Pickpocket Warning Sign

17 October 2017

Watch Your Numbers — And Your Wallet: A Programmatic Ad Fraud Case Study

by Freddy J. Nager, Founder of Atomic Tango + Writer Who Likes Numbers; photo by Cory Doctorow via Wikimedia Commons…

Digital advertising these days is like wandering into a room full of pickpockets. Obviously, you should avoid doing that altogether — you do have other choices — but if you must go, watch your wallet.

I know this from experience.

About a year ago I noticed some interesting ads at the ends of news articles. Instead of promoting a product or service, these ads featued other articles on other websites. Since I’m a writer who loves having readers (did I ever mention that you rock?), I decided to run a little self-promo.

So I tested a few ads hyping some Atomic Tango posts. The programmatic advertising service gave me the typical choice of paying for clicks or paying for views. Now as a follower of ad fraud researcher Dr. Augustine Fou, I know that paying for views is essentially asking to have your pocket picked:

  • If anyone in the advertising-and-publishing food chain is a crook, then many (if not all) of the “views” will be by bots, not humans.
  • If humans do “see” the ads, they likely won’t notice them because of ad blindness, ridiculous ad placement (at the bottom of pages), or corrupt placement (hidden behind other ads).
  • And even if every participant in the food chain is a law abiding human, in many cases, the same ad gets shown to the same person again and again and again and again… So many “impressions,” so little impact. (I call that “Facebook Syndrome.”)

So I paid for clicks.

Now, Dr. Fou has also shown that clicks are also easily faked — bots can be programmed to do anything a human can — but I had no other choice with this ad service. And clicks at least mean someone (or something) is coming to my website. I just had to watch my wallet and my web analytics.

And here’s what I saw…

Incoming!

Over a few weeks, my ads averaged fewer than 10 clicks per day. Meaningless but also inexpensive since I was paying only 15 cents per click. And I thought, who knows, one of those clicks could be an influencer who might love my article and make it go viral so that it attracts a publisher who offers me a multi-book publishing deal which in turn leads to public radio offering me my own show…

Amazing how much pleasure I derive from a 15-cent click.

Then, one morning, the number of clicks to one article hit 40 in just a few hours. Did that influencer arrive and share my post with a pack of voracious marketing junkies looking for their next hit? I dove into the data:

  • The up-to-the-minute report from WordPress showed 41 clicks from the ad… but 0 clicks on any links within my article. Strange: I usually get at least a few clicks on links. In addition, no other section of the website that day received more than single-digit visits. This meant that visitors to the article didn’t explore further — also unusual.
  • The ad service’s performance dashboard showed over 70 clicks on the ad (oh where did those missing 29 visitors go?), but not where the ad was running. I also noted that the click-through rate was about average at 0.03% (i.e., 3 out of every 10,000 people who saw the ad clicked on it — which, by the way, is another reason to never pay for views). That meant the number of views had skyrocketed to over 200,000 for no apparent reason. I immediately froze the ad.
Ad Fraud Analytics

Google Analytics report — click to expand.

  • The next-day report from Google Analytics confirmed my suspicions (see image). GA showed 42 clicks coming from the ad (about the same as WordPress) and a particularly glaring number: visitors from that ad stayed on my site for an average of only 3 seconds. That’s hardly enough time to read one sentence. By comparison, visitors from one of my LinkedIn posts stayed for an average of 6 minutes — 120 times longer. Now, I know that business professionals on LinkedIn would likely have more interest in my writing, particularly if I know them personally, but even visitors from Google averaged 44 seconds on my site.
  • Further, visitors from LinkedIn were clicking multiple pages. Nearly all the visitors from the ad bounced (leaving after only one page).

The Verdict

cats feasting

“Who ya callin’ fishy?!”

This smelled fishier than my cats’ breath after a Friskies feast.

People or bots were clicking on the ad but immediately leaving my site. While this didn’t cost much (70 clicks amounted to just over $10), it could have added up to hundreds of dollars if I didn’t monitor my site analytics — or if all I looked at was raw traffic.

I contacted the ad service, who assured me that they have fraud detection systems, but they also thought my numbers looked unusual. After investigating for a few days, they told me they couldn’t explain the results — not even the particular website generating the clicks.

Note: I never suspected the ad service itself of fraud, since they’re a well-known brand, and I had used them for weeks without issue. Rather, I suspected one of the sites on which the ad appeared. Since website publishers share in the ad revenue, they’re incentivized to cheat.

Even if the clicks were free, the resulting traffic may be worse than worthless. SEO experts claim that Google tracks user behavior on websites to determine site “quality,” and that affects Google search results. If users spend only a few seconds on a page, Google might downgrade it.

So I canceled the ads and haven’t run one since. Sigh. So much for my public radio show.

Key Takeaways

If you do decide to run digital ads, make sure to do all the following:

  1. Whatever you do in marketing, always test before you seriously invest.
  2. Remember that nearly every online action can be faked by bots, whether clicks or views.
  3. Check your site stats multiple times daily using multiple analytics tools.
  4. Compare your numbers to your usual site averages and (if available) industry averages.
  5. Look beyond raw traffic numbers to other measures of on-site behavior, such as time spent, number of pages visited per session, and activities such as comments.
  6. Note that bad traffic may be worse than no traffic. Instead of bouncing bots, you want numerous human visitors who view multiple pages over many minutes, comment, share, and otherwise engage with your site. That means targeting relevant customers with valuable content — but that’s a topic for a future post.

I’ll let you know as soon as I write it. Just be sure to subscribe to this blog to find out, because there’s no way I’m advertising it.

Tags : , , , , ,

Freddy is the Founder & Creative Strategist of Atomic Tango. He also teaches at the University of Southern California (go Trojans!), shoots pool somewhat adequately, and herds cats. Freddy received his BA from Harvard and his MBA from USC.

Leave a Reply

Your email address will not be published. Required fields are marked *