Empirically busting "the long tail"

In digital advertising, everyone seems to believe the theory of "the long tail." They use this belief to believe that they can reach more people with their ads if they took money away from the small number of mainstream publishers and spent that money on programmatic exchanges, so their ads would run on thousands of long-tail sites. This is a fundamental misunderstanding of the theory of "the long tail" as coined by Chris Anderson in his book from 2006 . To this day, advertisers buying digital media believe that buying through programmatic channels means they get more "reach." That tells me they have not looked at detailed placement reports, sorted by largest volume first, to see the distribution of where their ads went.

I have years of data from auditing campaigns for clients and measuring campaigns with FouAnalytics in-ad tags. Not only does this show advertisers where their ads went, it also shows the distribution of impressions within a campaign. The top part of the chart below shows the distribution of top sites within a campaign. Each number is the percentage of impressions. The color coding shows how rapidly the volume drops off after the first 5 - 10 rows. Much of the impression volume within campaigns are concentrated in the top sites; and in many cases, the relative volume drops to be fractions of a percent by row 50. So most of your ads ran in the top 50 sites, not 400,000. Remember when Chase reduced their programmatic reach from 400,000 sites to 5,000 sites (a 99% decrease)? They saw no change in business outcomes. You know what those other 395,000 sites were? Fake and fraudulent sites that have low to no humans. You can avoid wasting money showing ads on those long-tail sites that have low to no humans; and also avoid all those nasty MFA sites too.

Furthermore, remember how adtech vendors tell you they can measure down to the page url for brand safety, ad placements, etc.? That's a ton of bullsh*t, not because they can't measure it, but because it's entirely unnecessary work. If I asked you what was the most heavily trafficked page of any website? What would you intuitively say? Of course, the homepage. This is especially true of news sites or sites that have regularly changing content. The homepage is BY FAR the most heavily visited page. The deeper pages have some volume but the traffic falls off quickly after a single day's news cycle.

The empirical data in the bottom half of the data grid above shows the distribution of pageviews to the top page urls within sites. Note again how rapidly the volume drops off (each number is the percentage of pageviews). Most of the volume is in the top 10 pages. This implies, you can buy run-of-site and show ads to most people, and that all that adtech targeting which purports to deliver the right ad to the right person at the right time no matter what site or page they show up on is a bunch of snake oil that may work in theory but is entirely unnecessary extra cost. Marketers can achieve the same or better marketing outcomes if they advertises on an inclusion list of good publisher sites that have humans, with no other settings (i.e. run of site is fine).

Still don't believe me?

If you have FouAnalytics on-site tags on your landing pages, you will see that real clicks from humans come from real publisher sites, which have real human audiences. See the list of domains on the left, in the slide below. I am sure you recognize these sites. Humans go to sites they know of. Few humans go to long tail sites they have never heard of.

Long tail sites selling ad inventory through programmatic channels consist of a large number of MFA sites designed to slurp up your ad dollars and deliver lots of clicks so you think the campaigns are working. Your Google analytics only tells you how many clicks came from utm_source=programmatic, but it doesn't tell you about the quality, or lack thereof, of the clicks. In the slide below, you will see the FouAnalytics color-coding for the clicks coming from 15 different programmatic campaigns. Orange and red mean bots, dark blue means humans.

The vast majority of clicks from programmatic long tail sites are bot clicks. About 1 - 4% of the clicks are coming from humans. An easy optimization for your campaigns would be to turn off programmatic (the long tail) and buy media using an inclusion list of good publisher domains (and some mobile apps). This is directly parallel to turning off (fr)audience networks which again consist of long tail MFA sites that chomp on your budgets and give you no business outcomes.

See: (Fr)audience Networks Drive Massive Volume to Eat Up Your Budgets

The concentration near the top is the same on YouTube campaigns too

In case you were wondering about whether the above statements hold true in YouTube campaigns, they do! Citing data from good partners, here's an example of the distribution of impressions by YouTube channel. You can see it is a bit more spread out than the examples above, but not by much. The previous advice holds true: use a curated inclusion list of YouTube channels that are pertinent to your brand. This helps you avoid the million other channels that are not pertinent or potentially harmful to your brand. By doing it this way, you don't have to pay for brand safety vendors who lie about helping to prevent your ads from brand safety issues. You can do it yourself, for free. Time to save some money.

What next?

Hopefully this data shows you that 1) you were not getting the "reach" you thought you would get by spending your budgets in programmatic channels to reach the long tail, and 2) all that reach is unnecessary anyway, if not outright fraud and waste. As a media buyer, this means the following:

1. ask your agency or vendors to give you detailed placement reports with numbers of impressions by site/mobile app. Do the same with YouTube campaigns, so you can see impression numbers by channel. In an excel spreadsheet calculate the percentage of total impressions and sort by highest percent first. This way you can easily see if your distribution is highly concentrated at the top, or more evenly spread out.
2. buy media with an inclusion list of good publisher domains and some mobile apps; this helps you avoid 90% of the fraud and waste
3. stop wasting money on "targeting" and all manner of snake oil sold to you by adtech vendors and run your ads run-of-site on those good publisher sites

I can guarantee that you will see better actual outcomes, even if vanity metrics like numbers of clicks drop off. Be sure to tell me if I'm wrong on this, either privately or publicly in comments.

If anyone needs to see this, please share out, re-post, or tag them in comments for further discussion. Happy Thursday y'all.

For those of you having deja vu, you're right on. I showed a variation of this analysis in March 2021 . In each of the 4 examples below, 22 domains, 23 domains, 6 domains, and 47 domains respectively ate up 75% of the impressions in each campaign. You can barely see the long tail to the right of it, because the volumes of ads going to them are so small. Note also the number of different domains that had more than 1 impression each, at the top of each of the 4 charts below. Those numbers are in the hundreds, not hundreds of thousands, further corroborating the move that Chase made to cut 99% of the useless domains from their programmatic media buying in 2017.

https://www.slideshare.net/augustinefou/programmatic-reach-analysis-2021