As a partner to the world’s most popular apps, we’ve often heard the need from acquisition marketing teams to be able to track the performance of non-paid acquisition channels. Earlier this month, Localytics added device fingerprinting to our mobile attribution suite.
This enhancement enables mobile acquisition marketing teams to track installs and post-install behavior for channels beyond traditional mobile ad networks. From email to mobile Web, acquisition teams can now measure how earned, owned, and other media are driving new installs, engagement, and revenue.
Although device fingerprinting has existed as a fallback mobile measurement technique to IDFA or GAID-based measurement for years, it has suffered from mixed accuracy and match rates. Because our customers rely on our attribution data to build audiences for push and other post-install engagement campaigns, we made accuracy a priority for our flavor of fingerprinting. We never imagined we’d manage to achieve a rate of 99.96% accuracy. This level of accuracy is greater than what many of the pure play mobile attribution solutions guarantee, and ensures our audience-creation tools remain a reliable way to power downstream engagement.
Hunting for a Higher Return on Ad Spend
Paid mobile user acquisition costs have jumped 136% in the past 36 months to March. While there is still great ROAS to be had on these channels, they’ve become more competitive - and this has incentivized mobile UA teams to make exploratory investments into other channels. From SeatGeek’s offline radio campaigns to SoundHound’s acquisition efforts through Google’s iOS search, the shifting economics of mobile acquisition has accelerated a trend towards testing new waters in offline, email, and mobile Web.
"You'll never get a complete picture of how much revenue your offline marketing spend drives, but that doesn't mean you shouldn't try. The most important thing to think about when experimenting with offline is how to set up as many touch points you can measure as possible. In the past I've used a combination of market lift, promo codes, vanity URLs, and most importantly user surveys to get a sense of how an offline campaign is performing."
Director of Growth, Simple Contacts
Testing and scaling these channels has historically been harder for acquisition marketers to justify and measure, because they are typically unable to capture any sort of unique device identifier. Investing into workarounds like that implemented by the SeatGeek team has become more reasonable as the unit costs of traditional channels is driven higher by increased competition for inventory.
Preserving Accuracy: Keeping it Tight and Deterministic
Over the past three years, the various flavors of fingerprinting have diversified from naive, deterministic methods based solely off of a limited number of static parameters like IP address, to sophisticated methods that leverage statistical modeling to guess at the probability that observed events like installs were generated from a unique device.
Beyond IP addresses, device fingerprinting today takes into account a range of features that can be quickly and easily read from devices including OS type, version, and device model. Localytics’ fingerprinting considers a wide range of these parameters, building a unique profile for each device that the mobile ad industry has termed a ‘User Agent’.
Our approach differs from the probabilistic matching that solutions specializing in deep-linking and non-paid attribution use to extrapolate patterns observed from identified devices to non-identified devices. Additionally, our reliance on last-click attribution and a relatively short observation window of 2 hours minimizes the risk of generating false positives and negatives. The high accuracy of this approach comes at the cost of relatively lower match rates, but this is a tradeoff we’ve consciously made in order to preserve the integrity of our audience-creation and targeting tools.
Future Direction: Post-Install and Multi-Touch Attribution
Looking forward, it seems inevitable that marketers will continue to need some sort of device fingerprinting to serve as a fallback attribution tactic for those times when device identifiers are unavailable or unreliable. A key area for improvement and innovation is in weighted, multi-touch attribution. Let’s say I saw an app install ad on Facebook, and then a different ad creative for the same app on Twitter, and then an interstitial in a news reader app running on AdWords calling me to install. I click on the interstitial and install the app.
In a last-click, single-channel world, 100% of the credit for the install is attributed to AdWords. In the near future, statistical methods that extrapolate from what is observed across a larger sample population in order to attribute some proportion of credit to each of Facebook, Twitter, and AdWords will become table stakes. And eventually, we’ll get to a place where those proportions are weighted with respect to downstream, post-install behaviors rather than just the install itself. Startups that offer part of all of this multi-touch attribution vision are emerging, but the feasibility and accuracy of their approach is still being vetted by the market.
Thanks for signing up. Look for your first email shortly!
We’ll reach out shortly to schedule a time to talk.