Compass: a brokerage built on its own agent software

Context

Compass is a US residential brokerage that spent a decade arguing it was a technology company. The claim was simple: build the software an agent runs their whole business on, give it to agents who join, and the best agents will come and stay.

For years the numbers did not back the claim. In 2022, Compass lost $601.5m on $6.0bn of revenue, and adjusted EBITDA was a $210.0m loss. In the same year it spent $360.3m on research and development. The company cut about $338m of annualised non-GAAP operating expense between Q2 and Q4 2022.

By 2025 the picture had changed. Revenue reached $6.96bn, the net loss narrowed to $58.5m, and free cash flow was $203.3m. In January 2026 Compass closed an all-stock acquisition of Anywhere Real Estate, owner of Coldwell Banker, Century 21, Sotheby's International Realty, Corcoran and ERA. In Q2 2026 the combined company reported $92m of GAAP net income.

It also spent 2025 and 2026 in a public fight with Zillow over whether agents can market homes privately before putting them on the open market.

For an Indian channel partner or developer sales team, Compass is a useful case because its core bet is one you can make at any size: the software your people use every day is part of what keeps them.

What they built

Compass built a single platform for agents, covering client records, listing marketing, deal tracking and a client-facing view. It then layered products on top that give agents something to show sellers.

The agent workspace. Compass reported a record 20 average weekly sessions per agent in Q4 2025, roughly four a working day. That is the metric that matters for any CRM: whether people open it without being told to.

Compass One. A client dashboard, launched in early 2025, that shows buyers and sellers where their transaction stands at any hour. Compass says about 370,000 clients used it in 2025. It appeared in 17.4% of closed transactions in Q1 2025, 28.4% in Q4 2025 and 35% in Q2 2026.

Seller and buyer demand tools. "Make-Me-Sell" lets homeowners record the price at which they would sell a home that is not listed. It held more than 26,000 entries at the end of Q2 2026. A reverse-prospecting tool, which matches listings to buyers already in agents' books, was used by 11,400 agents in Q2 2026.

Integrated title and escrow. Compass says agents who use its one-click title and escrow integration attach those services at about twice the rate of those who do not. In Q2 2026 its title and escrow business closed 42,608 transactions.

AI assistant. By Q2 2026, about 15,000 agents had generated more than 97,000 conversations with the company's AI assistant.

A three-phase listing strategy. Compass encourages sellers to launch as a Compass Private Exclusive, move to Compass Coming Soon, then list publicly on the MLS. This is where the Zillow fight started.

The numbers

All figures are from Compass's own results releases unless noted.

The loss years:

• FY2021: revenue $6.42bn, net loss $494.1m, adjusted EBITDA $1.6m
• FY2022: revenue $6.0bn, net loss $601.5m, adjusted EBITDA loss $210.0m, R&D $360.3m

The turn:

• FY2024: revenue $5.63bn, net loss $154.4m, adjusted EBITDA $126.0m, free cash flow $105.8m. Compass says it was free cash flow positive in every quarter of 2024.
• FY2025: revenue $6.96bn (up 23.7%), net loss $58.5m, adjusted EBITDA $293.4m (4.2% margin, up from 2.2%), free cash flow $203.3m
• Principal agents at end of 2025: 21,190, up 19.4% year on year, with quarterly retention of 96.8% in Q4 2025
• Management said non-GAAP operating expense grew about 1% in 2025

After Anywhere (Q2 2026):

• Revenue $4.31bn, up 109% reported and 14.3% on a pro forma basis
• GAAP net income $92m (Q2 2025: $39m); adjusted EBITDA $363m
• 83,184 agents in the brokerage segment; quarterly retention 95.5%
• Long-term debt $3.14bn; cash $694m
• Year-one net cost synergies of $300m actioned five months early, target raised to $330m; three-year target $500m

At announcement in September 2025, Compass put the combined enterprise value at about $10bn including debt and projected more than $225m of net cost synergies. It has since raised that goal to $500m over three years.

One caution on reading these figures. Compass's revenue is mostly gross commission income, most of which it pays out to agents. A 4.2% adjusted EBITDA margin is the reality of brokerage economics, not a software margin. Compare that with Rightmove's 70%.

What worked, what broke

What worked: the platform as a retention tool. Quarterly principal agent retention stayed near 97% through 2024 and 2025. In Q2 2026, excluding agents who earned nothing that quarter, retention was 97.7%. Agents who run their pipeline, marketing and client updates in one system face real switching costs when they leave.

What worked: client-facing software. Compass One went from 17.4% to 35% of closed transactions in about 15 months. Giving the buyer or seller a live view of their deal is a feature agents can pitch when they compete for a listing.

What worked: cost discipline after the spend. The platform was built during the loss years. Profit arrived only when operating expense stopped growing while agent count and revenue kept rising. The software did not change. The cost base did.

What broke: the price of building it. Compass lost over $1bn across 2021 and 2022 alone, per the figures above. Few companies can fund a platform that way. The model worked for Compass partly because it could raise that capital.

What broke: private listings versus portals. In April 2025, Zillow announced listing access standards that bar listings marketed publicly without being put on the MLS within a day. Compass sued in June 2025, arguing Zillow held monopoly power. On 6 February 2026, a federal judge denied Compass's request for a preliminary injunction. On 17 March 2026, Zillow launched a pre-market product called Preview, and Compass dropped the suit the next day. The lesson: if your model depends on withholding inventory from the dominant portal, expect a long fight.

Still open: integration. Compass now runs a franchise business with far more agents than its own software was built for. It has started rolling its platform out to agents at non-Compass brands: over 4,000 by Q2 2026, with nearly 50,000 more expected. Whether agents in those brands adopt it the way Compass's own agents did is the test for the next two years.

How to apply this at your scale

The Compass lesson maps to one thing: a CRM that your sales team and channel partners actually use every day. Not a lead dump. A working tool.

Most Indian CP firms and developer sales teams run on a mix of portal inboxes, spreadsheets and WhatsApp. Leads arrive from 99acres, MagicBricks, Housing.com, Meta forms and walk-ins. No one can say which source closes, which executive followed up, or where a buyer stands on site visit, token or agreement. When a sales executive leaves, their buyers often leave with them.

What to take from Compass:

1. Measure use, not features. Compass tracks weekly sessions per agent. Track daily logins, follow-ups logged and site visits recorded. If your team is not opening the CRM every day, the build is wrong, not the team.
2. Give the buyer a view. Compass One is a client dashboard. For a developer, the Indian equivalent is a buyer page showing unit details, payment schedule, receipts and construction updates. A buyer who can check these facts alone calls your CRM team less often.
3. Tie every lead to a source and an outcome. Tag each enquiry with portal, campaign or CP at entry, and record site visit, EOI and booking against it. Within a quarter you will know which portal subscriptions pay back.
4. Keep pre-launch marketing inside RERA. Compass's private exclusives are a pre-market phase. In India, Section 3 of the RERA Act bars promoters from advertising, marketing, booking or offering units in a project before it is registered. Regulators read "advertisement" broadly, to include SMS, email and social media. A CRM can collect interest, but campaigns and bookings should wait for the registration number.
5. Build small, then grow. Compass spent hundreds of millions of dollars a year on R&D. You do not need to. Start with lead capture, assignment, follow-up reminders and a pipeline view for one project, and add modules only when the team asks for them.

The part of Compass worth copying is not its scale. It is that its agents open the software four times a day.