Regarding the first real estate AI Agent, Zhou Xin still hasn't answered three questions.

“Break out of the old framework, let go of the past, put aside your pride, and look forward.”

On March 19th, Zhou Xin, the Chairman of the Board of Directors of E-House (China) Holdings Limited, delivered a speech titled “Let Go, Learn, Look Forward, and the Chinese Real Estate Industry Has Great Potential” at a conference hosted by his company. Two weeks later, just before the Tomb-Sweeping Festival, he released “China's First AI Agent in the Real Estate Industry”, CRIC AI.

Zhou Xin said, “I once considered naming it ‘AI CRIC’. I thought we should subvert ourselves and use AI to recreate what CRIC (the research institute under E-House Group) has done.”

Currently, this AI agent is only available in web version and is only undergoing internal testing on a small scale. Its main functional modules include: Search, Articles, Reports, and Knowledge Base, and it only focuses on the vertical real estate field.

Internal testing page of CRIC AI Agent

After the first online live broadcast release of this product, on April 8th, Zhou Xin flew from Shanghai to Beijing to specifically demonstrate this AI agent to the media offline.

In the search module, after users ask questions, CRIC AI Agent combines public information from the external network and E-House's internal database, reads and analyzes the searched materials, summarizes and refines the results, and presents them in the form of listing key points and displaying charts respectively.

In front of the analysis and answer section, the sources of the searched information will be displayed.

Internal testing page of CRIC AI Agent

After the answer results, there are also relevant follow-up questions. This is quite similar to the design of AI products like Doubao.

Internal testing page of CRIC AI Agent

In the “Articles” module, the writing process of CRIC AI Agent consists of 7 steps: analyzing the writing idea (similar to the DS display process), collecting writing materials, organizing and analyzing the materials, building a material mind map, generating a first draft of the article, proofreading, and completing the article writing.

In the “Material Mind Map” section, you can replace and adjust the structure and materials in the mind map, and modify the writing idea. The final version of the article can also be directly modified on the online interface, with an experience similar to an online document.

After the article is created, a text message will be sent to remind you, and the manuscript can be directly exported.

Process of generating an article

It is reported that CRIC AI Agent autonomously executes according to the innovatively designed real estate industry writing workflow and will autonomously decide to call the CRIC database and RAG knowledge base to ensure the professionalism and accuracy of the articles. The types of article creation include real estate market evaluations, land plot interpretations, policy analyses, etc., covering multiple sub - fields such as residential properties, shopping centers, second - hand houses, and building materials.

However, during the test, a 36Kr author found that the image materials captured were not the latest ones but a mix of historical materials from 2017, 2020, etc.

Generated article

In response, Zhou Xin said that the more important role of the AI agent should be a tool. After using the article - writing function to complete a task, the user needs to make some proactive modifications and confirmations.

The “Reports” module is the core part of CRIC AI Agent. Its data mainly comes from the internal database of CRIC Research Institute, covering various sub - fields such as residential properties, land, office buildings, commercial properties, leasing, healthcare and elderly care, industrial parks, cultural and tourism, enterprises, and macro - policies.

Zhou Xin revealed that in order to adapt to the calling logic of AI, CRIC sorted out and reconstructed all the previous years' data and reports, which required a lot of effort and cost.

According to the introduction, CRIC AI Agent can generate writing outlines and page - by - page analysis content according to the report type and can automatically call the required data from the CRIC database. Whether it is a regional market monthly report, a national annual report, project research and judgment, or land investment calculation, it can be completed intelligently and efficiently.

In the traditional manual writing mode, it usually takes 7 - 14 days to write a professional industry report, while CRIC AI Agent usually takes less than 20 minutes to generate a report.

Internal testing page of CRIC AI Agent

During the test, 36Kr found that in sub - fields where CRIC has accumulated a large database and updates it in real - time, such as market monthly reports and city monthly reports, CRIC AI performs better. It also includes a large number of charts and data processing, and the structure is very clear.

However, in relatively niche sub - fields such as “foreclosed property evaluation reports”, there is still a problem of data lag.

Internal testing page of CRIC AI Agent

The “Knowledge Base” module has two main functions. One is to call various documents, manuscripts, and reports in the database that CRIC has accumulated over the past 20 years. The other function is for users to create their own knowledge bases.

Internal testing page of CRIC AI Agent

“We won't venture into other fields. Our project leader keeps telling me to ‘focus, focus’,” Zhou Xin said. The target users of CRIC AI are employees in the real estate industry, such as marketing personnel, researchers, planners and designers, investment and expansion staff. It is different from general - purpose AI agents like Manus. “We hope to reduce the working time and cost of those who originally used CRIC's services and improve their efficiency, and be more accurate in the real estate field.”

Zhou Xin summarized the ideal functions of CRIC AI Agent into three capabilities: integrated engineering innovation ability, industry cognition ability, and data precipitation ability.

• Engineering ability: CRIC AI Agent has made many innovations in both being an all - in - one agent in the real estate field and providing solutions for industry - specific scenario tasks. For example, it fine - tunes the underlying large model for the real estate industry and accepts user feedback and instructions during the result - generation process.
• Industry ability: CRIC AI Agent is deeply bound to the real estate industry. Through the “real estate thinking framework” accumulated over nearly 30 years of experience, combined with the database and knowledge base accumulated by CRIC over 20 years, it uses the industry RAG LLM model to effectively improve the accuracy of information retrieval. At the same time, through the execution and verification of information sources by RAG, the possibility of AI hallucinations in the results output by the model is greatly reduced.
• Data ability: CRIC's real - time data covering 400 cities across the country and more than 10 sub - fields (such as land transactions, project sales, and policy interpretations) endows CRIC AI Agent with strong data capabilities. Empowered by AI, big data is mined, activated, and reasonably applied to various work results with geometric - speed computing power using rigorous real estate thinking.

Internal testing page of CRIC AI Agent

“We've always emphasized ‘doing things with precision’,” Zhou Xin said. CRIC AI Agent is planned to be officially launched in May, and an APP version will also be released.

However, there is a real - world problem that Zhou Xin has not clearly answered or planned for, namely: the computing power support after the official release, as well as the charging model and the issue of rising costs.

It's worth noting that currently, CRIC AI Agent does not support large - scale simultaneous use. Once there are concurrency issues, the cost of resolution will rise rapidly. And for most AI companies, it is still a money - burning business at present.

In addition, the general target user scale of CRIC AI Agent is about millions. Among them, individual users have a weak willingness to pay, while corporate purchases need to measure the benefits brought by using AI or whether it is worth it by comparing the cost of replacing employees with AI.

There is also a question that Zhou Xin has not answered, which is: how to solve the problem of subsequent data pollution. Focusing on the vertical field does not mean avoiding data and text pollution, especially at the project level in the real estate field. In reality, there is a large amount of public commercial content. The capture and use of this content will affect the professionalism of the AI analysis results.

Currently, the team of CRIC AI Agent has fewer than 20 people, mainly composed of industry analysts and technical personnel, which is equivalent to the scale of a startup company. If the number of users increases, the team may need more people and more professional positions.

There is also a real - world problem in the real estate industry: after a four - year downward period, no company is willing to burn money anymore. Currently, AI applications in the industry mainly focus on practical fields, such as design, security, and merchant management. CRIC AI is mining gold in market data, while the real estate market itself is increasingly dominated by central and state - owned enterprises. The market openness is decreasing, and it is moving towards being dominated by finance and resources, which means the value of market research itself will decline accordingly. Can this field still achieve higher cost - effectiveness through AI?

“From 2020 to 2030 is a folded era for the Chinese real estate industry, a transitional stage from large - scale development to large - scale asset management. We are in such a stage, and it feels a bit cold and painful,” Zhou Xin said in a recent speech.

Ultimately, only the market can provide the answer.