(Fou)Analytics for your digital ads
You have Google Analytics for your website, so why don't you have analytics for your digital ads?
Sure, you've been paying for fraud detection for years. But what have you done with those useless reports? When they consistently report 1% fraud or 100% "fraud free" over the last 10 years, what do you do? Nothing, right? Everything's fine, keep spending. By now, after many many "scandals" [1], [2] documented for years, advertisers realize ad fraud is not the 1% that these legacy fraud verification vendors have been reporting, and the Association of National Advertisers had been parroting.
Legacy fraud verification vendors
The reason these legacy fraud vendors report such low fraud is because their tech is tuned for looking for invalid traffic ("IVT"), otherwise known as bots. Bots are fake users that repeatedly load webpages to cause ads to load. But obviously, digital marketing as evolved away from sites, to mobile and mobile app. The alarm clock app, brightest flashlight app, emoji keyboard and color by number apps that load ads continuously throughout the night, when the app is not in use and the mobile device is not in use are not bots hitting webpages. The legacy fraud vendors don't catch these forms of fraud (they might lie and tell you they do; so be sure to ask them for proof and to explain why something was marked as fraud or not fraud). There are many other forms of fraud that they are not even looking for (you don't have to believe me; but I have 10 years of data measuring head-to-head against these jokers). Because it is all "black box" and they won't/can't explain any of their measurements, you will never know whether they measured the fraud correctly or completely. Ten years of evidence suggests that they miss most of the fraud. The most recent examples include "the billions of bid requests with mismatched domains" [WSJ, 2022] and "billions of mis-represented TrueView ads" that meet Google's own definition of invalid traffic [WSJ, 2023].
The 1% is not all the fraud there is; it's all the fraud they can catch.
Further, in the slide on the right, above, you can see two columns -- "monitored ads" versus "measured impressions." Monitored ads are the total count of ads -- 781 million. Measured impressions means the fraud vendor ran a javascript tag to collect data -- 83 million. That means they measured only 1 in 10 ads with their javascript detection tag. 9 in 10 ads were not measured, but yet that row is still marked as 100% "fraud free" and 0% "SIVT" ("sophisticated invalid traffic"). Do any of the advertisers reading this think that this is correct? Is it right to assume there's no fraud in the 9 out of 10 ads the vendor didn't even measure? Which advertiser reading this thinks they should continue paying for these useless legacy fraud detection vendors? If you can no longer trust the 1% fraud they tell you, what do you do?
Analytics for your digital ads
Right. Upgrade your tools. Use analytics for your digital ads. If there were any other analytics platform for your digital ads, I'd tell you about it. I don't know of any others, so I will tell you about FouAnalytics for your digital ads. I started building these tools 11 years ago, so I can help advertisers audit their digital campaigns. I didn't trust anyone else's tech and I didn't trust anyone else's data. So I built my own tools to collect the data for analysis. It collects impression-level data, equivalent to log level data. You get this data by copying-and-pasting a FouAnalytics tag into your DSP or ad server, whichever is easier. This solves the challenges of getting access to log level data (ISBA can attest) and handling large data sets. Most advertisers don't have the analytics/big data folks to ingest and process this kind of data, let alone derive actionable insights from the data. The FouAnalytics dashboard was originally built for my workflow. That means it's been optimized to help me get to the actionable insights as quickly as possible -- for example what are the bad sites and apps that need to be added to the block list most urgently.
For advertisers buying programmatic ads without an inclusion list, the best way to make campaigns better is to progressively clean the campaigns by adding sites and apps to the block list. You don't need to add thousands of sites and apps at a time; you just need to add the "worst offenders" first, because they are eating up the largest numbers of impressions fraudulently. See: How to Use the Domain App Report in FouAnalytics. By adding the top 10 - 20 bad guys to the block list, and repeating the process every few weeks, you optimize away from the fraud and your ads flow to better sites and apps. You can also optimize towards humans. But this is a topic for a different article. See: Human CPM (hCPM) and Marketing to Humans.
FouAnalytics in-ad tags are copy-and-paste in various DSPs and ad servers. See: How to Deploy FouAnalytics Tags on Various Ad Platforms. For certain types of ads -- like Google search ads, Facebook, YouTube, and Tiktok ads -- FouAnalytics is not allowed to deploy a javascript tag in the ads themselves. So we place a FouAnalytics on-site tag on the landing pages and measure the quality (and quantity) of clicks that arrive from various paid media sources. We automatically parse out the utm_source that we see in the click through urls. Below are 4 examples. Note that each row has color coding too -- more blue means good, more red means bad. The percentages highlighted in yellow at the top of each data grid shows you the "prevalence." That means 36.6%, 23.6%, 49%, and 20.5% of the traffic to the landing page/website has "utm_source=" in them, suggesting that is the portion of traffic coming from paid campaigns. I can glance at this and know that the left-most example didn't turn off Facebook Audience Network in the Facebook campaigns (half dark red) and the right-most example didn't turn off Bing Search Partners (1/3 dark red). This gives me, and the advertiser, a quick way to troubleshoot what needs troubleshooting. The other rows in the data grids below look fine -- lots of dark blue and light blue -- so no action needed on those.
You can also compare the quality across channels. For example, paid search (D, E) both have more dark blue humans than paid display (A, B, C). Paid social (F - Facebook with Audience Network turned off) has far more human clicks than paid social (G - Twitter). You can optimize away from fraud by reducing budget to A; and optimize towards humans by increasing budget to C. And so on.
领英推荐
Below, you can see the quality of clicks when you have audience networks turned on versus turned off. You can even think of programmatic as a kind of audience network/extension, because most humans go to good publisher sites that they have heard of. There's not enough humans going to long-tail, niche content sites, so the vast majority of the traffic to those sites are automated (bots, forced redirects, popunders, etc.) as you can see from the color coding.
Not fraud, but also not optimal
With analytics for your ads, you can do far more than just look for bots and fraud. In the example below, note the 4 yellow boxes. In each, you can see the day tick over to the next day. The larger green bars show a surge in ad volume served at midnight, 1 am, and 2 am. Then lower volume green bars. By the time you reach 12 noon, all the ads for the daily budget are used up, leaving no ads to serve for the rest of the day when humans are awake and online. This is not fraud, but this is also not optimal for the advertiser. With these details, the advertiser can decide to set up day-parting, i.e. turn off ads between 1 am - 5 am (and save 4 hours of out of 24 hours of impressions so they can be re-allocated to waking hours). The advertiser can also set up pacing to more evenly spread out the volume across the hours. None of these details are provided by legacy fraud detection vendors or the reports provided by the DSP or ad server. (Most advertisers had never pulled such reports before; if they use FouAnalytics, they won't need to).
Let me end with a fun example of "what does the FouAnalytics data show me?" In slide 28 below, you see four 300x250 ad units and the click locations (green dots). What do you think the users were trying to click? Right, the close button. The users were trying to CLOSE the ad not CLICK the ad. So these should not be counted as valid clicks for the campaign. Whereas, in slide 29, we see clicks across the face of the 300x250 ad units (yellow boxes), and further supporting details of WHAT was clicked ("adclick.g.doubleck.net ... "). This supporting data shows you proof of where the click occurred and what was clicked so you can verify that it was measured correctly. Many many folks have asked how it is possible that FouAnalytics better measures fraud and other things than vendors that have 100s of millions of dollars of venture capital and quarterly revenues. Don't take my word for it, just "see Fou yourself" by measuring head-to-head against ANY legacy vendor you are using now.
And finally, if you have a question about cost.... FouAnalytics is free for small and medium businesses, just like Google Analytics is free. For enterprise customers (big advertisers doing 1 billion or more impressions) I charge an annual subscription like Microsoft Office or Google Analytics Enterprise. It's a flat fee based on the media plan for the year. I won't charge more if you go over the estimated quantity and we DON'T need to do the useless work of settling up to actuals every month. Further, if you are like the 6 out of 6 advertisers at the end of 2022, they are taking their existing budgets for fraud verification from legacy vendors and shifting it over to FouAnalytics. You don't even need to find new budget. Going forward, it will be an analytics budget line not fraud verification or brand safety.
You are welcome to run as many low-cost pilots as you want. Just message me to request access.
Watch more:?https://www.youtube.com/@augustinefou