The Inconvenient Truth About Ad Reconciliation
Mobile advertising is here to stay. Having grown by 30% year-over-year in 2018 to $184 billion, further growth is still to come with more and more advertisers shifting budgets towards mobile. Alongside this massive growth we’ll note an undeniable presence and industry concern over fraud. In many ways, fraud has been present in the advertising landscape since its early days and has been somewhat responsible for several developments and changes introduced into our industry over the years.
Perception about fraud’s true impact on the industry, however, paints a slightly different picture than reality.
Case studies, research and reports would have you believe that fraud is, or can potentially be, blocked before it ever touches your marketing budget, and is always managed in real-time, that fraudulent activity is detected as it occurs, and is filtered out by advanced anti-fraud tools designed to protect advertisers from ongoing scams.
While the above is accurate for the majority of the traffic identified and blocked, there’s an unspoken (and unpleasant) truth within our industry: some fraud slips through the cracks. As uncomfortable as it may be to admit it, we most definitely can’t ignore it. Machine learning algorithms do give marketers the ability to apply known fraud behavior patterns to detect and block fraudulent activity in real-time, but the human factor in play continuously tests the machine’s limitations in an effort to gain the occasional advantage. This means that some attempts at nefarious activity will succeed.
Marketers invest precious time utilizing different technologies in an effort to close the gap between expectation and reality. Attribution providers block fraud attacks as they occur, syncing with partners via postbacks to prevent payment being passed on for this activity. Other analytics tools are applied for deeper investigation of the data, to make sure nothing goes unnoticed, identifying cases of foul play in retrospect, followed by a rigorous process of reconciliation.
Fraud prevention solutions can indeed block known patterns, however machine learning algorithms take time to learn the unknown patterns. These patterns evolve constantly, continuously introduced into ongoing activity as they materialize. Fraud trends are essentially made of scattered data points with similar characteristics that we first need to aggregate to identify. The first instances could still be written off as single, unrelated, incidents at first, only to be later assigned to a trend some time into it taking place, once determined and labeled as fraudulent- thus requiring retrospective.
It’s Evolution, Baby
In the past few years, online advertising as a whole and in-app advertising in particular have gone through a process of evolution – maturing with the market and the forces operating within it. A distinct rise in post-install event measurement and LTV-focused marketing are setting the trend for a more quality-user focused market. Advertisers are steering away from the strictly CPI based models and are rewarding publishers for more engaged, high-quality users by introducing more CPA based goals, under the realization that not all installs are equal and some users have higher value than others.
Real-time blocking is still, and forever will be, a crucial part of fraud detection; however, it is just one layer of protection. The logic shown above simply means that our behavioral analysis must continue to evolve, inspecting data even after an install occurs. Our marketing models don’t stop there, which means fraudsters won’t.
The notion of fraudsters truly sticking to methods that are known to be identified in real-time is somewhat naive. Fraudsters are highly motivated, innovative, and very adaptive to whatever changes the market introduces, some capable of reacting to new blocks put in their way in as little as 2-3 days. This simply means that fraudsters have already upped their game and have modified their various tools into mimicking an active user flow well enough to go through the deeper, post-attribution events unnoticed, even reaching as far as actual purchases.
Moving out of our comfort zone is key in order to identify suspicious behavior trends that can only be flushed out post-install. Recognizing fraud that has managed to bypass all the known blocks put ahead of it, masquerading as a legitimate install. The current status-quo of keeping a blind eye from anything not identified and blocked in real-time is simply not aligned with the fact that some fraud does indeed go under the radar at first view. It is estimated that 1 in 4 app installs are fraudulent, this activity can and should be reconciled, how much of it is up to marketers keeping up and evolving their methods.
Facing the facts would be the more constructive approach at this point in time, understanding the growing requirement for a comprehensive solution that brings together continuous fraud analysis and attribution in a way that takes nothing for granted and faces harsh realities head on.