Machine learning

In order to streamline existing use cases and discover new ones, our client partnered with Onepoint to separate its B2B and B2C models. The goal was to identify specific AI use cases for each segment, estimate the workloads involved, and balance the AI investments between B2B and B2C models.

Some of the identified use cases included email treatment recommendations, customer propensity analysis for calls, and cancellation prevention. The ultimate goal was to help the client’s teams gain a deeper understanding of their current portfolio and improve their ability to predict future trends.

Solution

The study began with a scoping phase, which involved research, documentation review, interviews with teams, and exploration of available data. This allowed for a clear understanding of the business challenges and the data available for building models. It helped assess existing use cases and identify new opportunities.

After prioritizing and balancing the B2B and B2C use cases to avoid over-representation of one segment over the other, the teams proceeded with model development tailored to each use case.

Outcome

The iterative approach, working closely with business teams, ensured excellent agility and the definition of SMART intermediate objectives. These objectives were fully met, with use cases being reorganized, prioritized, and developed into models that provided consistent and actionable results.

To further its growth strategy, our client aimed to better understand its customers in order to offer them the most relevant services. Given the wide range of products offered by the company, the challenge of matching the right products to the right customers had been ongoing for some time. However, this issue was about to be addressed effectively with the help of artificial intelligence.

Solution

Throughout the engagement with the client’s teams, Onepoint began by exploring and restructuring customer personal data, equipment data, and transaction history. This allowed for a clear definition of business needs and their translation into 17 predictive models. These models were then developed and iteratively refined.

The results of the models highlighted key levers that could be quickly activated in the branches and revealed a segmentation of customers based on their likelihood to be interested in each banking product.

Additionally, socio-demographic and economic studies were conducted to validate the model results and ensure their relevance.

Outcome

Through this engagement, Onepoint delivered predictions of customer interest for each banking product, handed over the code to the client’s teams in a way that ensured understanding and maintainability, presented results to various management levels, and highlighted areas for improvement in the management of customer databases to facilitate their future AI needs.

During Onepoint’s engagement with the client, the dedicated teams worked on optimizing the volume of reprints when a book experiences a better-than-expected commercial launch. This case posed significant logistical and business implications, such as determining the appropriate number of copies to print while avoiding overstocking and stockouts, as well as reducing the generally high costs associated with reprints.

Solution

Onepoint’s investment began with an analysis phase focused on understanding the available data and preparing it for use. The clear definition of objectives and evaluation metrics enabled the team to select the most suitable algorithms for addressing the use case. As development progressed, additional complexities were identified, including the strong temporal dependency in the data, the interpretability of predictions, and the evaluation of the client’s actual performance, which were mitigated as much as possible.

Outcome

The delivered model provided a decision-support tool to better calibrate reprint decisions. The data used to train the most relevant model came from the client’s sales data between 2015 and 2020, allowing for a satisfactory generalization of the model’s predictions.

The client’s teams, satisfied with the model’s performance, have since moved forward with industrializing the model to integrate it into their operational processes.