Back to all discussions

  •   over 3 years ago

Channel-KPI causal structure in MMM

Now I’m working on implementing MMM, and I realized that sometimes we have a causal structure depending on the granularity of channel data we can get.
In that case, I wonder if the evaluation of the impact that advertising channels affect on KPIs can be biased because some advertisements do not directly affect clients’ KPIs.
I am considering one of the solutions to improve evaluation in consideration of the problem that can be plugged into modeling is DAG analysis to identify the causal structure and estimate the effect on KPI including mediation.
Does anyone consider that area as well and if so, it would be great if you could share your thoughts and discuss more.

  • 3 comments

  • Manager   •   over 3 years ago

    Hi Ryo, I am David Choe, one of the organizer of this Hackathon. Your suggestion is very interesting.
    Do you mean adding a casual structure between advertising channels? (e.g. Channel A would have impact on Channel B, and Channel B is more direct to the purchase)
    In this case, I think we need to be careful on the assumptions as these days consumer journey is very complex and non-linear. However, if you have specific methodology to quantify this and incorporate into modeling, please submit your idea and judges would be able to give better feedbacks and evaluations.

    If you are looking for team mates for working together, please refer to the recent update about collaboration and this Facebook post(I guess you already have seen) - https://www.facebook.com/groups/robynmmm/permalink/1297207077714089/

  • Manager   •   over 3 years ago

    Hi, Ryo, thanks for flagging this interesting topic here. I'm Ayae, organizing this hackathon with David.

    As you mentioned, that would be valuable to understand the causal structure between marketing channels for both prediction and explanation of your model. Perhaps the context is a little different, but there have been case studies in the past where intermediate KPIs have been captured in SEM(Structural Equation Modeling) while evaluating the channel in MMM. I guess there is a great opportunity for Robyn in this area.

    As you may already know, I found this paper "Marketing Mix Modeling Using PLS-SEM, Bootstrapping the Model Coefficients" - https://www.mdpi.com/2227-7390/9/15/1832, which address similar challenges in MMM. I hope this is helpful for you.

    Besides that, Robyn Facebook group (https://www.facebook.com/groups/954715125296621) has +1.3K members and you could get more attention for this topic if you want.
    Thanks again for your attendance!

  •   •   over 3 years ago

    Hi David and Ayae san, thank you for your kind reply. And your comments are definitely helpful. As Ayae-san introduced, SEM should be one of the most effective solutions from my viewpoint as well. I think there are many things to consider, such as how to assume the structure, but I think there is room to align with the existing MMM framework as a first step.

    Thank you for your introduction to the community as well. I will check it.

Comments are closed.