Multi - Channel Attribution In Walled - Gardens Era

By Giorgio Suighi, Regional Head of Data and Analytics, GroupM

Multi-Channel Attribution In Walled-Gardens Era2018 was a year when the main players as Google, Facebook, and Amazon who are representing  approx. 65% of the total digital advertising investment, strengthened their product strategies through isolation.

They did this mainly to prevent other actors in the ecosystem (adservers, DSPs, DCO, DMPs, SSPs) to connect to their systems, generating more than a headache for the professionals working on Multichannel Attribution, or even in simple channel performance analysis

The restrictions that were applied were mostly in terms of 3PAT (3rd party Ad Tracking) specifically on View Tracking (3rd party click trackers are still allowed), ID management in reports, and so on, with most effect in YouTube and Facebook.

In this scenario every major player started to push for his own Attribution solutions: • Facebook came out with an utm-based solution to try to have a local ‘attribution’, but the result is a product based on the old rule-based mechanism, with the limit that every activity outside Facebook universe need to be labelled and imported into Facebook as a ‘Platform connection’.

•Google pushed with his 360 platform, which means getting ‘stuck’ only to google universe which also limits our possibilities and let Google give us the results of their own performance (IMHO a questionable strategy)

•Even trying alternative paths as ADH will make Google needed for storage to load non-google data in their cloud (even if is a separated one) and to configure the matching tables to connect those 2 universes.

•Sizmek did an internal attribution too, still rule-based models, and all the 3PAT we’ve already described earlier.

"The complexity of human interaction with media channels makes it impossible to infer the motivations/ orientation of the customers during the purchase approach"

Rule-Based Approach The challenge became then how to get a real multi-channel attribution. First, let’s make clear that no approach or model is perfect. If the media mix is relatively simple, solutions, and forget about the Multi-channel complexity, if that’s not our situation, then something more is needed. we can opt for one of the previous solutions, and forget about the Multi-channel complexity, if that’s not our situation, then something more is needed.

As we speak, the biggest players are leveraging rule-based methods that have some evident limits. I’ll list only a few of them to give you an idea:

First Click: Useful if the focus resides on few channels (erg only Google or FB Ads), but when you have multiple channels involved is not the best solution. The first channel can have an “activation” function, but the others could be more important in terms of lead generation

Last Click: It considers marginal the other channels that may be played an essential role in your brand awareness and discovery.

Linear: the credit is divided equally among all the channels. I think no explanation is necessary here based on my previous notes

Position Based (u): As you know in this model first and last channel get the 40% of credit for the conversion, while the rest of the 20% is equally split among the channels that come in between, limiting the importance the first channel might have.

Time Decay: here the touch points that are closer to the conversion get more credit instead of those through which the conversion path started, and gives the least important focus on earlier channels that could have helped in creating awareness.

Algorithmic Approach

If the rule-based models aren’t enough, then the only path is through algorithmic modeling. We will take two methods in consideration: Shapley value and Hidden Markov Chains

Shapley: Shapley value derives from game theory and starts with the hypothesis that all actors (channels in our case) are working towards the same goal, while the ecosystem moves in a total unstructured way with the channels often challenging each other to get to the predefined KPI.

Shapley value states that a value of (e.g.) a Facebook Ads ad will be the same when followed by a paid search ad and a direct visit, but we know that every potential customer cannot be influenced from the same channel order, in the same way, every time.

So, even if better than rule-based methods, even Shapley have is own limits.

Hidden Markov Chains

With Markov Chains model we can represent every customer journey in terms of a sequence of channels/ touchpoints as a chain with the probability of transition between each of those touch points. That allows to know how a visitor’s propensity to convert changes as he/ she is exposed to different marketing channels over time.

Markov method allows us a more precise analysis of the channel’s contribution thanks to the Removal Effect principle, which states that to find the contribution of each channel in the customer journey, we just need to remove every single channel and see how many conversions are happening without that single touchpoint being in the user journey.

This approach makes a huge difference with (e.g.) Shapley value because it gets under consideration how the interaction between different touchpoints influences the customer to get to the conversion (or the desired KPI).

The complexity of human interaction with media channels (sometimes even in an unconscious way), makes it impossible to infer the motivations/orientation of the customers during the purchase approach.

Let’s keep in mind that our goal is to get as near as possible to the reality, there is always a degree of error.

Don't Miss ( 1-5 of 25 )