Inchcape LLM

Project Team 2
Project Team
High Level Architecture
Dataset

Inchcape AFS (AFtersaleS) Advisor Bot

Strategic Significance of Aftersales Services at Inchcape

Inchcape rigorously performs approximately 500,000 aftersales services annually across more than 40 countries. Aftersales services are a pivotal element in the Automotive Value Chain, offering significant opportunities to enhance customer retention, increase customer lifetime value, and generate a steady revenue stream. Notably, a typical vehicle requires servicing biannually. Despite the apparent benefits of the Aftersales segment, retaining customers beyond the initial year or post-warranty period remains challenging. While many customers aspire to maintain their vehicles in excellent condition, factors such as perceived inconvenience or prohibitive repair costs often deter timely service engagement. Over the past two years, under Inchcape's Accelerate Strategy, our team has developed an innovative Aftersales Churn Prediction Algorithm. This tool utilizes historical data to identify customers at risk of churn. However, the algorithm's efficacy is limited by the need for manual outreach, such as phone calls or emails, to remind customers, which is resource-intensive given our large customer base. Addressing this issue is a key objective of our current project, aiming to optimize revenue potential.

Project Description: Integrating AI for Enhanced Customer Engagement

Our project synergizes the capabilities of the Aftersales Churn Prediction AI with advanced Generative AI for personalized content creation. By analyzing churn probabilities and key customer features (e.g., proximity to service centers, previous service costs), our Generative AI algorithm crafts customized communication strategies. These include personalized emails, Chatbot interactions on Inchcape websites, and mobile app push notifications, all tailored to the customer's churn likelihood and offering targeted incentives. This automated communication approach enables us to engage the entirety of our high-churn customer segment across various markets, thereby encouraging timely service uptake and enhancing customer retention.

Development Process

The development involved using Retrieval Augmentation Generation (RAG) to enhance the context-awareness of our Language Model (LLM) with sample text files. We integrated the output from our Churn Prediction algorithm to refine our model further. The model was trained using a variety of prompts to accurately represent customer churn personas. We implemented a database feature for user-specific communication, ensuring privacy in Chatbot interactions. Our knowledge base includes standard offers, which are supplemented by custom offers derived from the churn model insights.

Encountered Challenges

We faced hardware limitations when operating the LLM on local machines or due to restricted cluster capacity. Additionally, the integration of the nascent Pandas AI library presented some challenges.

Key Accomplishments

We successfully integrated RAG-based methods with our existing Predictive AI framework. This integration is particularly significant given that Inchcape's aftersales segment accounts for nearly £1 billion in annual revenue. Even marginal enhancements in this domain through our project could translate into substantial revenue increases. Our achievements also include basic Natural Language Processing (NLP) querying and the innovative combination of Predictive and Generative AI techniques.

Future Directions for Inchcape LLM

Development of an Aftersales Bot capable of directly contacting customers via internet telephony for service bookings, utilizing LLM-driven conversations.
Implementation of automated customer communication journeys in CRM systems like Salesforce.
Expansion of the communication strategy to include offers and notifications via mobile apps, moving beyond the current website-based Chatbot interface.