Multi-device attribution is a bumpy road but mobile offers a bright side

Knowing how your customers came to book with you is a complicated business and it’s not getting any easier

In a multi-device world it is imperative to understand where the clicks are coming from, but it isn’t easy

To understand how customers are using their devices, marketers are leveraging the available data to deliver a superior customer experience. This can be done in a number of ways. By evaluating the most searched pages, exit pages and events/clicks, for example it is possible to understand how a mobile site is being explored.

EyeforTravel’s Ritesh Gupta talks to Jonathan Isernhagen, director, marketing analysis, Travelocity, about the challenges of attribution in the multi-device environment.

EFT: How challenging is it to assess the performance of an online travel brand in the multi-device world?

JI: This is the greatest obstacle to good attribution. Before you even think about modeling, you need to de-duplicate your records so that 12K shoppers that visit your site via three different devices look more like 12K than 36K shoppers. This is accomplished in two ways: 

1) By marshalling your own internal e-mail and checkout point of purchase data: ‘This desktop cookie and this smart phone cookie both correspond to this user ID’, and ‘the messages addressed to this e-mail address were opened on this tablet and this mobile device….’

2) There are third parties who have their pixels across dozens of sites and are working to build profiles, and by virtue of their networks they have better odds of figuring out which 20 cookies roll back to one person. They are more immune to the effects of rampant cookie clearing than those of us with individual sites.   

EFT: What about attributing mobile-related marketing expenditure? 

JI: It’s tougher, because not only do you have to have to synthesise a single image of a consumer coming at you via multiple channels and devices, but since the call-to-action with mobile devices literally results in a call, you have to be able to track conversions in the call centre and tie them to their click/impression stream. You pull everything available into the attribution model, find your ROI by channel and point-of-sale, and adjust spend accordingly. On the bright side, consumers are much more likely to remain logged in on mobile browsers, so the shopping data can be richer. 

EFT: How has attribution evolved in the last 12 months?

JI: The increase in the number of devices and channels has only made it more complicated. The only thing that’s ‘simpler’ is that understanding of the need for algorithmic attribution — what it is and why it’s needed — has become far more widespread, so if you’re an attribution evangelist, there’s a chorus of voices now backing your message.  

EFT: Where is attribution headed next?

JI: Getting the attribution model output to the site monitoring/dashboarding tools so it can be used by frontline decision makers is the challenge of the moment. Google Analytics has the capability to show you which online channels get how much credit under different arbitrary (ie. first click, last click and so on) attribution schemes, but they haven’t yet woven their algorithmic attribution directly into the interface. Visual IQ and Adobe can each do this but it’s an expensive custom integration. So the analysis consumers constantly have to toggle between the site monitoring tool and the model output. “According to SiteCat, our brand advertising drove 60% of the bookings and is ROI positive, but the attribution model says that credit really goes to metasearch.”  The sooner algorithmic attribution is a standard feature of site monitoring tools, the better off we all will be.

EFT: It can’t be easy to understand the role of each device. How do you it?

JI: According to the new IBM Benchmark survey, on Black Friday and Cyber Monday, smartphones drove 19.7% of all online traffic compared to 11.5% driven from tablets, but only 5.5% of all online sales vs. 11.7% for tablets. Modelling clicks separately by device captures their respective ROIs, but won’t necessarily tell you that say a TripAdvisor click is 3x more effective when followed by an e-mail message. Visual IQ claims that their scenario-planning interface handles the synergistic or antagonistic effects, which channel exposures have on each other. 

This whole topic is the essence of understanding customers by knowing their behaviour. If, for some odd reason, a meta impression delivered on a tablet one week before a given purchase occasion is the magic bullet that closes the sale, the model’s going to pick up on that connection.   

EFT: What are the major hurdles in knowing where each impression or click has come from?

JI: It has less to do with the devices than where the impressions are coming from. Facebook will give you ad impressions but not newsfeed impressions, or at least not at a level of detail, which supports modeling. 

Twitter provides an API, but you see the data by Twitter handle, not IP address, and you’re just seeing who’s publishing, not who views. 

Beyond the challenge of obtaining the impression data, is just the magnitude of the impression data. You can house click data in a relational database, but impressions data at normal commercial site traffic volumes is a ‘big data’ problem in the truest sense.

Jonathan Isernhagen, director, marketing analysis, Travelocity is scheduled to speak at the forthcoming Social Media & Mobile Strategies for Travel 2014 conference. It is scheduled to take place at Hotel Nikko, San Francisco (March 17-18).

Related Reads

comments powered by Disqus