Developing a Personalized AI Recommendation Engine for India’s Top Automaker

Success Story

About the Client

The client is India’s largest automobile manufacturer with annual revenues of more than US $17 billion. It played a key role in starting the country’s automobile revolution and today holds over 40% of the market. With more than 28 million customers, the company continues to be a trusted and leading name in the Indian automotive industry.

Country:

India

Industry:

Automobile

Download PDF

Business Situation

Customer expectations have shifted significantly in recent years. Studies indicate that 71% of buyers now expect personalized interactions, and nearly 76% move away from brands that fail to provide them. At the same time, 91% of customers prefer to engage with companies that share relevant and timely offers.

Personalization has therefore become central to capturing attention and building long-term loyalty. For our client, this was an important priority given its large customer base. The company wanted to strengthen engagement by tailoring communication to individual needs and milestones.

The objective was to design a system that could deliver customized recommendations. For example, long-term customers who had owned a vehicle for more than five years would receive suggestions for new models, while recent buyers would be targeted with service offers and maintenance discounts.

Why a Custom Digital Platform Was Needed

As India’s largest automobile manufacturer with over 40% market share, the client manages a customer base of millions. Reaching each customer with personalized suggestions presented a major operational challenge.
The company initially relied on a manual approach to manage its messaging. Teams had to review large datasets—such as customer names, car models, purchase year, and other details—before segmenting users, identifying their needs, and drafting relevant offers for each group.
This approach proved difficult to scale. The manual process consumed significant time and resources, yet delivered minimal impact on customer engagement or business growth. The client recognized that a custom digital platform was necessary to streamline personalization and make it truly effective.

The Solution

Unthinkable worked closely with the client’s team to gain a clear understanding of their objectives and identify process gaps. The engagement began with a Discover and Frame workshop, where the scope of the project, technology stack, and user experience design were defined in detail.

The result was the development of a custom web application designed to deliver personalized, high-impact messages at scale. To achieve this, the team used GenAI-driven LLM models, along with Python and Streamlit, to build the application framework. OpenAI’s GPT-3.5 model was integrated to generate customized recommendations and messages for individual users.

From an infrastructure perspective, the solution was deployed on Amazon EC2 Instances, ensuring scalability and performance under high user traffic and large data volumes. Databricks provided a unified workspace for model training, while Jupyter Notebook supported data exploration, training, and visualization throughout the development process.

Tackling Privacy Concerns with Synthetic Data

Although the client manages millions of customer records, data privacy protocols restricted the sharing of complete datasets with Unthinkable. Only a limited portion of user data could be accessed, which created a major challenge for building accurate recommendation models.

To address this, Unthinkable applied synthetic data generation techniques supported by machine learning. The team expanded the available dataset by creating artificial data points that reflected real-world patterns and behaviors. This augmentation strengthened model training, improved accuracy, and enabled the system to produce more reliable and personalized recommendations without compromising data privacy.

Tackling Privacy Concerns with Synthetic Data

How the Solution Works

The platform uses targeted filters such as birthdays, anniversaries, maintenance schedules, and other key criteria to identify customers most likely to respond to personalized communication.

Through the application, the client can view individual customer profiles, refine selections based on factors like region, current car model, and budget, and then choose from AI-generated message options. This eliminates the need to manually sift through large datasets or draft messages for each user.

The system tailors recommendations to customer history: owners with vehicles older than five years receive suggestions for new models, while recent buyers are sent maintenance offers and service reminders.

How the System Improves Customer Interactions

Car Recommendation Engine

The solution incorporates a recommendation engine built on collaborative filtering techniques to provide personalized vehicle suggestions. These recommendations draw on each customer’s purchase history and stated preferences.

To predict the likelihood of interest in a particular car, the system applies matrix factorization methods. This allows it to estimate how customers might rate vehicles, even if they have not interacted with them directly.

The engine processes multiple data sources, including customer demographics (age, gender, location), vehicle attributes (pricing, fuel efficiency, specifications), ownership history, and feedback from test drives. By training the model on these user profiles, the platform generates tailored car suggestions that align closely with individual customer needs.

True Value Prediction
The platform includes a feature to estimate the resale value of a customer’s current vehicle. This is particularly useful when customers are considering an exchange, upgrade, or trade-in. By providing accurate value predictions, the system enables relevant and timely communication, strengthening customer trust and loyalty toward the brand.

Best Time Prediction
To maximize engagement, Unthinkable combined the capabilities of Large Language Models (LLMs) with automation to generate natural, personalized messages. A rule-based system determines the most effective moments to connect with each customer.

The rules include:

Special Occasions: Sending personalized greetings or offers on car purchase anniversaries, birthdays, and other key milestones.
Festive Days: Scheduling tailored messages and greetings around cultural or religious festivals to build stronger connections.
Maintenance Reminders: Using purchase and service history to notify customers about upcoming maintenance needs.
New Offers: Targeting customers due for an upgrade or showing interest in newer models with promotions, launch details, and exclusive offers.

Message Generation

The message generation process was built on Large Language Models (LLMs) to convert customer data and insights into personalized communication. Using GPT-3.5, Unthinkable trained the system with clear guidelines so messages would consistently reflect the client’s brand voice—whether professional, friendly, or engaging in tone.

Through prompt engineering, the solution was further optimized to apply content-based filtering. Keywords or phrases that did not align with brand policies were excluded, while approved language was embedded into the LLM framework to maintain consistency across all communication.

The recommendation engine was also trained using examples of previously successful messages. These were adapted and refined to match the unique profile and needs of each customer.

To make the experience seamless, the platform integrates contextual data—such as car recommendations, service schedules, and special dates—into natural, human-like messages. This approach ensures that every interaction feels timely, relevant, and personal.

The Impact

The solution enabled a more efficient and customer-focused marketing process, allowing the client to connect with the right customers at the right moment. Automation reduced the time previously spent on sorting users and sending recommendations, while the AI-driven recommendation engine tailored suggestions to individual preferences.

As a result, the client experienced higher customer satisfaction, increased engagement, and improved conversion rates. Personalization has now become a central element of the client’s marketing strategy, strengthening loyalty and reinforcing their leadership in the automotive market.

Developing a Personalized AI Recommendation Engine for India’s Top Automaker

About the Client

Country:

Industry:

Business Situation

The Solution

Tackling Privacy Concerns with Synthetic Data

How the Solution Works

How the System Improves Customer Interactions

The Impact

Awards & Accolades

Delaware, USA

London, UK

Dubai, UAE

Gurugram, India

Discover Unthinkable

Industries

Services

Domain Expertise