วีดีโอ

Hello! As of the time of recording the series (around 2 years ago now) and I believe still until today Robyn does not support categorical variable. What you'd need to do is eventually build several different models, one per each segment (so 5 models in your specific case). Same input variables, ideally same configuration, just changing the output variable.

​@cassandra4533 thank you!

Hello! Glad you liked this. Not sure about Tableau itself but with a bit of work we managed to integrated this with different front-ends. Typically saving the outputs to a database or some sort of GoogleSheets and then reading from those dinamycally.

Hey Praveen, we don't have one yet but if you look through our channel you'll find more content on how to build a regression-based MMM either in Excel, R or Python

Unfortunately we don't have a decomposition over time tutorial yet. but If you add new columns next to your dataset and multiply each transformed media variable (after adstock and diminishing returns) by its coefficient (from the linear regression) you can get the contribution of each variable and analyze it in a decomposition chart

This is a smart question When there are high lag effects between spend and conversion, we would need to change the adstock transformation to one that takes into account the lag effect One you can use is called “weibull adstock” With that we can model the delayed/lag effect over time of our campaigns and accurately account for long sales cycle in B2B Weibull adstock is really difficult to use it in excel, I suggest to use either Cassandra or Robyn to model your media mix if that is the scenario

Hei mate, can you be specific on what parts I was not clear enough? In this way I'll be sure to create a video that clarifies your doubts

We have a complete tutorial series on building an MMM, starting from the dataset to be used. Here's the video: th-cam.com/video/HE8qZSzPWfc/w-d-xo.html&ab_channel=Cassandra In short: you should include all the variables that have an impact on your output KPI.

@@cassandra4533 Thank you. In this example you only used paid channels, no direct or organic traffic. Was it on purpose or only an example?

@@aaakmm1785 Only an example

Hello there, Google's refers mostly to traditional MMM that have been using a wrong methodology for adstock. Traditional MMM uses geometric adstock effect of media campaigns. This methodology does not take into account the delayed effect between the spend and the effect on sales generated. This leads to overestimating Lower funnel campaigns rathern than upperfunnel's. The easiest way to solve this problem is by leveraging weibull adstock, that takes into account that. Nested MMMs are incredibly difficult to replicate, even more difficult to become actionable and useful only on handful of situations. As soon as Meridian comes out, I'll do a video about it, and if it has a feature for nested MMMs I'll be happy to show it.

hello, please reach out on linkedin: www.linkedin.com/in/gabriele-franco-hybrida/

@@cassandra4533 please accept the request

Hey Adityia, we don't have that planned at the moment but we'll take the feedback inside and consider as per the next contents. I'd suggest to look out for Google Meridian anyway, which is the newer and official Google's library coming out very soon.

@@cassandra4533 Thank you!

Hey Tarek, Intercept sums up with other variables to be considered as those sales that you'd do anyway no matter of your marketing investments. More info here facebookexperimental.github.io/Robyn/docs/features#model-onepager

We're working to automate the GeoExperiments with the platform. But for now, you'll need to edit your campaign or launch it, based on the Geo Experiment design

Hey, we haven't really been searching for it but not that we are aware of

Hey, glad you liked the series :) The most simplistic explanation you can use is: Intercept represents all those sales you'd do anyway, regardless of your paid and organic marketing activities. Contribution of the intercept varies a lot from business to business and it's much influenced by the amount of historical data as well as the length of the modeling windows selected. So if you feel like your intercept is not correctly attributed you might want to try and add more data as well as change the windows selected.

Hey Guido, with newer versions of Robyn (compared to this video) we do have Confidence Intervals as well as an output. Typically we do look at a mix of things to ensure the model makes sense statistically wise including general KPIs of the model, Confidence Intervals, Adstocks and Contribution of Baseline vs Other Factors. If that looks good from a statistical point of view we do compare that with the business knowledge of the client itself which tipically means comparing with their known CPO/ROI up until that point just to make sure it's somewhere realistic (still assessing uncertainty for channels where the C.I. are too wide). We're not expecting ROI/CPO to match with their "known" ones (tipically cookie-based) obviously. From there it's a matter of both understanding why there are differences (and using data to validate hypothesis) and relying on experimentation as the definitive validation.

​@@cassandra4533 Super clear mate! Thanks! 🙌

Hey Lewis, we didn't produce any additional episode for this serie in the end. We might get back to it in the near future but nothing confirmed yet.

Hey, great question, I'll try to simplify this to its core. When you run a model in Robyn you're getting a set of optimal models and that is because statistically wise it's impossible to pick only 1 model and say that's the overall best compared to all others. This plus the fact all those models will slightly vary from each another (one might favor more Variable 1 and another might favore more Variable 3) as of today still require a human input. That is for two reasons: 1) Business-wise you want a model that your stakeholders will trust, if you present something that's far away from their current knowledge right from their start they'll just go "this doesn't make sense" and won't even consider it - On the other hand if you build trust first in the model with something close to their knowledge then you've room to make them switch overtime 2) Statistically-wise there are many different ways to achieve an accurate model, that's why we use the RSSD as well in order to exclude those less "realistic" options already but still you might get some models that are a bit off on the attribution Hope this answers your question :)

​@@cassandra4533 Hey thx for your response. The deviation between the models seems so big I'm wondering if the dangers of using this aren't greater than benefits. What would have to happen for us to be able to objectively pick the best one?

@@meheh002 Though to say with no context at all unfortunately. I'd probably suggest to start from the data, ensure those are complete and comprehensive and you did actually include all the relevat variables. Once that is out of the way I'd focus on the model. For cases where's there's lots of uncertainty dig deeper on why that's really happening, what the cause is. Often it could be driven by something like Multicollinearity, once you identify the cause you can define the solution. It could be an Incrementality Experiment for instance using GeoLift.

Hey Guido in general I'd recommend not to go too granular at a Campaign leve but rather to stay somewhere in the middle like at a Campaign Type level (e.g. P.Max, Video, Search ...). Then by default you'll get outputs at the same granularity of your inputs and hence there's no automated aggregation but that would be pretty easy to do it yourself as you'll get all the detailed output directly from Robyn and it will just be a matter of displaying them out as you like.

Awesome, sounds good. Thanks a lot mate!

Hello! This is not shown in this introductory tutorial but we talk about it in other videos in our channel like this one th-cam.com/video/PU9zuR1axUA/w-d-xo.html&ab_channel=Cassandra In short you do apply additional transformations to your data before running the model itself. For the time decay and lagged effect we specifically recommend an Adstock transformation called Weibull PDF which can handle both of them.

Hello! Glad you liked the video, you can find the dataset and the notebook itself linked in the Description!

Hello@@cassandra4533 , that's what I'm saying. The data set linked in the description is different than what is used for the OLS demo in the video. For example, the linked data set is daily granularity with 595 observations. The column headers are also quite different.

@@MN-UnicornUnderstood and apologies for the incovenience! I've checked internally but as the video is more than 1 year old we couldn't find the original one unfortunately. There's this interesting library from facebook however to generate MMM datasets for tests if that might come in hand: github.com/facebookexperimental/siMMMulator

Hello Luna, we're really happy you liked our course! As this is an introductionary series we used a simply Adstock transformation which doesn't handle Lagged Effect. However in Cassandra as well as other production-level MMMs other functions such as Weibull PDF are used to handle such behaviour. You can find more information on this other playlist in our channel: th-cam.com/video/PU9zuR1axUA/w-d-xo.html&ab_channel=Cassandra

Hello, in that case we'd suggest to go back and start from the input data. An RSQ that low could be caused by a number of things such as: 1) Incorrect input data 2) Incomplete input data 3) Completely missing relevant input variables that affect your output variable 4) Not enough historical data, if it's a new business with low spend and revenue volumes you might simply not be able to use MMM yet

@@cassandra4533 Yes, that's what is happening in my case. Thank you!

Ciao Marcello, sì e viene fatto nella fase di configurazione del modello tramite gli Iperparametri di Scale e Shape che gestiscono la trasformazione chiamata Adstock. Questa non solo descrive, se presente, l'effetto ritardato di un'attività di marketing ma anche il suo effetto nel tempo. Più info nell'episodio #3 th-cam.com/video/ohe3Gt0Qu38/w-d-xo.html

@@cassandra4533 grazie per il chiarimento ;)

Hello Tomas, we would suggest asking directly in Lightweight's GitHub.

@@cassandra4533 thanks for your response. Aslo thanks for all the MMM videos, they've really helped me grasp the logic behind it!

Hello, ROIs are generated by dividing the amount of Revenue attributed to each channel (by the model) by the total budget spent for each channel.

Hello, I'd first check on the suggestion from the error itself about git. Then if the issue persist you might want to report it on the Robyn's GitHub Issues: github.com/facebookexperimental/Robyn/issues Or eventually in the Facebook Group: facebook.com/groups/robynmmm Where you can find someone else that might have had this issue already.

These are dynamic values. You need to try the model multiple times by checking RMSE to get the optimized Alpha and Beta