MFAs, junk inventory, & how Flywheel is helping brands navigate a broken ecosystem

This blog post was written by Patrick Miller & the Flywheel Data Science Team

‍

Flywheel is on a mission to eliminate ad waste and inefficient media spend - and this begins with developing solutions to what we call “junk inventory,” or, what is more commonly known as “Made for Advertising” (MFA) sites and ad inventory. In this blog, we’ll talk about what MFA is, why this inventory still exists, the impact on advertisers, and how we’re developing automated, brand-specific, solutions to avoid it as a means to improve campaign performance. (skip to the end if you’re interested in reading about our solution first).

‍

Made for advertising, and defining junk inventory 

While there isn’t a standard definition of what MFA is, it is broadly considered any website where individual pages have been overloaded with ads, forcing dozens of impressions on a single page load simultaneously. Often, these websites will also force users to click through multiple pages to get through the content they’re consuming, creating more and more opportunities for the publisher to cram the page with ads.

For a classic example, search any recipe and you will be bombarded by ads with every scroll - too many ads for a single person to even consider looking at. These types of ads are what we would consider junk inventory, and they are a common indicator of an MFA site. If you’re curious to learn more about MFAs, go check out the recent report published by Adalytics.

‍

How did we get here?

With tens of thousands of sites serving up junk inventory, it’s natural to wonder how we got to this point. Once upon a time, digital advertising was manually purchased via insertion orders. Advertisers and agencies would go directly to the digital publishers and buy batches of their ad inventory to fit into their media plans. The concept of “audience targeting” was based on who the publishers' audience was. 

Then programmatic advertising came along and flipped this model on its head. We saw the explosion of ad tech - specifically SSPs, DSPs, and ad tech point solutions. SSPs (or supply-side platforms) started going to publishers to manage their display and video inventory and enter that inventory into an open auction for sale. On the other end, DSPs (or demand side platforms), lined up advertisers in the ecosystem, pitching them the ability to buy audiences across the open web. Those DSPs were able to layer precise targeting capabilities into their tech stacks, such as demographics, geo-location, and behaviors, all thanks to the cookie. Programmatic provided massive scale and heaps of ad inventory at brands’ fingertips.

This ultimately created a supply and demand ecosystem, and publishers looked to capitalize on increasing digital demand by creating new forms of ad inventory with the likes of display banner ads, interstitials, native in-feed ads, etc. While the buys were more efficient, via automation, the automation also made it easy to buy across thousands of sites, far more than a human could curate.  Over time, and under economic pressure via the walled gardens, many open web publishers realized that the solution for low yield on a per-ad basis was to increase the number of ads on the page.  Brands, chasing low CPMs, incentivized this behavior of forcing more ad inventory onto the digital page leading to what we know as “made for advertising” websites today. 

By generating SEO clickbait content that pulls users in, publishers could actively sell ad inventory to DSPs that would be deemed brand-safe, have high viewability, and generate scale. And on the other end, advertisers were relishing the massive scale, reach, and inexpensive CPMs. With increasing demand from advertisers and increasing supply from publishers, a vicious cycle of ad waste ensued.  When brands relied solely on clicks, viewability, or reach as the key metrics, MFA sites could appear as a great deal.

‍

Why did we begin researching this?

Flywheel has been using Amazon Marketing Cloud to deep dive into site performance for the past few years.  Because Amazon offers a closed-loop ecosystem, we were able to go beyond views and clicks and figure out if sites were also contributing to a sale on Amazon.  While we were quietly working away, in December of 2023, the Association of National Advertisers analyzed $123 million of ad spend with 35.5 billion impressions and published a report on programmatic supply chain transparency which began with an eye-opening stat:  

“Our research found that only 36 cents of every dollar that enters a DSP effectively reaches the consumer.” 

With the ANA’s findings, Flywheel doubled down with AMC to dive deep into this subject and determine the best course of action for brands to take. Many of the findings are related and, before AMC, were impossible for large consumer brands to solve:

• We know that through Amazon DSP campaigns, ads can serve across tens of thousands of sites. To help allocate advertiser budgets effectively, we initially looked across more than 24k sites and ranked them based on the performance of a handful of closed-loop metrics.
• At first glance, these sites would be considered ideal inventory -  inexpensive CPMs, great viewability, and strong detail page view (DPV) rates. However, the narrative shifted when we used AMC to see if these DPVs converted to a sale, with a focus on new-to-brand sales. Unfortunately, they largely didn’t - even when looking at them both in terms of first and last-touch attribution.
• There were thousands of sites that didn’t drive a single purchase, thousands with a conversion rate of less than 0.5%, and hundreds more that regularly saw less than 25% of activity coming from users with recognized IDs, making it difficult to track the value gained from serving impressions there.

‍

How Flywheel is solving: site filtering powered by AMC analyses

While we don’t have the power to eliminate all MFAs with the snap of a finger, we have been building our custom solutions to help brands avoid them and shift delivery towards sites that deliver desired outcomes. This has materialized as a site-filtering algorithm that our team has built for Amazon DSP which blocks delivery on sites that aren’t delivering closed-loop, deterministic outcomes.  It is important to note that some sites will perform well for one brand, but not another, as such our algorithm is unique to each brand. 

So far, the results of our filtering have been positive. The first thing to notice is that CPMs increase once inventory targeting is set, and this is expected behavior! These impressions provide more value to advertisers, leading to the increased cost. MFA sites often have inexpensive CPMs, making them enticing when just looking at the number of impressions, but half the amount of impressions on effective inventory has a much stronger impact.  We see this as similar to how the Top Search placement is worth paying more as the conversion rate is often 2x the other placements–we want to trade on value, not cost.

In February, we continued testing with a brand using our DSP site filtering algorithm.  The results for 02/01/2024 - 02/24/2024 are shown in the table below.

‍

CPMs continue to be higher in the test, however, the value of these impressions is much greater resulting in the test having a lower cost per purchase and CPDPV. The test had a DPVR of more than double the control, further emphasizing the quality of the impressions being served. Additionally, the filtered sites have driven more NTB purchases which implies that these impressions may be more incremental. 

‍

‍

In conclusion

To recap, programmatic inventory is under the microscope. The ANA’s recent report found that 36 cents of every dollar that enters a DSP effectively reaches the consumer. After doing independent research and testing, Flywheel found that grading site inventory based on the actions they were driving, and then applying site filters to future campaigns helped increase detail page view rates by more than 125% and reduced cost per new-to-brand purchasers by more than 30%. While these initial results are promising, we understand that site performance will look different for each brand, and the process of filtering out inventory will likely not be a one-size-fits-all approach.

Are MFAs going anywhere? Probably not right now. However, we are going to continue to build solutions that protect brands from poor-performing ad placements and bias towards sites that ultimately drive sales.

‍