A broker who learns about a tenant move when the space shows up on LoopNet is already competing with every other broker in the market. The tenant's plans were visible months earlier, in a layoff notice, a funding filing or a lease end date, to anyone watching. AI tenant sourcing is software that watches those signals and tells a broker which companies to call first.
The question we hear from brokerages is how to get one: buy a platform, build an agent, or rent the data and build only the logic on top. The most specific version came from a small industrial brokerage in Hawaii. Its plan, drafted with an AI assistant, pulled data from the CoStar and Reonomy APIs and joined them with n8n or Make.com workflows. Hybrid is a sound shape for a plan like that. The CoStar half of it does not work, because CoStar sells no API a brokerage can plug into, and its terms forbid automated extraction.
TL;DR: Buying gets you property and ownership data fast, but no product we found combines tenant signals with lease data. A custom build from scratch is the slowest and most expensive path. Hybrid, renting data and owning the matching and scoring logic, fits a brokerage working one market. Plan it without CoStar as a data feed: Reonomy and public sources have APIs, CoStar does not.
What is AI tenant sourcing for CRE?
Tenant sourcing is finding companies likely to lease, expand, shrink or leave commercial space before they start looking. Brokers have always done it through cold calls, referrals and driving their territory.
AI tenant sourcing automates the watching. Software tracks public and paid signals, such as layoffs, funding rounds, lease expirations, new business registrations, mergers and bankruptcy filings, and flags the companies most likely to move. Tenants with larger spaces often start looking 12 to 24 months before a lease ends, so the useful window is months, long before a listing goes public.
One signal rarely means much alone. A funding round can go into product rather than office space. The "AI" part usually means the layer that matches records about the same company across sources and combines several weak signals into one ranked alert. We cover each signal, where it comes from and what it costs in the digital breadcrumbs that hint at a tenant move.
This works upstream of a CRM and a listing platform. A CRM tracks relationships a broker already has. A listing platform such as LoopNet shows space already on the market. Tenant sourcing surfaces the deal before either one would.
Buy: what you can purchase today, and where it stops
Two kinds of products come close.
CRE data platforms cover properties, owners and leases. Reonomy costs $4,800 a year, or $400 a month, per user, and offers an API and bulk data feeds, per CRE Daily's review. It covers more than 50 million US properties, traces LLC ownership back to the real owner, and includes a likely-to-sell score built on debt and sales history. That score is about owners selling buildings, not tenants moving. CoStar's average contract is about $15,130 a year, per Vendr's transaction data, and its strength is comps and market analytics. Lease end dates, the most direct tenant signal, come from Reonomy or CompStak.
B2B sales data tools such as Bombora, 6sense and ZoomInfo track company events, including funding and layoffs. They cost $25,000 to over $300,000 a year, per Autobound's 2026 comparison, and score companies for software sales teams. They do not tie a company to the building and lease a broker cares about.
We found no product that combines layoffs, funding, registrations, M&A and bankruptcy filings with lease data for CRE prospecting. Buying gets you good inputs. The matching and scoring across them is still missing.
Build: what a from-scratch AI system costs
The most detailed public estimate comes from Smart Capital Center, a company that sells a ready-made CRE AI platform, so read it as a vendor's case against building. Its 2026 analysis puts a production AI platform for a CRE firm at 18 to 24 months and a team of at least three engineers. At about $206,000 loaded cost per engineer, that is about $620,000 a year in people alone, and the vendor puts the total above $1 million a year once integrations, security and iteration are added.
Those numbers describe an enterprise-wide platform that runs on a firm's own property, accounting and deal data. A tenant-sourcing tool for one market is a smaller system, but it shares the hard part: data from many sources that does not line up. The same analysis cites MIT NANDA's 2025 finding that 95% of enterprise generative AI pilots show no measurable profit impact, and an S&P Global and Forrester figure that 42% of companies scrapped most of their AI initiatives in 2025, up from 17% in 2024. Both figures cover companies in every industry, not real estate alone.
The causes repeat across those failed builds:
- Data that does not line up. The same company appears under different names and addresses in every source.
- Maintenance after launch. Feeds change format, APIs change terms, and a scoring rule that worked last year drifts.
- No owner on the business side. A model nobody checks against real deals stays a demo.
Large firms hit the same wall. Josh Panknin of Columbia University told Commercial Observer that real estate companies "are not technology development companies," in a piece on AI efforts backed by Blackstone and Brookfield.
Why CoStar can't be a data source in your build
Plans for tenant-sourcing tools, including the Hawaii plan above, often list CoStar and Reonomy side by side as two data feeds. Only one of them is.
CoStar has no public or self-serve API for CoStar Suite, COMPS, Market Analytics or LoopNet, per Supergood's API report card, which grades it F. Supergood sells API access to hard-to-integrate platforms, so it has an interest in the gap, but the facts are checkable: access is through CoStar's portal, CoStar's terms of use prohibit automated extraction, and CoStar has sued companies that pulled its data programmatically under the Computer Fraud and Abuse Act. The integrations CoStar does offer for its Real Estate Manager product are built by its own professional services team, not opened to customers. Reonomy, by contrast, sells a documented API.
For the build, that means:
- Reonomy's API for ownership, debt and distress data, and Reonomy or CompStak for lease end dates.
- Public feeds for the rest: WARN notices for layoffs, SEC EDGAR for funding and M&A filings, state registries for new businesses, PACER for bankruptcy cases, county records for deeds and permits.
- CoStar as a manual check a broker runs on a flagged company, outside the automated pipeline.
An AI assistant can draft an architecture like the Hawaii plan in an afternoon. It rarely checks which vendor's terms forbid the use the plan depends on. That check costs an hour before a contract and a rebuild after one.
Hybrid: rent the data, own the logic
For a brokerage working one market, hybrid is the path we recommend: rent the data no brokerage can collect itself, and build the layer that decides who to call.
What you rent. Data runs from free to a few thousand dollars a year per seat. SEC EDGAR is free. WARN Firehose has a free tier and a $49-a-month API. OpenCorporates has a free tier. PACER charges $0.10 a page, capped at $3 a document. Reonomy is $4,800 a year per user. A one-market stack can start under $6,000 a year in data.
What you own. Record matching across sources, the link from each company to a building and lease, the scoring rules, and the record of which alerts turned into deals. This is the part that reflects your market, and it is the part a bought product cannot tune to it.
Why one market. Several sources are run state by state, from WARN thresholds to business registries, and lease data is thinner in markets with few reported comps. A system tuned for Oahu industrial space needs new sources and new weights before it works in Phoenix offices. Starting with one market keeps the first build small enough to test.
What it costs to build. Our estimate for the owned layer on one market, built by an AI-native team, is $25,000 to $40,000 and 6 to 8 weeks to a first version: one backend and data engineer full time, an architect part time, and a part-time PM and QA. The first alerts from three feeds can run in week 4. A team working without AI-augmented delivery would need roughly $40,000 to $60,000 and 10 to 12 weeks for the same scope. Running costs after launch are about $100 to $300 a month for hosting and language model calls, on top of the data.
Who keeps it running. Feeds change, and scores need re-checking against real moves every quarter. Plan for an owner on the brokerage side who reviews alerts weekly, and for 15 to 25 engineering hours a month of upkeep, about $700 to $1,200 a month at our rates, for feed changes, matching fixes and score updates.
What the owned layer has to do
The owned layer does three jobs, and each one is both a data problem and an engineering problem.
- Match records. "Acme Logistics LLC" on a WARN notice, "Acme Logistics, Inc." in a state registry and "Acme" in a funding filing are one company. Until the system knows that, a broker gets three weak alerts instead of one strong one.
- Score by direction and timing. A funding round points to growth, a layoff to shrinking. Event signals lose weight as they age. Lease expirations gain weight as the date gets closer.
- Test against real moves. Take tenants in your market that moved in the last two or three years, and check how many the score would have flagged, and how early.
A hypothetical example: a 120-person logistics company files a WARN notice for 60 layoffs at its Kapolei warehouse, and its lease there ends in 14 months. Two signals point the same way, shrinking, inside one window, so the score ranks it high and the broker calls about subleasing or downsizing. If the same company had also raised a round last month, the signals would point in opposite directions, and the alert would say read the news before calling.
The full method, with sources, costs per signal, the open-source matching library and a version you can run on a spreadsheet, is in our guide to the signals that hint at a tenant move.
We in Brocoders have built this kind of multi-source work in real estate, though not a tenant signal score yet. For AreaButler, our engineers joined the client's team and built features that combine CRM data, a real estate data platform and OpenAI into one property analysis for brokers. For Lake, we rebuilt a vacation rental platform's backend in 3 months, and Lake then grew from 500 to 40,000 connected properties through its property management system integrations. Both sit within our proptech and real estate work.
Where to start without overcommitting
Start by hand. Track the six signals for your market in a spreadsheet for a few months and record what each flagged company actually did. The starter framework in our signals guide lays out the steps. If the signals predict moves you would have wanted to catch, the build is worth pricing.
Before you sign a data contract, check each source in your plan for three things: does it sell an API, do its terms allow the use you plan, and does it cover your market's tenant sizes. The CoStar gap above is the one we see most often.
If you want a second opinion on the plan, we in Brocoders run a 2-day paid feasibility review with an architect and an AI engineer. In those two days we check every source in your plan for API access, terms of use and coverage of your market, draw the architecture, and price the build. You get a verified source list, an architecture diagram, a build estimate and a 60-minute walkthrough. Two days cover a desk check. If you also want a sample pull from each feed and a test of how many records match, plan for 5 business days. The same review applies to any proptech or real estate workflow built on scattered public data.
Sources: Commercial Observer, 3 Aug 2026; Smart Capital Center, 28 Aug 2026; CRE Daily, Reonomy review; Vendr, CoStar pricing; Supergood, API report card: CoStar; Autobound, 2026 intent data provider comparison; WARN Firehose; PACER pricing.
Frequently Asked Questions (FAQ)
It means using software to watch signals such as layoffs, funding rounds, lease expirations, new business registrations, mergers and bankruptcy filings, then rank which companies are most likely to move before a listing goes public. It works upstream of a CRM or a listing platform.
No public or self-serve one. CoStar Suite, COMPS, Market Analytics and LoopNet are available through the portal only, its terms of use prohibit automated extraction, and CoStar has sued companies that pulled its data programmatically. Integrations for its Real Estate Manager product are built by CoStar's own professional services team.
Yes. Reonomy sells a documented API and bulk data feeds, with standard pricing of $4,800 a year per user. CoStar's average contract of about $15,130 a year buys portal access.
Layoffs (often visible through WARN Act notices), funding rounds, upcoming lease expirations, new business registrations, mergers and acquisitions, and bankruptcy filings. No single signal is reliable alone. The value comes from matching them to one company and seeing two or more point the same way. See the digital breadcrumbs that hint at a tenant move.
Data alone costs from free to $4,800 a year per user for Reonomy, and $25,000 to over $300,000 a year for B2B sales data tools, but none of it matches and scores signals. A vendor estimate for a from-scratch enterprise AI platform runs 18 to 24 months and over $1 million a year. A hybrid build for one market rents data for under $6,000 a year and builds only the matching and scoring layer. Our estimate for that layer is $25,000 to $40,000 with an AI-native team, and $40,000 to $60,000 without AI-augmented delivery.
Reonomy's API for ownership and distress data, Reonomy or CompStak for lease end dates, public feeds for filings and registrations, a workflow tool such as n8n or Make.com to run the pulls, and an owned layer that matches records, scores signals and sends a ranked alert. CoStar stays a manual check.
A prototype on a few feeds can come together in weeks: in our estimate, first alerts in about 4 weeks and a first version in 6 to 8. A production system needs record matching that holds up on real data, monitoring for feed changes, and a score tested against two or three years of real moves. We estimate 3 to 4 months and $40,000 to $65,000 in total for one market with an AI-native team.