Attribution modeling. Anyone who’s had a professional conversation with me in the last four years has probably heard me harp on the subject. For those who are unfamiliar, the concept of attribution modeling is actually pretty simple.
First, take every single data source you can think of from your marketing mix and centralize it. Second, track conversions against each data source to determine every time a customer engaged with one. Third, analyze those touchpoints to determine what your actual conversion funnel looks like from awareness to sale.
Okay, so maybe it’s not that simple after all. In reality it’s a complex web and giant undertaking to integrate all of that data in such a way where you can actually map conversions to the sources. Then there’s the question of what model to look at – linear, time-decay, upper funnel or Bayesian. Attribution companies aren’t cheap, and they don’t employ the best statisticians and analysts in the world for nothing, right?
For brands jumping into the attribution game it can be a daunting undertaking. Even after you’ve figured out how to integrate all of those data sources and gone through the process of set-up, there are still questions that remain. Now what? What am I going to see? Are my assumptions all wrong? What should I look at, and why?
Don’t fret. Attribution can be a major game-changer. As with anything new, the most difficult step is just getting started. For clients and brands taking that leap, here are the most important things I suggest you keep in mind:
Think Big.
Everyone talks about data, data, data. With the overwhelming volume of data, the old adage “analysis paralysis” is hard to avoid when you first begin scrutinizing your models. Ultimately you shouldn’t get too caught up in the details. Start with a high-level plan of the metrics you want to analyze, and stick with it. Ask yourself what trends can have the greatest impact on your end sales, and use that to guide your initial analysis. Once you’re comfortable these questions are addressed, then dig into the minutiae.
Forget About Last Click.
This is probably the most challenging aspect of attribution. 99% of digital marketers think in terms of converters. In fact, the entire industry is driven by tracking last-touch – floodlight tags, Google Analytics, the Facebook pixel – ALL of these ONLY focus on last-touch. In order to fully understand the value of attribution modeling, you have to virtually eradicate this line of thinking and instead think about a true marketing funnel. Eventually, understanding each portion of the funnel allows specific strategies to be created around origination, frequency and lower-funnel conversion tactics.
Understand Lag.
Conversions simply don’t happen in a vacuum. As much as we might like to think this would be the case, single-session conversions are an extreme rarity. For this reason, understanding lag time is critical in order to effectively evaluate your various models. Until lag time stabilizes, the data is virtually worthless. For some clients, and especially those with more traditional brand drivers like TV in the media mix, lag times can average months. You can’t effectively evaluate 30 days of data when your customers take 90 from origination to conversion.
Match the Hatch.
I’ve had numerous clients ask the question, “How do I use the data to inform my marketing mix?” It’s a tough question to answer because every client’s goals are different, but generally speaking, you should try to effectively match the model with your efforts. Specifically for paid media, we focus optimization strategies on two things – finding originating media sources that have shorter lag times to conversion and finding placements driving conversion through lower frequency. This is exactly the opposite of most optimization strategies where emphasis is placed on spending toward last-touch tactics. With this in mind, I encourage you to re-read point number two above. You HAVE to forget about last click and think about originating and shortening a purchase funnel.
Always Question the Data.
No matter how perfect your model may seem, there will always be anomalies. For example, we have a client with heavy exposure in prime-time television. By the nature of their product category, they also see heavy web traffic in the evenings. Having TV spots coinciding with natural lifts in web traffic means the attribution models suggest that TV is the prime source of this traffic. Although some of this may be true, a more granular analysis tells us that the impact isn’t quite as strong as the math suggests.
So if you’re thinking about jumping into the attribution game, don’t be afraid, and get in touch with us. It truly is a game changer. Happy modeling everyone!
