Sell Homes With an AI Agent: How Much Can a Lean US Brokerage Make?

The Real Brokerage closed 185,314 transactions and reported about $2 billion in revenue in 2025, but its gross profit was $165.7 million, which works out to roughly $894 per transaction side (a calculation from the reported figures). That is the core lesson for anyone who wants to sell homes with an AI agent: automation cuts your costs, but a brokerage still earns thin margins per deal and needs a licensed human on every file. This post walks through example math, funded case studies and what a $5k to $50k founder can realistically do.

What is the AI real estate brokerage opportunity?

The opportunity is that AI now does much of the clerical work of a home sale, while commissions still price that work as if it were scarce. An Alloy Advisors report, covered by Real Estate News on 11 June 2026, found that AI beats humans on 10 of 23 agent tasks in a home sale, mostly clerical ones such as MLS listing entry, listing descriptions and showing scheduling.

The same report found agents clearly outperform AI on only three tasks: negotiation execution, hyperlocal tacit knowledge, and emotional support and crisis management. It estimated a typical $400,000 home costs $39,660 to sell and that $17,000 to $22,000 of that is "overpriced." The authors say these are bottom-up estimates from freelance and professional market rates, and that no controlled studies compare AI-assisted and traditional sales.

Who pays today? Sellers and buyers. Clever Real Estate's survey, updated 1 October 2026, puts the average commission at 5.46%, split 2.76% to the listing agent and 2.70% to the buyer's agent. Treat it as a survey estimate, not a fixed rate.

The demand side is uncertain. The Alloy report cites a December 2025 YouGov survey in which 65% of people trust AI to compare prices but only 14% trust it to act on their behalf. A lean brokerage therefore sells AI-assisted service with a licensed human, not a robot.

How much can a lean AI brokerage make?

An example estimate: a one-person brokerage closing two deals a month would show about $1,500 a month in profit on paper, while a 27-agent operation closing 20 deals a month would show about $20,000. These are not forecasts or earnings claims. They are arithmetic on stated assumptions, and your results could be zero or negative.

Assumptions (all example estimates, not sourced averages):

• Home price: $400,000 (the figure used in the Alloy example).
• Brokerage fee kept per deal side: 1%, or $4,000. This mirrors the fee Homa says it keeps on the buyer side (see below). Your fee may differ.
• Payout to the licensed agent or specialist doing tours and negotiation: 50% of the fee.
• Fixed monthly costs (insurance, association and MLS dues, software, ads, a coordinator in the larger cases): $2,500, $6,000 and $20,000. These are placeholders, not quotes.
• Profit shown is before owner pay, taxes and any one-time startup costs.

Check the headcount. The National Association of REALTORS (NAR) says the typical member completed nine transaction sides in 2025. At that pace, the large case (240 sides a year) needs roughly 27 typical agents, and the medium case (72 sides) needs about eight. Recruiting that many is the hard part.

Now stress the numbers. If your gross profit per side were $894, like Real's company-wide figure, instead of $2,000, the medium case brings in about $5,364 a month against $6,000 of fixed costs, a monthly loss of about $636. Real runs a different model at far larger scale, so the comparison is rough. The point is that small changes in per-deal margin swing the result.

The small case also falls short of a normal salary. NAR reports a median gross income of $59,200 in 2025 for REALTORS, and $8,000 for those with two years or less of experience.

Which companies are already building AI brokerages?

Several startups are attacking different parts of the brokerage. Only some actually sell homes; others sell software to brokerages. Here is what each says, with funding where public.

Homa. According to its Business Wire release, carried by VentureBeat and Pulse 2, the company says that when a seller offers a buyer-broker commission of up to 3%, it keeps 1% and credits the remainder, typically about 2%, back to the buyer. The company says credits have averaged more than $10,000. It says it uses one in-house broker per state to write and negotiate offers, with local specialists handling tours for a flat fee. All of this is the company's own description. Note that Homa is buyer-side only, so it does not list your home.

Spotlight Realty. Per its Y Combinator launch post, Spotlight says it charges only 4% commission to lease rentals, and that traditional rental commissions have been north of 12%. Its Y Combinator profile describes a full-service sell-side residential brokerage that screens and schedules showings with an AI agent. This is a rentals business in New York City, so it is only a partial model for a home-sale brokerage.

MaxHome. In its December 2024 release, the company said it automates over 70% of tasks in managing real estate transactions. That is a company claim, not an independent measurement.

RealPact. Its Y Combinator page says it builds AI agents that find property records and fill out contracts, organize documents and track deadlines. It lists two employees and says it currently sells to brokerages in New Hampshire.

The Real Brokerage. HousingWire's coverage of its 2025 results reports a net loss of $8.1 million, adjusted EBITDA of $62.9 million and 31,739 agents at year end. A brokerage of this scale still reported a net loss.

What do funded teams do with millions, and what can you do with $5k to $50k?

Homa says its seed round funds expansion across California and in-platform mortgage lending, and it already uses a licensed in-house broker in each state. You probably cannot match that with $5k to $50k. What you can do: pick one city, one price band and one narrow service, such as AI-prepared listing paperwork plus a flat fee. Then test whether sellers will pick you over a 2.76% listing fee. Prove demand before building any software.

Do you need a licence to run a brokerage?

Yes. You need a broker-level licence, and the rules are set state by state. Washington's Department of Licensing is one example: to upgrade to a managing broker you need at least three years of full-time experience as an active broker, gained within the last five years, plus 90 hours of approved education (30 each in brokerage management, business management and advanced real estate law). The state also offers alternate-qualification and reciprocity routes.

Other states differ in experience, hours and exams. Start at your own state's real estate commission website, not a third-party guide. Also confirm any state rules on advertising fees and passing commission credits to consumers.

The licence requirement is also why AI cannot replace the broker. The Alloy report says human presence is required by law for two home-sale tasks, including licensed fiduciary accountability.

How much does it cost to start a brokerage?

McKissock Learning says starting a brokerage can cost anywhere from $30,000 to over $70,000, depending on scale. It does not publish a line-by-line price list, so treat the range as a rough guide as of 15 October 2026.

If you have $5k to $50k, the lower end of that range suggests you should start as a licensed agent under an existing broker, or run virtual, before opening your own brokerage. That is a judgment, not a sourced rule.

How do you start an AI-assisted brokerage, step by step?

Start with the licence, then prove one service before you build anything. These steps are a suggested order.

1. Check your state commission's broker requirements and calculate how long the experience and education will take.
2. Get licensed as an agent, if you are not already, and work under a broker to learn the deal flow.
3. Choose one city, one price band and one service (for example, listing a home for a flat fee).
4. Write down your fee, the agent split and your break-even deals per month using the table above.
5. Price E&O insurance, association and MLS dues, and software with real quotes.
6. Map the 23-task split from the Alloy report onto your workflow: put AI on clerical tasks, humans on negotiation and judgment.
7. Pilot with a few sellers before you hire anyone, and record time per deal.
8. Open the brokerage entity once the numbers work, and register it with your state.
9. Recruit agents who already close deals. Remember the NAR typical figure of nine sides a year.
10. Review each month: fee kept per side, hours per deal and complaints.

Which free tools and resources help?

Public data and calculators cover the research phase. Pricing for brokerage software is generally not published, so get quotes.

Where does AI fit in a brokerage?

AI fits best in transaction paperwork, listing copy and scheduling. The Alloy report puts MLS listing entry, listing descriptions and showing scheduling among the tasks where AI beats humans. MaxHome and RealPact both sell tools aimed at contract-to-closing work, though neither lists pricing on the pages cited above.

The same logic applies to lead response and follow-up. For lead handling, see our look at reselling AI receptionists to contractors and the lead follow-up agency model. For regulated-intermediary economics, buying an insurance agency with AI is a close cousin. If you want a related service, the landlord bill-pay automation business serves a similar audience of property owners.

Do not let AI sign, advise on price as if it were a licensee, or negotiate without a licensed human in the loop.

What are the risks, and what's still unclear?

The biggest risk is thin per-deal profit. Real's $2 billion in revenue produced an $8.1 million net loss in 2025, per HousingWire. New agents also struggle: NAR reports a median gross income of $8,000 for agents with two years or less of experience.

Other risks:

• Trust. Only 14% of people trust AI to act on their behalf, per the YouGov survey cited by Alloy.
• Legal. A licensed human must carry fiduciary duty. An AI-written disclosure or contract that is wrong is still your liability.
• Evidence. The Alloy report is built on estimates, not controlled studies.
• Self-reported claims. Homa's credits, MaxHome's 70% automation and Spotlight's fee comparisons come from the companies themselves.
• Funding gaps. Spotlight Realty's and RealPact's funding amounts are not stated in their Y Combinator profiles, so their runway is unknown.

What is still unclear:

• Whether lower fees win enough volume to offset the thinner margin per deal.
• Whether AI-assisted sales close at different prices than traditional ones (no controlled studies exist yet).
• Whether the 5.46% average falls. The Alloy authors predict the standard breaks within five to seven years, which is a forecast.
• How Homa performs at scale. It expanded to Texas in June 2026 and soft-launched in California on 1 September 2026, per its release, so results are early.

This is research, not legal, tax or financial advice. Dollar figures above are example estimates with assumptions shown, not income forecasts or earnings claims.