Cian's analytics teams focus on two primary areas: optimizing CRM communications based on user behavior and ensuring the credibility of property listings. Analysts work to identify user intent, distinguishing between those actively searching for property and those in an exploratory phase, to tailor push notifications and recommendations effectively.
The platform utilizes machine learning models, including LightFM for cold-start scenarios and transformer-based models to analyze the sequence of user actions. These tools help predict which properties are relevant to specific users based on parameters like location, price, and property type, while also accounting for the evolving search criteria of customers.
A/B testing is a critical component of the CRM strategy. Experiments have shown that push notifications containing specific details, such as the number of rooms and price, significantly outperform generic messages in both open rates and target actions. Analysts also monitor for cannibalization to ensure new communication scenarios do not negatively impact existing ones.
To combat fraudulent listings, the company employs a multi-layered approach combining automated heuristics, machine learning models, and manual moderation. The system evaluates various risk signals, including account history, price deviations, and image authenticity, to identify suspicious activity. Analysts play a key role in preparing datasets for training models to detect new types of violations, such as AI-generated interior renders.
The team also manages trust mechanics, such as 'super-agent' and 'super-host' statuses, by analyzing behaviors associated with reliable property listings. Analysts continuously monitor the impact of these statuses on user engagement and platform safety, ensuring that the moderation process remains effective as new challenges, like synthetic imagery, emerge.
