If you are using dummy variables for events (control variable) how are you distinguishing between ROAS during event vs ROAS outside of event? For example ROAS during chirstmas vs ROAS rest of year?
Great presentation! I had the opportunity to attend to some live lectures on probabilistic modelling by Bruce Hardie and Peter Fader, who actually developed those models you talked about in the CLV section. I really enjoyed that part! I also have some questions about the Media effectiveness studies (MMM) that you mentioned. Are there any examples/notebooks on the way one can go about modelling time varying coefficients by means of a GP? I believe the PyMC team actually modelled a parameter of the saturation transformation as a time varying coefficient rather than the actual channel coefficient. Is there any guidance about what option may be more desirable? On the topic of time varying coefficients, one point that is often overlooked (although it is both very interesting and quite challenging) is the ability to disentangle long term and short term effects of media channels on sales. One way to do that Is to allow for a time varying intercept. That is traditionally achieved in a DLM/state space model framework by specifying a state equation for the intercept/baseline. Normally that equation includes an AR component and allows for covariates that are assumed to capture long term effect of upper funnel media channels. Any suggestions on how to model a time varying baseline in a bayesian framework? Would looking at GPs again be a solution? Thanks a lot!
+1 In my experience as well I've used Dynamic Linear Modelling with Kalman filter to get time varying coefficients as well as using UCM to get a dynamic base. Looking forward in knowing how this can be implemented in Bayesian framework? Btw. Are you familiar with any free/open source libraries to create Bayesian Belief network with our MMM data? Wanted to check on the possibilities of exploring network analysis as well with the libraries...
Really nice session. Can you please share materials around CLV & Causal inference?? Just out of curiosity I am also interested to know what type of solutions are you offering in marketing Analytics space? (Other than MMM)
Amazing presentation. Thanks, Juan.
Great stuff, gang. Thanks so much for sharing these approaches and for the great libraries.
If you are using dummy variables for events (control variable) how are you distinguishing between ROAS during event vs ROAS outside of event? For example ROAS during chirstmas vs ROAS rest of year?
Hi!! Following up on this question.
That was an excellent presentation! Thank you.
Great presentation! I had the opportunity to attend to some live lectures on probabilistic modelling by Bruce Hardie and Peter Fader, who actually developed those models you talked about in the CLV section. I really enjoyed that part! I also have some questions about the Media effectiveness studies (MMM) that you mentioned. Are there any examples/notebooks on the way one can go about modelling time varying coefficients by means of a GP? I believe the PyMC team actually modelled a parameter of the saturation transformation as a time varying coefficient rather than the actual channel coefficient. Is there any guidance about what option may be more desirable? On the topic of time varying coefficients, one point that is often overlooked (although it is both very interesting and quite challenging) is the ability to disentangle long term and short term effects of media channels on sales. One way to do that Is to allow for a time varying intercept. That is traditionally achieved in a DLM/state space model framework by specifying a state equation for the intercept/baseline. Normally that equation includes an AR component and allows for covariates that are assumed to capture long term effect of upper funnel media channels. Any suggestions on how to model a time varying baseline in a bayesian framework? Would looking at GPs again be a solution? Thanks a lot!
+1
In my experience as well I've used Dynamic Linear Modelling with Kalman filter to get time varying coefficients as well as using UCM to get a dynamic base.
Looking forward in knowing how this can be implemented in Bayesian framework?
Btw. Are you familiar with any free/open source libraries to create Bayesian Belief network with our MMM data? Wanted to check on the possibilities of exploring network analysis as well with the libraries...
Really nice session. Can you please share materials around CLV & Causal inference??
Just out of curiosity I am also interested to know what type of solutions are you offering in marketing Analytics space? (Other than MMM)