Technology is rapidly reshaping India’s real estate market, bringing new levels of intelligence, efficiency, and transparency to how homes are discovered, marketed, and sold. With regard to the same, Tech Achieve Media recently spoke to Sunil Mishra, Chief AI and Strategy Officer at ANAROCK Group, who is leading the development of ANAROCK.AI across 49 identified use cases. In this conversation, he dives into the core problems the platform was built to solve, the growing role of AI-assisted bookings, the company’s privacy-first approach to data, and the expanding opportunities for AI in pricing intelligence, premium housing, land optimisation, and beyond.
TAM: What core problem in India’s residential real estate market were you aiming to solve when conceptualizing this platform?
Sunil Mishra: There are essentially two levels at which we examine a problem statement. The first is on the B2B or institutional side, which involves real estate developers. They typically face a range of sales and marketing challenges. The second is on the home-buyer side, representing individuals like us who, at some point, aspire to purchase a home. Each segment has distinct problem statements, which I’ll outline briefly.
For real estate developers, the core challenge is a classic sales and marketing issue. They are overwhelmed by a deluge of millions of home-buyer leads and tens of thousands of brokers or channel partners. These two complex ecosystems operate simultaneously, making it crucial to identify the right customer and the right channel partner while ensuring neither is overlooked. This is the fundamental need behind any sales and marketing intervention in real estate, and it is exactly what most of our tools at Anarock.AI are designed to solve.
Amidst this vast sea of leads and brokers, we help developers determine which customers are genuinely serious about buying and which channel partners are genuinely active and reliable. That addresses the institutional side.
On the home-buyer side, the challenges are equally significant. Sales teams often fail to provide adequate attention. A buyer enquiring about a 2 BHK may be contacted for a 3 BHK instead. Important work moments, such as performance appraisal discussions, are interrupted by follow-up calls on leads dropped days earlier. The entire home-buying experience is riddled with inefficiencies.
While Anarock.ai is fundamentally a B2B enterprise AI platform, many of its interventions ultimately enhance the home-buyer experience as well making engagement smoother, more consistent, and more intuitive.
TAM: AI-assisted bookings now account for up to 45% of sales in some projects. Do you envision AI becoming the primary sales engine in the next few years?
Sunil Mishra: Even if the percentage of AI-assisted bookings eventually reaches 100%, the emphasis will still firmly remain on the word assisted. That part is not going away anytime soon. There is a fundamental physicality to real estate buying in India, and it plays out in two ways.
First, buyers want to see what they are purchasing. Whether it’s a completed home, a plot of land, or even a project still on paper, people prefer to visit the site. They might discover something on-ground, like a drain behind the plot, that no online listing would reveal. And given that purchasing a home typically involves anywhere from Rs 50 lakh to Rs 50 crore, representing 30 to 40% of an individual’s wealth (and in some cases up to 80%), the need for physical inspection is natural.
Second, the sales process itself is inherently physical. Brokers or channel partners need to meet customers face-to-face to build trust and guide them through the decision. This in-person interaction remains a critical part of the transaction. For these reasons, AI will continue to play a supportive role, enhancing efficiency, improving targeting, and helping sales teams perform better, but not replacing the human element.
To answer your question on efficiency: yes, AI will enable developers to sell more homes and generate more revenue with the same number of employees. However, we are not yet at a stage, at least for the next three to four years, where AI will reduce headcount or fully automate the sales process. Instead, it will make life easier for both sales teams and home buyers.
TAM: With proprietary data from over 7 million customer enquiries, what safeguards are in place to ensure privacy, compliance, and ethical AI deployment?
Sunil Mishra: The DPDPA has been in development for the last seven to eight years, so we have always kept it in mind while building our technology. The first important point is related to how we train our machine learning models. These models never required full mobile numbers of customer leads. From the very beginning, we masked the last five digits before feeding the data into the models, because the models simply do not need that information. What they do need are insights such as when the lead was dropped, which project it related to, the telecom operator (which is identifiable from the first five digits), the day of the week the lead came in, and similar metadata. These signals are sufficient to train the model effectively. In other words, private or personally identifiable information was never part of the model training process.
The second dimension relates to generative AI. While the first part deals with predictive AI and traditional machine learning, our generative AI tools, chatbots and voicebots, are built on large language models. As we all know, LLMs became mainstream after OpenAI launched ChatGPT in November 2022. Many chatbots in the market today rely on models trained on the open Internet.
We have taken a very different approach. Our models are trained on highly curated, project-specific datasets. For example, if we are selling a project in South Bombay, the chatbot is trained the documents containing all relevant details about that specific project. It is also integrated with the CRM, so information about site visits and customer interactions is part of its knowledge base.
Crucially, our chatbot does not access the Internet for information. It does not go to any external websites to pull content. We have intentionally built it like a focused, blindfolded racehorse, extremely smart, but operating strictly within boundaries. It queries the LLM, produces contextually accurate responses, and remains tightly grounded in the project data we provide. This prevents hallucinations and avoids the “garbage in, garbage out” problem that arises when models rely on uncontrolled Internet data. With these two measures, privacy-focused machine learning and tightly governed generative AI, we are confident that we meet not only current DPDPA requirements but will also remain compliant with any future regulations related to data privacy and responsible AI usage.
TAM: Despite a 9% drop in units sold, overall sales value surged 14% in Q3 2025. How is ANAROCK.AI positioned to help developers navigate this demand for premium housing?
Sunil Mishra: First of all, these are market-wide numbers for the entire real estate, specifically the new home sales segment, which has seen a 9% decline in unit sales but an increase in overall value.
When home sales are struggling or not growing, tools like ours become even more important. Developers need better returns on their marketing investments, and efficiency-oriented solutions gain prominence in such periods. When the market is booming and everything is selling, such tools often don’t receive the recognition they deserve. So the timing for adopting efficiency-enhancing technologies is absolutely right.
Now, coming to your specific question about premium projects: our models work equally well across different price segments. Historically, one of the challenges in digital marketing, something widely discussed over the past decade, was that lead generation for high-end projects was weaker compared to mass-market projects. A similar pattern can appear in AI as well, and there is a scientific explanation for it.
AI models rely on data. Lower-priced projects naturally generate far more leads because more people can afford them. As you move into Rs 5 crore, Rs 10 crore, or Rs 25 crore properties, the volume of leads drops significantly. Smaller datasets mean the models have less information to learn from, which can affect prediction quality.
Because of this inherent limitation, both digital marketing and AI can sometimes be less effective for ultra-premium projects compared to mass-market ones. However, based on our experience across more than 90 projects that our models have been trained on over the past two years, including at least 15 high-end projects, we have seen strong results. Even in premium projects, the system has been able to identify missed bookings and overlooked brokers effectively.
TAM: As AI matures, do you foresee ANAROCK.AI expanding into areas like property pricing intelligence, developer risk scoring, or smart project launch planning?
Sunil Mishra: The answer is yes. We actually have a total of 49 use cases identified. I have a detailed slide deck outlining all 49, and so far, Anarock.ai covers about 20 of them. That means there are still 29 use cases to unlock. One major high-value use case is price prediction for projects, based on a wide range of macroeconomic and micro-market indicators. Another important area is expanding beyond residential real estate. Until now, our work has largely focused on the residential segment.
The other asset classes, retail leasing, retail sales, and hospitality, are also significant. We have already begun building models for these categories as well, though data availability is more limited. We are making steady progress. Additionally, we are exploring integrations in upstream processes such as construction and land acquisition. For example, determining the optimal use of a parcel of land is an important use case where generative AI can be applied effectively.
So yes, we will definitely move in those directions over time. At the moment, our priority is to streamline sales, marketing, and customer experience functions for residential real estate developers. We are going deep into the 20 existing use cases across nine tools. Meanwhile, we are also assembling teams and collaborating with external partners to work on several of the remaining 29 use cases.
