Meet Robyn, Facebook’s Experimental Automated Marketing Mix Modelling Code

Facebook has released a new experimental measurement approach, Robyn 3.0 – a take on marketing/ media mix modelling. Here, Qassem Naim takes us through a few of its elements that are designed to help improve your current marketing analytics across all media investments.

No not PBA (people-based attribution), or MTA (multitouch attribution), or the strongly recommended server-side CAPI (Conversions API), or the offline conversion API, of the SDKs, or any other type of tracking related direct measurement approach, Robyn 3.0 is a new take on old school Marketing Mix Modelling aka Media Mix Modelling (MMM) aka econometrics.

The traditional media industry and brand agency favourite indirect measurement approach to measuring stuff that’s hard to measure.

And fair enough. Facebook’s attribution has gotten significantly harder to measure lately, with all the privacy changes across the digital ecosystem. So, it only makes sense they’d come back around to the old school method of understanding media effectiveness.

With a few upgrades of their own of course.

While still reliant upon time-series analysis and statistics, Robyn has a few key elements that are new and interesting. The Facebook Marketing Science team has already created a great quick start guide, but here I’ll take you briefly through some of the major talking points:

Robyn can learn from “ground truths”

Digital attribution can still have a role in the Robyn 3.0 world.

Through experimentation and what they’ve called ground truths, you can refine and guide model selection based on these more tangible outcomes. So, treatment and control groups aren’t going anywhere either.

This opens the door to all kinds of experimentation, including creative or brand effectiveness. 

They’ve taken steps to mitigate analyst bias and speed things up

As they put it: “Building MMM manually is a time-consuming process that involves subjective decisions based on analysts’ modelling experience and trial and error over hundreds of iterations.”

This has also made it historically expensive, particularly in small markets and so slow it’s often out of date, especially in pandemic times – keeping networks and agencies racing to keep up and develop automated solutions of their own.

Delivering total media optimisation, not just Facebook

As with all MMM, Robyn optimises your total media investments, on and offline.

Expanding beyond the walled garden once again, it will measure the effectiveness of your total media mix and provide budget optimisations between channels (yes even TV).

Robyn pays particular attention to organic media 

They’ve accommodated organic metrics in the modelling process, allowing for blogs, email, organic social, and web activity to be considered. This as Facebook expects them to decay and saturate in a similar way to paid activity.

Probably the best part about Robyn is that it’s (relatively) easy. And open-source

Facebook’s version of MMM should be much more accessible to businesses than traditional econometric modelling and could disrupt the current marketing analytics paradigm if it proves effective.

It calls for experimentation, so we’ve begun testing Robyn and putting it up against other modelling techniques.

I’m certainly excited to see how it performs.  

Avatar photo

About Qassem Naim

Qassem Naim is Head of Data & Tech at FCB New Zealand.

Leave a Reply

Your email address will not be published. Required fields are marked *