×

Dan Robinson, Artemis Manager, MPG Media Contacts 'DFA Attribution Modelling – Quick and easy attribution solution, but does it pose more questions than it answers?'

Over the last 12 months the attribution debate within digital marketing has slowly moved from whether or not advertisers should be doing attribution beyond the last click, to what kind of attribution they should be doing.

Last week saw the launch of DoubleClick’s Attribution Modelling tool within the beta section of DFA reporting. For those who are unaware, this new tool sits within the Multi-Channel Funnels section of the interface and allows a campaign analyst to quickly and easily apply a variety of fixed attribution models to an advertiser’s digital media data and observe how each one affects the campaign’s outcomes and KPIs.

Sounds great, and it is. Sort of. Advertisers and agencies have long had the ability to look at the effect of these sorts of fixed models for themselves in Excel using their raw campaign data, but the ability to run them simply and easily within the DFA reporting interface certainly makes this process less arduous.

However, what can these fixed models actually tell us about the digital media that we’re planning and buying though, and which model in particular is the most important to look at?

Well, at the moment there are six different fixed models available within the DFA tool, some of which allow for customisation to some degree. Each of these models has a different effect on a campaign’s results.

To look at these effects, let’s take a fake example. Jodie is looking for a new electricity company. She begins by doing a Google search for a generic electricity keyword and clicks on a PPC link for our example electricity supplier, let’s call them ElecX, and browses their site.

She leaves to think about this and a few days later, while web browsing, she gets targeted based on her previous online activity and served some ElecX display impressions. She is reminded of her search for a new supplier and visits an affiliate site so she can compare the prices offered.

She clicks on the ElecX link after comparing the prices she’s offered. They weren’t the cheapest on the market, but her previous interaction gives her confidence in their brand and so she decides to make them her new electricity supplier. She browses some other sites while she gets her current supplier information together and then does one final Google search for ‘ElecX’ and clicks on the PPC link and becomes their newest customer. Success! But who gets the credit?

This depends on which model you choose to use. Jodie’s conversion journey looks like this:

It looks fairly typical, but the six fixed models offered by the DFA Attribution Modelling tool offer varying views of which activity should be assigned conversion credit.

So, the changes to a campaign that would be made from looking at each of these models vary, and this is only looking at one conversion! Applying any one of these models at random across all of a client’s activity would result in significantly different results than if any other one were applied.

We’ve taken on one real client and looked at their digital activity across a month, the model that assigns the highest number of conversions to PPC makes it appear more than 400% better than the one that assigns the lowest.

Of course there are manual changes that can be applied to these models but we are always left with the question of which one to trust. Fixed attribution models, while they will always be quick and easy to run, and DFA’s latest update offers the quickest and easiest way yet, will always suffer from the same problem: they cannot infer causality.

The fact that a piece of activity was served to a customer who went on to convert does not even guarantee that it was SEEN, let alone that it made a difference and helped to drive that conversion.

The only way to answer the causality question is to include non-converting users into the attribution analysis.

Comparing the activity served to both groups of users, and understanding which pieces converting users are more likely to have been served, is the only way to get close to understanding what worked, rather than just what was seen. That is where algorithmic attribution solutions become so important.

So, the DFA Attribution Modelling tool can provide a quick and easy way of attributing conversion credit beyond the last click. However, each model offers a very different picture of media performance and there is no way to tell which one (if any) to place faith in.