Sunil Mishra – CEO-Growth Businesses, Chief AI & Strategy Officer
What is the Omni-AI part (in Anarock.AI)?
Anarock.AI is a combination of some Predictive AI Machine Learning Models and a bunch of conversational AI Agents, powered by Gen AI LLMs. Instead of running these independently and isolated from each other, we have made many Business Use Cases employ a combination of such ML models and AI Agents, to give a much more effective and enhanced business output.
1. AI x AI = AI-squared :
In the most basic form of an Omni-AI system, a Predictive ML Model predicts which customer or Channel Partner (CP) is more serious than others. The Models rank the customers or CPs in descending order of propensity to contribute business to us. The AI Agent (conversational chatbot or voicebot) goes through the output of the Predictive AI Model and engages more intensely with the higher ranked elements vs the lower ranked ones.
Some examples of use cases employing such an Omi-AI instance are as shown below :
Astra Platinum (ML Model) rates the home-buyer leads in real-time basis on propensity to buy a home. This Model uses the Anarock data of 8 years of 7mn leads of which 100,000 have successfully purchased a home through us. The Walk-in Genie Whatsapp
Chatbot starts engaging with such leads, with the intended objective of convincing the leads to come for a project site visit. The Walk-in Genie connects with the Platinum leads (top 10% of leads) 3 times in the first week vs one time with non-Platinum leads.
Another similar use case is a combination of Astra Phoenix model and the Junked & Failed Walk-in Genie, as shown above.
2. Multi-agent multi-model Use Case :
There are some business situations where the outputs of a Predictive AI model are picked up in a segmented fashion by one AI Agent, which carries out some actions. After that, another AI Agent picks up the action and carries out a different set of actions to execute some business output. Please see the schematic shown below.
This is almost like a set of multi-layered AI actions taking place, and theoretically, can be extended to a few more agents if the process is more complex.
Anarock.AI does employ such a multi-layered AI in action, when we work on the CP Management piece. Please see the flow below :
The first step involves CP Ranker, a predictive AI ML Model used to rank CPs for a particular Project in any city. The Model has been developed by crunching data on CP behaviour taken from multiple sources – Anarock Residential Mandate CP data from 950+ Projects (70000 CPs performance data over 50,000 homes sold through CPs), RERA data, data from property portals.
Leaving the top 100 CPs for the Project to be met in person by CP sales team, CP Genie starts interacting with CP Nos 101-2000, as per the ranks given by CP Ranker. The CP Genie shares project information, brokerage information with CPs and convinces them to bring their buyers to the project site visit. Some CPs also share the mobile numbers of their customers in the chat with CP Genie. These numbers get captured automatically in our CRM as a home-buyer lead.
The Walk-in Genie then gets into action and starts conversing with these home-buyer leads and shares project information with them, answers their queries and convinces them to come for a site visit.
Thus, we have created a multi-agentic Omni-AI system, which actually simulates the entire CP Management Team of a real estate project.
3. How the outputs improve in Omni-AI setup
Now that we have understood the setup of an Omni-AI system, it is not tough to understand some “multiplicative” benefits of the setup, from the feedback loop. Consider the Omni-AI use case below :
Whatsapp conversations and Voicebot recordings of Walk-in Genie with customers are fed back into Astra ML model as an input in the ML modeling. This improves the predictive ability of Astra Phoenix Model and it is able to predict with much higher accuracy.
Astra Predictive Models (Platinum and Phoenix) are getting more refined now with this outputs from AI Agents being fed back to them. This is almost a positive feedback loop created.
