Robyn
Robyn copied to clipboard
Attribution for Brand Direct
I am working with a dataset of 2.5 yrs history with 6 media channels and 3 context variables. KPI is number of new customers. I have a question about understanding the output from Robyn model. The brand direct channel(search campaigns that target brand keywords) is being attributed the highest, on one side it seems intuitive since this channel is not expensive/super low funnel and is correlated with the target. But I also think the other channels are misattributed (underestimated). How can I change the model setup/feature engg to solve this issue?
I am looking for suggestions to improve my modeling and hopefully move towards capturing better attribution for all the channels.
Thanks!
Have you checked the analyst guide to MMM? https://facebookexperimental.github.io/Robyn/docs/analysts-guide-to-MMM/ and specifically the feature to calibrate MMM with ground truth (usually experiments): https://facebookexperimental.github.io/Robyn/docs/features/#calibration-with-experiments
MMM is modelling so you won't be able to make sure you are misattributing or not, so either use existing (from the past) experiments or design some in the future and use the results to calibrate your MMMs.
On Fri, Jul 21, 2023 at 6:18 AM ar-asur @.***> wrote:
I am working with a dataset of 2.5 yrs history with 6 media channels and 3 context variables. KPI is number of new customers. I have a question about understanding the output from Robyn model. The brand direct channel is being attributed the highest, on one side it seems obvious since this channel is not expensive/super low funnel and is correlated with the target. But I also think the other channels are misattributed (underestimated). How can I change the model setup/feature engg to solve this issue?
I am looking for suggestions to improve my modeling and hopefully move towards a better attribution for the channels.
Thanks!
— Reply to this email directly, view it on GitHub https://github.com/facebookexperimental/Robyn/issues/789, or unsubscribe https://github.com/notifications/unsubscribe-auth/ARF52KAIU2MUGRMTO3V5GLDXRH7HTANCNFSM6AAAAAA2SJRBSE . You are receiving this because you are subscribed to this thread.Message ID: @.***>
Try starting from multiple different seeds. That way maybe you will converge to different local optima that makes more sense to you.