Why FouAnalytics is NOT automated, NOT set-and-forget

As more and more large advertisers adopt FouAnalytics, a recurring question has been "can we automate the detection and blocking of bad sites and apps?" I have said over the years, that FouAnalytics is not fully automated for a reason -- specifically that there needs to be a "human in the loop" to look at the data in the context of the campaign set up, business objectives, target audience, and other factors. Data in isolation often leads to mis-diagnoses, bad business decisions, or incorrect actions or inaction.

Context is crucial to proper interpretation of data

Let me use a simple example to illustrate. The data on the left shows hundreds of ad impressions per fingerprint (an anonymous representation of a unique device-browser). But is that good or bad? How can we tell? We need to understand the context -- for example, the campaign set up. If the campaign had frequency caps ("fcaps") set at 3 per user, then seeing 100s of ads per user in the data means something is wrong and corrective action needs to be taken -- e.g. check the fcap set up, see which DSP is not enforcing fcaps properly, etc.

If, however, the campaign parameters called for 500 ads per user per month, the data corroborates that the campaign is executing properly, and no action is needed. As you can see, the same data can be interpreted differently depending on the context of the campaign set up.

Don't let algos spend your money

Similarly, you have heard me say over the years "don't let algorithms spend your money." This is because algorithms use certain input signals -- like clicks -- to do their optimizations. If algorithms optimize for higher clicks, they may inadvertently send more of your budget to bots and fraudulent sites, because bots click on ads more than humans do. Simple. Most platforms, including the largest ones, don't have sufficiently advanced bot detection to ignore the bot clicks in order to ensure their algorithms don't optimize for the wrong things. We know this by observing that for the last 10 years, large advertisers budgets have been flowing away from good publishers and towards programmatic channels. Algorithms were optimizing for more clicks, lower CPM costs, higher win rates, etc. all of which are easily gamed by fraudsters and botmakers.

"Humans in the loop" are crucial because their experience and "gut feeling" lead them to investigate things that look abnormal, including things that are too good to be true. Using a slide from 2016 (above), you may see 9.4% click through rates in your campaign. Knowing that humans only click 0.1% - 1% of the time, a human analyst should ask "how" it's possible that this campaign is getting 10X higher click rates. They might pull a more detailed report that shows line item details -- e.g. some sites exhibiting 100% CTRs (which are blended away and hidden in overall averages). With this crucial detail, you can easily see which fake sites -- like pudmed .org (real site is pubmed.org) -- need to be added to block lists to make the campaign run better.

FouAnalytics supplies the insights so clients can make business decisions

The role of an analytics platform is to supply the data that advertisers can use to make more informed business decisions. FouAnalytics is not a fraud detection tech platform that stupidly blocks sites and apps because the IVT is greater than some arbitrary threshold like 3%. I have shown over the years that these legacy vendor platforms have caused more harm than good. For example, they incorrectly labeled legit publisher sites like esquire, reuters, foodnetwork, etc. as high IVT and automatically blocked them. Not only did this defund the legit publisher domains, it also prevented advertisers' ads from going to real publisher sites that had real human audiences. Not only were their algorithms wrong, the set-it-and-forget-it blocking was done behind the scenes and the advertisers and publishers didn't know the harm was being done. These vendors' crappy tech and automated blocking caused campaigns to get worse over time, not better.

FouAnalytics, on the other hand, is an analytics platform for digital ads. As such, it presents the evidence to clients so they can make the business decisions themselves. FouAnalytics does NOT automatically block sites and apps because they exceed some arbitrary threshold like 3% IVT. In fact, in the following example, FouAnalytics presented a site like peopleenespanol.com as having 30% invalid traffic and did not automatically block the domain. The client reviewed the data and decided to NOT block it because they knew their Hispanic audience still visits that site. So despite the 30% dark red, they made the business decision not to block it. You can also see in the list below, some MFA sites have more dark blue (humans) while others have a lot more dark red (bots and invalid traffic). Advertisers can choose which to block and dig into the supporting details to understand why something was marked as dark red versus dark blue.

Reviewing data with clients and comparing what is observed in FouAnalytics to their campaign set up is how I have been using the platform for the last 12 years to deliver actionable insights and recommendations to clients. I am also training analytics professionals to understand the data presented to them in FouAnalytics and how to use that data to support business decisions.

Folks have also asked me if I can give them a long list of bad sites and apps to block. I can. But so many of those sites and apps are no longer in existence it makes that data unwieldly. The recommended process is to tag your current campaigns with FouAnalytics in-ad tags and see which bad sites and apps are currently impacting your campaign. We can then quickly add them to a block list. The slide below show that after just 5 days of measurement, we could cut the dark red in half by adding the right sites and apps to the block list. So no, you don't need a universal block list from me, compiled over the last 10 years.

One final word for those who still worry that using FouAnalytics without automation will be a large amount of work. Have no fear, think about layers. For those who want to cut to the chase and see which sites and apps they should add to block lists, go straight to the Domain App Report and look at the top 10 worst offenders. Those are the large volume bad guys you should consider blocking first. You can repeat this process week after week, to make your campaign cleaner. I don't recommend continuously updating the block list. I recommend adding 10 - 20 bad guys into block lists, letting the campaign run for a week so you can SEE the change (like the slide above) and then making another change. Otherwise you can't see the changes you are making. For those who like analytics and want to dig into further details to understand things, all of the supporting data is there for you to explore. Check out the following case study where we use detailed supporting data from FouAnalytics to determine whether the large surge in red was ad fraud or not.

So what?

Don't let algorithms spend your money for you; FouAnalytics is the way to keep those algorithms in check. Advertisers and agencies who want a "set it and forget it" platform should not apply to use FouAnalytics. Keep using the legacy tools that tell you there's 1% fraud and that automatically block bad stuff for you. (Have you ever wondered whether they actually blocked the bad stuff?). Doing good digital advertising requires effort. Using FouAnalytics to help you optimize campaigns is the effort that's worth making. The analytics are sufficiently detailed so you can understand "why?" before making business decisions. Have no fear, FouAnalytics is here.

Further reading: 620 other articles by me on digital marketing, ad fraud, and analytics