Related Posts
Anyone knows about the base for chase PCA?
FCB Health? Thoughts?
What’s everyone doing this weekend 🧐
Hi Guys,
Can anyone refer me to BCG
TIA
Additional Posts in Data and Analytics in Advertising & Marketing
Best resource to learn SQL?
New to Fishbowl?
Download the Fishbowl app to
unlock all discussions on Fishbowl.
unlock all discussions on Fishbowl.
I just input details into DSP and see the AI forecast. Other alternative is to guess the rates and calculate based on past campaigns?!
In your terms, what does accurate look like? Is it being around 95% of what you forecasted? Cause that's kinda hard 90 days out.
Our data science team provides tools for campaign managers. They previously used tools like Meta's Prophet or Uber's Orbit but now we've switched to a a tool developed in-house that has been more consistent to what we needed to optimize ROAS. Can do 85%+ accuracy for most campaigns that don't forgo any structural changes whilst they're running.
It's doable but requieres a fair amount of software engineering. We don't trust the ad platforms forecasts BTW.
Sorry not being to help with that. This was mostly done via experimentation and over time.
Prophet?
I do not understand this question.
Nobody should be forecasting media spend. Name your independent variables and dependent variables and research which models would fit the criteria.
What are you trying to forecast? What is considered accurate? How are you looking to validate?
If your model periods are quarterly, you could easily get 90 dates out, if your periods are daily it might be harder… but usually MMMs do weekly periods, to which forecasting 90 days is trivial.
If you’re straight-up forecasting you could use ARIMAX? That’s quick and easy
I’m sorry, I misread your original post. Thank you for listing out your IVs and DVs.
I don’t think your variables are spot-on.
With such models, we assume a closed system: there isn’t anything else outside your variables that can affect revenue. However, in your list of variables, none accounts for competitor activity. Your revenue is probably more affected by what a competitor did instead of number of sessions. By that logic, Share of Spend (SOS) outweighs straight-up spend. You’d want to look at SOS and SOM instead of these IVs.
It’s also better to pick non-correlated variables. In your IVs, I believe sessions, CVR, AOV, orders would be highly correlated.
What’s the timeframe of your predictions. Like within 10/15 days etc or on daily basis?