Robyn with SEM

Inspiration

Many ad players like Marketers and Advertisers are strongly interested in Marketing Mix Model to evaluate the overall structure of the performance of channels, but sometimes it is not good to evaluate the effect of channels equally. They serve ads not only for leading directly to conversion but also for ads that have quite important roles in driving awareness, their consideration or search, and so on. Thus it seems better to add a model that can also consider these indirect effects.

What it does

In the development I worked on this time, we can align a model based on structural equation modeling (SEM), that can consider the causal structure between channels with existing model implementation.

How we built it

• Added one python package "semopy" based script (named "sem_model.py" in the same directory as all other R code.)
• semopy link: https://semopy.com/index.html ## Challenges we ran into
• Unstable with running (sometimes stop in "robyn_run" without any error message)
• More convenient if the logic to explore the optimal causal structure will be introduced (I have one idea to utilize "causalnex", python-based package) ## Accomplishments that we're proud of
• Improve flexibility to apply MMM practically (i sometimes got a comment this approach seems not good because Robyn cannot consider these indirect effects) ## What we learned
• Can extend more variety of model framework to align with current model script ## What's next for test
• Improve stability to run (clarify problem)
• Can view semplot

Built With

• python
• r