So now that you’ve got the basics down, it’s time to take a deeper dive into user acquisition and attribution as it stands today. To do that, we also need to look more closely at advertising.
In my last article, I discussed data-driven models for advertising attribution, and why the industry is leaning more toward these models (versus models like last-click attribution). To recap, when you use data to determine the exact sources of attribution, you get a better sense of your true ROI, and you can better optimize your ad spend. Thus, you’re going to need both an attribution company to get the publisher tags, as well as a post-install analytics company to connect the users with their in-game behavior. Make sure that your attribution and analytics companies are compatible. They should be, but not all are.
Now, the problem with models that aren’t data-driven lies in accountability: Namely, there is none. As a data scientist, with an emphasis on the science, I say there has to be a better way. When we’re divvying up the credit based on theory (like the theory that the majority of credit should go to the last-clicked source), rather than deriving insights based on observable patterns, there’s no way to justify your spend on each source.
Another problem with attribution models as they stand today is the tendency for these models to sap the creativity out of the process. I know it sounds strange. How would relying more on data put the creative back in advertising?
Let’s look at an example. You’re working with a linear attribution model, which divides credit up equally among all ad sources. This means that each source is equally important, and you need to make sure every source is carrying your audience member down to that eventual last click before they buy. Your advertising becomes stiff, optimized, even formulaic. You have ads on every platform, and you need to make sure that every ad is hitting exactly the right points, cramming every key message into each one -- because you never know at what point the viewer is coming into the funnel.
If you’re treating every user the same way, there also isn’t room to fine-tune advertising messaging. Good messages and creative ideas are often lost in the noise of competing and incongruent messages -- or worse, they’re targeted to the wrong demographic. First contact messaging, for example, is very different than advocacy messaging. Let’s approach this in the same way developers approach a tutorial or new user experience. Do you cram all the information about the game down users’ throats all at once? Of course not -- you take compelling and informative, yet light, first steps into a complex world.
To achieve this, though, we need a data-driven perspective. With data-driven models, you can see exactly what source is bringing consumers in, which source they’re passing over, which source finally drives them to buy. And with all this information, you can allocate your resources accordingly. Banner ads aren’t working, so let’s not waste our design resources with those. We know our audience is responding well to video ads, so let’s get creative with those -- really grab our viewers’ attention. A good attribution model eliminates the source or placement of the ad as the reason people aren’t buying, giving creatives the freedom to experiment with messaging.
So a new attribution model would benefit creatives, but what about the publishers? These are the sites that price and sell the ads. They rely on attribution models to figure out how to value their ad, but that has gotten more complicated in recent years with the introduction of real-time bidding. Instead of having an ad in place when people load the site, real-time bidding turns an advertising space into an auction floor as a page loads.
The publisher is the auction master, setting a price and giving information on the user whose page is loading. Advertisers are the bidders, placing their bids for who gets their message in front of that user. This helps with targeting. In the pre-real-time bidding world, if you were trying to sell a sports video game, you might place your ads on sites that had to do with sports: those that your target player would likely frequent. Now, when a target user pops up on a site completely unrelated to sports, you can jump in and place your ad in front of them. Sure, you might have to outbid a few other companies, but it’s still more cost-effective than an ad wash.
With this new technology, publishers are living in a brave new world, trying to figure out the best way to price ads and still make a good amount of money. What’s good for ad publishers is the ambiguity that comes with current attribution models. This isn’t to make them the bad guys: They’re trying to run a business and maximize their profits, just like everyone else.
However, it does make things difficult for advertisers. Right now, they try to justify their spend by using ROI metrics. That’s good, but it’s not the most accurate because of the lack of data. With more data, we can turn to a better metric: total lifetime value (LTV). Predictive analytics programs can calculate this just like in games (see my series on prediction for more information on how that works). Based on their spending patterns, these programs can calculate the value of users based on their predicted spend, and marketers could target their programs accordingly.
This adds another layer to real-time bidding. If you only go after the most valuable players, you know exactly how much to spend. But what’s the best way to find them, and how do you know what way to engage them? My final article of the series will explore this concept more.