Natural Language Search Implementation for Customer Loyalty Program (ChatGPT)
Our client, a digital performance marketing business has amassed a wealth of valuable behavioural data based on user demographics and usage patterns. Although this data holds enormous potential for discovering new business opportunities, it remained buried inside a set of large data repositories, which were challenging to interrogate. This meant that the marketing and member engagement teams were unable to efficiently access this valuable business intelligence.
Customer
A market-leading digital business operating a nationwide membership scheme serving circa. 2 million members across the UK. In addition to an award-winning membership, events and voting platform, our customer also provides a suite of performance marketing services to national and international brands, all built around the central tenet of ‘community first’.
Problem Statement
The untapped potential in the behavioural data was well understood by our customers. However, they faced a significant challenge in accessing and decoding it. Traditional business intelligence tools like Power BI were considered but fell short of the customer’s need for a flexible, interactive and user-friendly method of integrating the data. Our customer sought to enable users to ask questions in plain, natural language and receive prompt, accurate and insightful responses. Our AI Engineering team designed and built a solution which converts user queries into often complex structured database queries and executes them against the data repositories to generate rapid results. LLM (Large Language Models), ChatGPT, and prompt engineering emerged as key enablers of this solution.
Data Potential Realization
User-Friendly Search Experience
LLM and ChatGPT
Natural Language Query
Solution Development
The project started with a workshop-based discovery phase, following which we conducted a short engineering-lab-based R&D / feasibility activity. Having established the boundaries and feasibility of the design concept, we executed a rapid, interactive development process. A team of eight AI/ML Engineers were assigned to build the PoC service. Further AI Engineering work is ongoing.
The project hinged on the innovative use of Large Language Models, specifically ChatGPT, to facilitate natural language search and query design. The following key components played a vital role:
- Structured Query Conversion: User queries, expressed in plain natural language, were translated into structured queries that could interrogate the data repositories effectively.
- Data Repository Integration: The structured queries were executed against the data repositories to quickly extract accurate results based on user intent.
- Advanced Interpretation: using prompt engineering, the AI Engineering team created a context for ChatGPT which enabled it to achieve a high level of proficiency in accurately interpreting user queries. It could handle various query structures, including those with unstructured elements such as flexible date formats like “Next year,” “within 12 months from now,” “Q4 2023,” and more.
Requirement Collection
Structured Query Conversion
Data Repository Integration
Outcomes
The project has already delivered substantial value and generated tremendous enthusiasm within the customer’s organisation. Some notable achievements include:
- Exceptional Interpretation: our AI Engineering team’s design enabled ChatGPT to interpret user questions in a manner which surpassed previous natural language processing efforts undertaken in earlier initiatives. It can decipher complex queries with ease, accommodating various formats and expressions.
- User-Friendly Interaction: Users now enjoy a user-friendly, conversational interface that allows them to seek data insights in a frictionless manner by simply typing their queries in natural language.
- Enormous Promise: The PoC, currently operating via a command line, demonstrates great promise, with the potential to revolutionise how its users interact with and extract insights from the ocean of behavioural data.
The client’s pursuit of an AI engineered solution, empowering users with a natural language search capability, represents a significant step towards unlocking the hidden potential of their valuable data. The combination of Large Language Models, ChatGPT, and prompt engineering has opened new horizons for data exploration and integration and has the potential to elevate its position in the market. As the project progresses, it is poised to render increasingly precise and commercially valuable intelligence, demonstrating the transformative power of AI Engineering and advanced language models in making data accessible and actionable.
