How Anarock turned a broker network into a data business

Context

Most primary housing in India is sold through brokers. A developer launches a tower, hands inventory to a panel of channel partners, and pays each one a brokerage fee when their client books. Most of those partners are small firms: a principal, a few sales staff, a phone book and a WhatsApp list.

Anuj Puri founded Anarock in April 2017 to run that business at scale. The pitch to developers was simple: give us an exclusive mandate, and we bring our own sales team, a large channel partner network and data on who actually buys. By 2023, the company's standard press release said it managed "over 80,000 established channel partners." By its November 2025 AI launch, it said it had data from 90,000+ units sold across 950+ projects.

Anarock is private, so there are no quarterly filings. What we have are the chairman's annual revenue interviews, company press releases, third-party summaries of statutory filings, and Anarock's own product pages. This case uses those sources and says where they disagree.

What they built

An exclusive-mandate sales business. This is the core. Anarock takes a developer's project, runs the sales team and marketing, and pulls in channel partners to widen reach. Business India reported in 2022 that it had delivered over 450 exclusive mandates and facilitated sales of over 50,000 homes. It also set up a separate unit, Trespect, for open-market brokerage.

A lead-scoring model trained on its own data. ASTRA launched in July 2021, built with data analytics partner G-Square Solutions. The company's 2023 patent announcement said it was built on behaviour data from over 5 million leads, ran 74 models for projects of different sizes and stages, and had been used in over 200 projects across 21 cities. What it does is ordinary CRM work done carefully: rank enquiries by likelihood to book, so the sales team calls the right people first.

A channel partner platform. In 2022, Anarock told Business India it planned "a digital distribution platform for aggregating channel partners," including a route into Tier II and III cities. That became ANAROCK Channel Partner (ACP), an app and web portal. The ACP site today lists client registration, booking tracking, "secure client tagging" and brokerage credited within 72 hours of booking validation. It claims ₹500 crore+ GMV transacted, 150+ exclusive projects, 12K+ CPs onboarded and ₹10 crore+ in brokerage paid.

An AI layer on top of the CRM. ANAROCK.AI went live in November 2025. It has nine tools in three groups:

• Genie suite: chatbots and voice bots for walk-ins, channel partners, reputation management and referrals.
• Astra suite: lead scoring, reviving dead leads, hiring and sales team performance.
• CP 360: a "CP Ranker" that scores which partners to push a project to.

A June 2026 ElevenLabs case study says the voice agents use regional voices matched to each market. They qualify leads on budget, configuration and locality, then push for a site visit.

Adjacent businesses bought, not built. In January 2021 Anarock bought 100% of ApnaComplex, a society management app serving over 600,000 households in 80 cities, now sold as ANACITY. In April 2022 it bought 75% of coworking marketplace myHQ for ₹125 crore in cash and stock. On 24 September 2026 it announced a 51% stake in architecture firm DSP Design Associates, terms undisclosed. It raised ₹200 crore from 360 One Asset Management in 2024.

The numbers

Revenue (group, as stated by the company; fiscal years April to March):

Sources: NewsVoir release (FY23–FY24), PTI via Deccan Herald (FY25), PTI via Outlook Business (FY26).

Where the sources disagree:

• The FY24 press release gives ₹566 crore in its headline and ₹556 crore in its body.
• Inc42's profile, built from statutory filings, shows ₹517.4 crore for FY24 and ₹677.3 crore for FY25. That is lower than the group figures, probably because it covers one legal entity, not the whole group.
• Net profit is not disclosed in any source we could verify. Third-party databases give conflicting summaries (one reports EBITDA down about 43% in FY25), so we leave profit out.

Sales facilitated: ₹16,435 crore of housing in FY26. For the three years to 2022, Business India reported over 41,500 homes worth over ₹51,500 crore.

AI results (company claims from the November 2025 launch, not independently checked):

• Four core tools ran from 2024 across 80+ projects. They contributed to 700 sales worth ₹750 crore, which was 10–15% of sales in those projects.
• AI-assisted bookings made up 15–45% of sales in leading Anarock-mandated projects, and 60% in one Chennai project.
• The top 10% of leads flagged by Astra Platinum produced 40–60% of bookings.
• Astra Phoenix re-engaged 8–9% of failed leads, which contributed 5–6% of total bookings.
• The top 5% of channel partners generated 40–50% of bookings.

The ElevenLabs case study adds that nearly 30% of customer conversations now happen outside working hours.

Headcount: 1,800+ (2022–23), 2,200+ (2024), about 2,300 (2025 and 2026).

What worked, what broke

What worked

• The data became the product. Each mandate adds leads, visits and bookings to one database. ASTRA, CP Ranker and Phoenix only work because years of sales records are stored in a form a model can learn from. A broker whose history lives in personal WhatsApp chats can't do this.
• It targeted the partner's real problem. ACP leads with 72-hour brokerage payouts and client tagging, not features. Its testimonials keep coming back to cash flow and protection against client poaching. Those two issues decide whether a CP brings a buyer to your project or someone else's.
• It measured where bookings come from. The claim that 5% of partners drive 40–50% of bookings is only possible if every booking is tagged to a partner. That tagging is plain CRM discipline, and it's the most useful thing in this case.
• Out-of-hours response. Nearly 30% of conversations happen when human teams are offline. Night enquiries used to wait for a morning callback, and by then the buyer had often called another project.

What broke, or stays unclear

• Residential is slowing as a share. Group revenue grew 25% in FY26, but residential revenue grew about 5% (₹420 crore to ₹442 crore). The core broking business now brings in less than half of revenue, and growth is coming from leasing, advisory and acquisitions.
• Targets were missed. Anarock set a ₹800 crore target for FY25 and reported ₹755 crore. In 2024, vice chairman Santhosh Kumar said the company was "confident of close to tripling" revenue to ₹1,500 crore by FY27. From ₹946 crore, that needs about 59% growth in one year.
• The AI numbers are self-reported. Every attribution figure comes from company press releases or a vendor case study. "AI-assisted" isn't defined, and none of the figures has been audited.
• The platform is still small compared with the network. ACP lists 12K+ onboarded partners against the 80,000+ the company says it manages. The two numbers may be defined differently, but most of the network is not yet on the app.
• Research numbers get challenged. For April–June 2026, Anarock reported top-seven-city sales down 6% to 90,715 units. The same week, PropEquity reported 19% growth to 1,12,458 units. When data is part of your pitch, a competing dataset weakens it.

How to apply this at your scale

You don't need 90,000 sold units or an AI suite to use Anarock's main lesson. You need every enquiry, site visit and booking tagged by source and partner in one system. Then you can see which partners and which leads actually close. For a channel partner firm or a mid-size developer, that system is a CRM built around how property sales work.

Practical steps:

1. Tag at the first touch. Every lead gets a source (portal, ad, referral, CP name) and a timestamp when it arrives. Store tagging rules in the CRM, not in a manager's head. That settles who owns a client before anyone argues about it.
2. Rank your partners every quarter. Count bookings per partner, not registrations. If your numbers look like Anarock's, a small group does most of the selling. Give that group early inventory access and faster payouts.
3. Show brokers their payout status. A partner who can see "booking validated, payout due on this date" calls less and sends more clients. You don't have to promise 72 hours. You do have to show an honest date.
4. Work your dead leads. Before buying more leads, run a structured call or WhatsApp sequence on last quarter's lost enquiries. Anarock says this produced 5–6% of its bookings. Your number will differ, but you already have the leads.
5. Cover the night shift. Log enquiries that arrive after hours and reply to them first thing, or with an automated first reply that asks budget, BHK and locality.
6. Define your metrics before you claim results. Decide what "influenced" or "assisted" means before you report it, so your numbers hold up when a developer asks.

Start with clean tagging and a partner leaderboard. Scoring models and voice bots come later, once you have a year of data worth training on.