Mixed Media Models are a state-of-the-art predictive tool for forecasting marketing spend on a channel basis, but they are difficult to design and interpret without the help of a data scientist. This talk with discuss how we used Plotly Dash and CVXPy to build an application to allow our marketing partners to explore different budgeting scenarios using constrained optimization.
As the complexity real-world scenarios grows, data scientists often rely on powerful modeling frameworks that are more and more opaque, whose underlying decision making is not intuitively obvious. Therefore, data scientists must also emphasize the usability and interpretability of their modeling, and consider how business partners can best utilize and trust the results of these black-box approaches. Here at ThirdLove, we built an application using Plotly Dash to enable our marketing team interact with our marketing Media Mix Model (MMM), which forecasts orders as a function of dollars spent in each channel. By obviating the legacy approach our marketing team had been relying on with this usable MMM dashboard, we've witnessed a profound impact on our budgeting, and a significant return on investment. I will demo a simple version of the application and share the code that we used to build out the user interface in Dash, as well as the budget optimization feature using CVXPy.
This talk will be relevant for Data Scientists who want to increase adoption of their models, as well as any practitioner who is interested in applications of convex optimization in business applications. The audience should be familiar with python. Any other expertise in marketing or convex optimization is not required.