 95% of the parcel in the county atlas554 split with agricultural or residential districts
No part in the SFHA3520 in or partly in a mapped flood zone
Within 40 m of an FDOT segment with 2025 AADT of 25,000 or more1619 on lower-volume roads or roads FDOT does not count

One parcel the appraiser's file codes CG appears on the atlas as CI under rezoning 25-0931. Car washes are permitted in both, but the FAR, lot, and coverage limits differ, so the screen used the atlas.

The finalists

The 16 survivors were ranked by competition: parcels coded 2710 or 2720 within 2 miles, across city lines. Five of them show the range.

ParcelAcresDistrictPlan category2025 AADT within 40 mCar washes within 2 miFlagsNext step
Folio 0139860900, Flagship Dr, Lutz1.41CGNMU-450,000 (segment Debuel Rd to Sunset Ln)0None from the dataSite visit
Folio 0602600000, 11513 N 301 Hwy1.37CGSMU-6 on 57%, R-4 on 43%25,500 (Fowler Ave to Harney Rd)0Split plan category; endnote 29 applies in R-4Review, then visit
Folio 0613530000, 301 Hwy2.00CGUMU-20171,500 on a bridge segment; 28,500 on Harney Rd to Jefferson Rd0Top count may be a road the site cannot useReview
Folio 0411600000, 6525 E Hillsborough Ave2.98CGOC-2040,5001None from the dataSite visit
Folio 0292240000, 6205 N Dale Mabry Hwy1.09CGOC-2067,00011Saturated trade areaDeprioritize

The third row is the reason a person reviews every finalist. The highest count within 40 meters belongs to a bridge segment carrying 171,500 vehicles a day. Whether the site can take access from the road with the traffic is a question for the map, the driveway permit rules, and the drive.

A screen like this runs in minutes once the rules exist. The hard part is writing the rules from the code correctly, and keeping them current.

What does the site visit still have to settle?

The maps get a team to the curb. The visit decides:

  • Access and turning movements: median openings, signal spacing, and whether a driveway permit is realistic. Hillsborough sets access spacing in LDC Part 6.04.00.
  • Visibility and stacking: sight lines from the approach and room for a queue on site.
  • The real frontage: which road the parcel actually touches and at what grade.
  • Neighbors: residential next door changes the required buffers; see endnote 11 of Sec. 6.01.01.
  • What the layers miss: wetlands, drainage, easements, a recent clearing, a sign that says "Sold".
  • Utilities: confirm water and sewer at the curb, then get a capacity letter.

After the visit, the parcels that survive go to due diligence and the land due diligence checklist.

How does it scale across markets?

The method stays fixed; the rule tables change. Each new county or city needs its own use table, district standards, plan categories, and data sources, each with a date. The same letters can mean different districts across a city line, as CG does in Tampa and unincorporated Hillsborough. A multi-market build keeps one rule table per jurisdiction, re-reads the code when a new supplement posts, and re-runs the screen when the parcel or zoning layer changes.

How Plotzy builds this for your team

Plotzy runs zoning Q&A agents, owner contact research, and automated report generation in production today. A custom AI site selection build includes:

  • Your buy box and criteria, written as gates and scores, with one rule table per jurisdiction.
  • Runs across your pipeline, a whole county, or many markets, on a schedule.
  • Data pipelines from county assessor feeds, zoning layers, public datasets such as FDOT counts and the Census ACS, licensed sources, and your internal systems.
  • Output into your CRM or spreadsheet, one row per parcel with scores, flags, and owner contacts from public sources.
  • Human review points for planned developments, split zoning, access questions, and anything the model is unsure of.
  • Citations back to the code section on every zoning answer.

An engagement follows the timeline on the custom software page: a working prototype in the first 2 weeks, a proof of concept within 30 days, and production in 8 to 12 weeks.

Self-serve, a team can already search 150M+ parcels across 3,200+ US counties with a zoning category and designation on every parcel. Filters cover zoning, lot size, land use, and geography (polygon, radius, ZIP, county), with folders, notes, CSV export, and owner contact lookup. Zoning AI answers zoning questions from the ordinance text with the section cited, on the Production plan and up. Start with Site Selection.

If your team screens the same asset across several markets, book a 30-minute call below.

Questions to ask before you build or buy

  • Does the zoning gate come from the code text, with sections cited, or from a generic district label?
  • Is there one rule table per jurisdiction, and who updates it when the code changes?
  • Which date does each layer carry, and how often is it refreshed?
  • Are gates and scores separate, and can you see why each parcel ranks where it does?
  • How does it handle city lines, planned developments, and split-zoned parcels?
  • Does it find the owner, or stop at the parcel?
  • Where does the shortlist go, and who reviews the flags?
  • If you already license a trade-area tool, can the two share a parcel list?

Frequently asked questions

What is AI site selection?

It is software that screens parcels for a specific asset across one or many markets. It tests zoning, parcel size, traffic, demographics, flood, and utilities, then ranks the survivors for a person to visit.

How is AI site selection different from Placer.ai or Buxton?

Placer.ai and Buxton start from consumers, visits, and trade areas to judge how a location will perform. A land-first workflow starts from parcels and the zoning code and finds where the use is allowed and the site fits. The two fit together: a land screen first, then a trade-area tool on the finalists.

What data does site selection software use?

Parcels from the county property appraiser, zoning from the county or city layer and code, future land use from the comprehensive plan, traffic counts from the state DOT, demographics from the Census ACS, flood zones from FEMA, and utility service areas.

Can AI replace the site visit?

No. The screen cannot see median openings, sight lines, grade, wetlands, or what changed last month. It decides where to drive, not what to buy.

How long does an AI site selection screen take?

Once the rules exist, a county-wide screen runs in minutes. Writing and testing the rules for each jurisdiction is the real work, and it is repeated whenever the code or the program changes.

Does this work for data centers and industrial sites?

Yes, with different layers. Data centers turn on power, substations, water, and fiber, and often on an unlisted use. See data center site selection and the industrial zoning types guide.

Sources

The codes, data layers, and vendor sites on this page were checked on November 19, 2026, and the parcel counts were pulled that day. Codes and layers change often, so confirm the current version before relying on any detail. This page is general information, not legal advice.

Related reading