About the Client
The client operates one of India’s largest and most accessible passenger vehicle retail networks, with over 3,000 outlets across more than 2,500 towns and cities and a customer community exceeding 82 lakh people nationwide. Their platform offers a tech‑enabled car‑buying experience that supports seamless exploration, customization, and purchase journeys for a wide range of popular models, helping it lead the market in reach and engagement.
Business Situation
Many customers today expect personalized interactions, and relevance is key to capturing their attention. The client wanted to engage a vast user base with timely and meaningful offers, tailored to each customer’s preferences and history. Delivering the right recommendation to the right segment required a system capable of understanding user behavior and preferences across different scenarios.
The client aimed to enhance engagement and drive loyalty through personalized messaging. Their goal was to send context-specific recommendations- for example, offering new car suggestions to long-term owners and maintenance discounts to recent buyers, ensuring every customer received relevant communication at the right time.
After a thorough evaluation of more than 20 software engineering companies, the client zeroed in on Unthinkable Solution for our vast expertise in developing AI-driven personalization engine.
Some of their key requirements included:
Deliver personalized recommendations and messages to customers efficiently and at scale.
Accurately segment and filter customers based on attributes like birthday, anniversary, maintenance dates, car model, and purchase history.
Predict customer preferences and ratings to suggest relevant car models or offers.
Ensure messaging is sent at the optimal time to maximize engagement and response.
Maintain user privacy while enabling AI-driven model training and data analysis.
Generate brand-aligned, human-like messages automatically using AI and LLMs.
The Solution
Our team worked closely with the client to understand their goals, requirements, and key challenges. During the product discovery phase, we clearly defined the project scope, chose the right technologies, and planned the user experience. This helped us move forward with a clear and structured approach.
We then built the web application using GenAI-based models along with Python and Streamlit. GPT-3.5 was used to create personalized messages for users. The solution was hosted on Amazon EC2 to handle growing usage, while Databricks and Jupyter Notebook supported model training and data analysis.
Some of the key features include:
Addressing Data Privacy with Synthetic Data
The client could not share its full user dataset due to strict privacy protocols, which limited the data available for model training. This created a gap that could have affected recommendation quality. To work around this, the team expanded the dataset using synthetic data. Artificial data points were generated to mirror real user behaviour, allowing the model to train on a broader dataset without exposing sensitive information. This improved accuracy while keeping user data secure.
Car Recommendation Engine
The recommendation engine suggests vehicles based on past purchases and user preferences. It uses collaborative filtering to find patterns across users and matrix factorisation to estimate a customer’s interest in cars they have not interacted with. Inputs include demographic details, vehicle specifications, ownership history, and test drive feedback. This helps generate suggestions that are relevant to each customer rather than generic listings.
Vehicle Value Prediction
The system estimates the current value of a customer’s vehicle to support exchange, upgrade, or trade-in scenarios. This allows the client to send messages that are tied to real opportunities, such as suggesting an upgrade when the timing and value align. It also supports retention by giving customers practical, data-backed options.
Best Time to Reach Customers
A rule-based system determines when to contact each customer. Timing is based on events like birthdays, purchase anniversaries, festivals, and expected service intervals. It also identifies when a customer may be ready for an upgrade or likely to respond to new offers. This ensures that communication is sent when it is most relevant, rather than at fixed or random intervals.
Rule-Based Messaging Triggers
The system uses defined rules to automate outreach across key scenarios. It sends messages on special occasions such as birthdays and purchase anniversaries, schedules greetings around major festivals, and reminds customers about upcoming service needs based on usage patterns. It also targets users who may be ready for upgrades or interested in new models with timely offers.
Message Generation and Personalisation
Messages are created by combining customer data with predefined guidelines for tone and style. The system follows clear rules to keep communication consistent with the brand, whether the tone is formal, friendly, or conversational. It also filters out words and phrases that do not fit brand guidelines. Past successful messages are used as a reference to improve relevance. Each message includes contextual details such as car recommendations or service timelines, so it feels specific to the customer rather than templated.
The Impact
The solution made the client’s marketing process more efficient and customer-focused. It helped them reach the right customers at the right time and cut down the time spent sorting users and sending recommendations.
Unthinkable’s AI-powered recommendation engine delivered personalized suggestions based on each customer’s preferences. This not only boosted customer satisfaction but also increased engagement and conversions.
As a result, personalization became a core part of the client’s marketing strategy, helping them build stronger relationships and achieve real business impact.
