Enhancing Advanced Analytics with Generative AI

Leveraging large language models to improve data preparation, model development, and results interpretation of advanced analytics.

Generative AI, especially large language models (LLMs) like ChatGPT, can significantly boost the capabilities of advanced analytics by assisting in data preparation, accelerating model development, and facilitating a better understanding of results. While the hype around generative AI is growing, it’s crucial not to overlook the proven value of advanced analytics in business decision-making, such as predicting customer behaviour and optimising supply chains.  

Rather than viewing LLMs and advanced analytics as competing technologies, organisations should recognise their complementary nature. LLMs can help address challenges in both predictive and prescriptive analytics by incorporating unstructured data, translating business problems into analytical models, and explaining complex results.  

Unstructured Data for Predictive Analytics  

Predictive analytics are essential for data-driven decision-making. However, incorporating complex, unstructured data sources like customer reviews into predictive models can be challenging. LLMs simplify this process by efficiently extracting relevant information, enhancing the quality of predictions. For example, in a telco company’s project, LLMs were used to analyse customer complaints, significantly improving the prediction of the next best action in debt collection processes.  

Explaining Predictions  

Deploying predictive analytics often involves communicating complex results to non-technical stakeholders. Tools like SHAP analysis can be technical and hard to understand. LLMs bridge this gap by providing clear, simple explanations of model results. For instance, in a churn prediction model, LLMs helped explain the impact of variables like the number of products a customer has and their age on churn rates, facilitating better business decisions.  

Developing Prescriptive Analytics  

Prescriptive analytics, which optimise decision-making processes, can also benefit from LLMs. Developing effective prescriptive models involves translating business challenges into mathematical formulations, a task that LLMs can streamline. By engaging in dialogues with decision-makers, LLMs help define goals, constraints, and decision variables, quickly generating accurate optimisation models.  

Explaining Prescriptive Analytics  

Understanding the results of prescriptive models can be difficult for business users. LLMs assist by explaining model outcomes and identifying trade-offs in plain language, increasing trust and adoption of the models. For example, in a retail distribution problem, LLMs helped identify why a given algorithmic solution was infeasible, pinpointing supply constraints and suggesting improvements.  

Ensuring Model Quality  

Monitoring the performance and quality of advanced analytics models is critical. While integrating LLMs introduces new opportunities for errors, careful supervision and testing can mitigate these risks. By improving the incorporation of unstructured data and explaining model outputs, LLMs enhance both the efficiency and effectiveness of advanced analytics projects.  

Conclusion  

Integrating generative AI with advanced analytics offers significant benefits, including streamlined processes, improved model accuracy, and better stakeholder understanding. Multidisciplinary teams are essential to realise these opportunities, with business users empowered to take more active roles thanks to the accessibility of natural language processing. This integration not only enhances the quality of outcomes but also promotes greater trust and adoption of advanced analytics solutions.

By Pedro Amorim, Research Coordinator