Project Overview:

In our exciting marketing mix modeling project for a leading pharmaceutical company in the United States, we are delving into the world of data and analytics to help optimize their marketing strategies and support business in allocating advertising budgets across more than 30 channels. Integrated with the client’s data science team, our tasks involve first fine-tuning and calibrating the model, aligning it with business intuition to ensure it accurately reflects real-world scenarios. Secondly, we provide budget recommendations using linear optimization, ensuring a strategic allocation of resources. Additionally, we implement MLOps best practices (MLflow, GitHub CI, etc.) to make the system scalable, automating tracking and deployment tasks to easily support business users. This not only ensures project scalability but also allows business users to effortlessly run models for multiple products across various countries.

Example of hypothetical response curves built using regression bayesian models


Example of contribution plot giving the incremental impact of each channel


  • Analyse and understand data in preparation for modelling (get a prior belief on the impact, chose the transformation to apply and get a sens of the variation that we expect the model to detect).
  • Update the model and fine-tune it to align with business expectations.
  • Assist business users during the execution of optimization scenarios, including clarifying technical concepts and resolving issues related to the optimizer (CBC, Gurobi).
  • Streamline the workflow and incorporate MLops best practices to enable non-technical users to execute and analyze models efficiently.


  • the project initially focused on a single product that garnered leadership attention. As it progressed, there was a transition from a single-team setup dedicated to the product to the establishment of a core team with the responsibility of building a Platform as a Service (PaaS) product for power users.
  • One notable advancement in the project was the automation of report generation and model comparison. This automation aimed to facilitate easier access to data for business teams, enabling them to make decisions independently of the technical teams. This shift towards automation not only streamlines processes but also empowers non-technical teams by providing them with tools to access and analyze data, fostering a more efficient decision-making process.
  • Expanding the scope from a single product to a broader PaaS offering suggests a strategic move towards serving a wider audience or catering to more diverse needs.



Snowflake, AWS, pyStan, CBC, Mlflow