AI workflow

Build vs Buy AI for Commercial Real Estate: A Sourced Cost Model

What it costs to build AI in-house at a CRE firm, from published wages, parcel data prices, model API rates and hosting, and when buying software or partnering with a specialist builder wins instead.

By Nathan Robinson, Founder, Plotzy Published 13 min read

Build vs buy AI for commercial real estate comes down to whether the workflow is standard or yours. Buy software when the job is standard, such as finding parcels by zoning or looking up owners. Build in-house when AI is core to how you win deals and you can staff a team. The cost model below puts that at about $0.5 million to $0.9 million in year one, mostly salaries.

The third path sits in between. Partner with a specialist builder when the workflow is specific to your firm but must be in production within a quarter.

Every figure below comes from a published source checked on November 19, 2026, unless it is marked as an estimate. Prices change, so confirm current rates before you budget.

What does it cost to build AI in-house at a CRE firm?

The AI implementation cost for a CRE firm that builds in-house is about $500,000 to $920,000 in year one, under the assumptions below. People are 80 to 85 percent of it. Data, model usage and hosting are the rest.

Line itemLow caseHigh caseBasis
Team, loaded with benefits$417,914$741,543BLS May 2025 mean wages, plus benefits at the private industry average
Parcel data (50,000 records a month)$62,100$92,400Regrid self-serve API list prices; volume this size is quoted as enterprise
Model API (50,000 zoning checks a month)$12,000$48,000Anthropic list prices, Sonnet 5.5 with batch discount to Opus 5.5 standard
Hosting$6,000$36,000Estimate, anchored to DigitalOcean list prices
Year one totalabout $498,000about $918,000

Not included: recruiting fees, equipment, software tools, legal review of data licenses, and the time your acquisitions team spends defining rules and reviewing output.

People: what an in-house AI team costs

The salary line is the biggest one, so it deserves the most care. The Bureau of Labor Statistics (BLS) Occupational Employment and Wage Statistics survey for May 2025 gives these annual wages.

Role (BLS occupation)National meanNational median75th percentileFlorida meanTampa metro mean
Software developer (15-1252)$148,100$135,980$171,980$137,010$133,630
Data scientist (15-2051)$126,800$120,230$158,880$119,150$116,600

BLS has no separate data engineer occupation. The closest, database architect (15-1243), has a national mean of $144,440. Computer and information systems managers (11-3021) average $192,160.

Wages are not the whole cost. BLS's Employer Costs for Employee Compensation for June 2026 puts wages at 70.0 percent of private industry compensation and benefits at 30.0 percent. Dividing wages by 0.70 gives a loaded cost.

  • Lean team: one software developer and one data engineer. Wages $292,540; loaded about $417,914 a year.
  • Fuller team: two software developers, one data scientist and half of a manager's time. Wages $519,080; loaded about $741,543 a year.

National means are averages across all experience levels. Engineers who have shipped AI systems in production may sit nearer the 75th percentile, which would push both cases up.

Parcel and zoning data

Parcel data has published list prices at small volume and quoted prices at scale. Zoning data at scale is mostly quote-based.

  • Regrid self-serve API, viewed November 19, 2026: Standard is $375 a month with 2,000 parcel records, then $0.10 per record. Premium is $500 a month with 2,000 records, then $0.15 per record, and adds land use classification and a zoning type and sub-type field. Regrid's full Standardized Zoning product is not listed on either self-serve plan.
  • At 50,000 records a month, those list prices imply $5,175 (Standard) to $7,700 (Premium) a month. Regrid's page directs anyone needing more than 10,000 records a month to enterprise licensing, which is priced by quote.
  • County files are another route. Regrid's Data Store lists Hillsborough County, Florida at $300 (Standard) or $500 (Premium), plus $650 for its Standardized Zoning add-on.
  • Ordinance text is free to read but costly to collect. Each jurisdiction's code has to be gathered, parsed and re-checked after amendments. Scripted requests to Municode's code library returned a bot check in November 2026.
  • Owner contact data comes from vendors with their own terms. Budget it as a separate line.

Read the license before you build on any dataset. Storage, model use and redistribution rights vary by vendor and by contract.

Model API costs for 50,000 parcel zoning checks a month

Model usage is the smallest line for most CRE workflows. Assume each check sends about 15,000 input tokens (the district's code sections plus the parcel record) and returns about 1,000 output tokens. That is 750 million input tokens and 50 million output tokens a month.

Model (list price per million tokens, input / output)Cost per checkMonthly, standardMonthly, batch (50% off)
Claude Haiku 5.5 ($0.10 / $0.50)$0.002$100$50
GPT-6 Luna ($0.10 / $0.50)$0.002$100$50
Claude Sonnet 5.5 ($2 / $10)$0.04$2,000$1,000
GPT-6.2 Sol ($2 / $10)$0.04$2,000$1,000
Claude Opus 5.5 ($4 / $20)$0.08$4,000$2,000
Claude Fable 5.2 ($8 / $40)$0.16$8,000$4,000
GPT-6.2 Astra ($8 / $40)$0.16$8,000$4,000

Three things move this number:

  1. Caching cuts it. Checks on parcels in the same district share the same code text. Say 10,000 of the 15,000 input tokens are a cached prefix. Sonnet 5.5 or GPT-6.2 Sol then drops from about $2,000 to about $1,050 a month, before cache-write charges.
  2. Reasoning raises it. Thinking or reasoning tokens bill as output. An extra 3,000 per check adds $1,500 a month on Sonnet 5.5.
  3. Agent loops multiply it. An agent that calls the model five times per parcel re-sends context each time. Expect input cost to rise several times over unless the context is cached.

Prompt length matters on OpenAI's API. Above 272,000 input tokens, GPT-6.2 input prices double and output rises 1.5 times. Anthropic bills its current models' 1M-token window at standard rates, except Haiku 5.5 above 100,000 tokens. Keep zoning checks short by retrieving only the relevant sections.

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

Hosting

Hosting for one production workflow is modest. Plan on about $500 to $3,000 a month (an estimate). At DigitalOcean's list prices, two 16 GiB general purpose servers cost $126 a month each. A 16 GiB managed PostgreSQL node costs $244.35 a month. That is about $500 before a standby database, staging, backups, logging and monitoring, which push toward the top of the range.

Time to first production workflow

PathTime to first workflowBasis
Buy softwareSame dayA self-serve trial; Plotzy's needs no credit card
Build in-houseAbout 4 to 9 monthsEstimate: 2 to 4 months to hire, then 2 to 5 months to build, test and launch
Partner with a specialist8 to 12 weeks to productionPlotzy's stated timeline, with a prototype in 2 weeks

The in-house range is a planning estimate, not a published figure. The biggest swing is hiring.

Worked example: screening Hillsborough County's vacant commercial and industrial land

A land team wants every vacant commercial or industrial parcel in Hillsborough County, Florida checked against its buy box each month. The Hillsborough County Property Appraiser's daily parcel layer, last edited November 18, 2026, sizes the job.

ItemFigureSource
Parcels in the county531,626Property Appraiser daily parcel layer
Vacant commercial (use code 1000) and vacant industrial (4000)2,808 parcels, about 5,456 acresSame layer
Distinct zoning codes among them83Same layer
Parcels with a PD-type code742Same layer
One county file with zoning$1,150 ($500 Premium plus $650 zoning add-on)Regrid Data Store
Source date of that file's zoningApril 17, 2026Regrid Data Store
Model cost for one full screenAbout $112 on Sonnet 5.5, $56 with batch; about $6 on Haiku 5.5Anthropic list prices, assumptions above

What the numbers say:

  • The model is cheap; the rules are not. One full screen costs about $112 in tokens. The real work is encoding four jurisdictions' codes, 83 district labels and 742 planned development (PD) parcels. Each PD has its own ordinance and site plan.
  • Freshness varies by source. The county layer was edited the day before this check. The vendor file's zoning dates from April 2026. Check the source date on every zoning field.
  • The zoning field is not the map. The Property Appraiser's zoning field is a record. Confirm finalists on the official Tampa or Hillsborough County zoning map.
  • Scale is where paths split. Fifty thousand checks a month equals this screen run for about 18 buy boxes, or one buy box across many counties.

Buy, build or partner: how do the three paths compare?

Each path trades cost against fit and speed. The table compares them on the points that usually decide it.

Buy software (SaaS)Build in-housePartner with a specialist builder
Upfront costNone to lowHiring plus months of salary before launchA scoped build, priced per engagement
Ongoing costPer seat or per lookupTeam, data, model and hostingA run or support fee, or your own team takes over
Time to first workflowSame dayAbout 4 to 9 months (estimate)8 to 12 weeks
Fits your buy box and dataOnly as far as the product's filters goFullyFully, for the scoped workflow
Who maintains data and ordinancesThe vendorYouAgreed in the contract
Main riskThe product cannot model your workflowHiring, attrition, and a build that stallsDependence on the partner if you do not own the code

The honest case for buying software

Software wins when your workflow looks like everyone else's. Finding parcels by zoning, lot size and land use, then looking up owners, is a solved problem. A vendor spreads data and maintenance costs across many customers, so you pay a fraction of the in-house cost. Plotzy's self-serve plans run $85 (Scout), $180 (Starter) and $300 (Production) a month. The comparison pages list other vendors' prices with dates. The limit is fit: if your edge is a proprietary buy box or your own data, a product built for everyone will not encode it.

The honest case for building in-house

Building wins when AI is part of how you compete, not a single tool. A team you employ keeps the knowledge, iterates every week, and connects to every internal system. Firms with several workflows, proprietary data at volume and a long horizon get the most from it. The costs are real: two to four hires, months before the first result, and a permanent line for data and ordinance upkeep. If the work stalls when one engineer leaves, the build was not really in-house.

The honest case for partnering with a specialist builder

A partner wins when the workflow is specific to you but you cannot wait for a team. A builder that already runs parcel, zoning and owner pipelines skips the learning curve. You get production software in a quarter and decide later whether to hire. The risks are dependence and fit. Insist on owning the code and data, on documentation, and on a handover plan. A partner without CRE data experience will rebuild the same pipelines on your budget.

When does each path win?

Your situationPath that usually winsWhy
A small team needs parcel search, zoning filters and owner lookupsBuyStandard workflow; the vendor carries the data cost
You need one high-value workflow (a county-wide buy-box screen, automated diligence memos) in production this quarterPartnerSpecific to you, faster than hiring
AI and data are central to strategy, with several workflows planned and budget for a standing teamBuildOwnership and iteration speed pay back over years
You are testing whether AI helps at allBuy, or a partner prototypeLearn before committing salaries
Your data cannot leave your environmentBuild, or a partner build deployed in your cloudControl over where data lives
You already have engineers but no CRE data experiencePartner for the first workflow, then buildTransfer the pipeline knowledge, then own it

How to decide: five questions

The same test works for most build vs buy software decisions in real estate.

  1. Is the workflow standard? If a product already does it, start there.
  2. How many parcels a month? Tens a week suits software. Thousands a month suits a pipeline.
  3. Is it your edge? If the buy box is proprietary, plan to own the logic, whoever builds it.
  4. Can you staff it? If you cannot hire two or more engineers, building in-house is not yet an option.
  5. How fast? If it must work this quarter, buy or partner.

How Plotzy builds this for your team

Plotzy is the partner path, built on its own parcel, zoning and owner data stack. Plotzy runs zoning Q&A agents, owner contact research and automated report generation in production today.

What a build includes:

  • Your buy box and criteria, written as rules a model applies and a person can audit.
  • Runs across your pipeline or a whole county, such as the Hillsborough screen above.
  • Data pipelines from county assessor feeds, your internal systems and third-party APIs.
  • Outputs into your CRM or spreadsheet.
  • Human review points for PD parcels, split zoning and anything a deal depends on.
  • Citations back to the code section on every answer.

How an engagement runs: a working prototype in the first 2 weeks and a proof of concept within 30 days. Production follows in 8 to 12 weeks, depending on complexity. Engagements are scoped on a call, so pricing depends on the workflow. See custom AI software for the full offer.

What you can do today, self-serve: search 150M+ parcels across 3,200+ US counties in Site Selection, with a zoning category and designation on every parcel. Filter by zoning, lot size, land use and geography, and export to CSV. Zoning AI (Production plan and up) answers zoning questions from the ordinance with the section cited, in the Florida jurisdictions Plotzy has loaded. Plans and the free trial are on the pricing page.

If you are weighing build against buy for a specific workflow, book a 30-minute call below.

Questions to ask before you build or buy

  • What exact decision will the output support, and who signs off on it?
  • How many parcels or documents a month, and how often does each need a re-check?
  • Which data licenses cover model use and storage of results?
  • What is the source date of every zoning and ownership field?
  • Who updates the rules when an ordinance is amended?
  • Does every answer cite a code section and a record?
  • If you partner, do you own the code, prompts and data at the end?
  • What does one full run cost in tokens, and who watches the bill?

Frequently asked questions

How much does it cost to build an AI tool for commercial real estate?

Under this page's assumptions, an in-house build costs about $500,000 to $920,000 in year one. Salaries are 80 to 85 percent of that. Model API costs for 50,000 zoning checks a month run $50 to $8,000, depending on the model.

What does an in-house AI team cost?

At BLS May 2025 national means, a software developer earns $148,100 and a database architect (the closest occupation to data engineer) $144,440. With benefits at 30 percent of compensation, that two-person team costs about $418,000 a year.

Is the LLM API the biggest cost of an AI workflow?

Rarely. At 50,000 zoning checks a month, Claude Sonnet 5.5 costs about $2,000 at list price, or $1,000 with batch processing. People and data cost far more.

How long does it take to get an AI workflow into production?

Software is usable the same day. Plotzy's custom builds reach production in 8 to 12 weeks, with a prototype in 2 weeks. An in-house build takes about 4 to 9 months by the estimate on this page, mostly because of hiring.

Can a team start with software and build later?

Yes, and many teams should. Start with a product for standard work, write down the rules it cannot handle, and build or partner for those. The written rules become the spec.

Sources

Related reading