Frequency Distribution and Fruit
Many impressions are delivered to a small part of the audience, and a few impressions are delivered to a large part of the audience. It's an inefficient way to buy media. Why haven't publishers dealt with it?
Reach and frequency were a huge topic in online advertising as recently as a year ago. But it sort of quieted down and isn’t viewed as such a burning issue these days.
A major reason is the industry reach and frequency champion, David Smith, CEO of Mediasmith, stopped fighting for it. I remember a conversation we had at an industry event. He told me he was burning out on the issue and had to focus his energies elsewhere.
Though the industry isn’t hyperfocused on reach and frequency, it’s still a high-urgency issue for many media people. It strongly affects media buying efficiency.
The online ad industry buys and sells media by raw impressions. If you buy impressions from a publisher, you purchase a set number of impressions. Let’s say a million. Publishers use two variables to deliver those impressions to the audience: the number of impressions you bought and the time frame they must be delivered in. Publishers set their ad servers and inventory control systems to deliver those impressions as evenly as possible across the timeframe.
The problem is in how audience usage patterns spread on the Internet. About 20 percent of the audience absorbs 80 percent of page views. Some studies stretch that out to 40 percent of the audience eating about 90 percent of page views.
Overlay that behavior on the ad impressions you’re buying. You’re delivering a very high frequency to a very small part of the audience, and a very low frequency to a very large part of the audience. This is an inefficient way to buy media. You’re feeding most of your impressions to a small group of users.
Why haven’t publishers dealt with this? It’s one of the hardest technical problems in the industry. It has to do with inventory management and prediction (or inventory forecasting). The biggest publishers, who deal in hundreds of billions of monthly impressions, must overcome huge technical hurdles. It’s a technology scale problem.
Big publishers built their inventory prediction systems to work with CPM (define) media buys. Add more complicated methods of buying media, such as behavioral targeting, frequency capping, cost per acquisition (CPA), or CPC (define), and it gets much more difficult to predict inventory (hence, to sell it).
When a publisher deals in smaller impression numbers (say, fewer than 10 billion monthly impressions), it’s relatively easy to predict inventory with multiple measures, such as CPM, CPC, and CPA. But even these smaller systems start to choke when more media selling methods are added.
A few companies have systems that reportedly do a fantastic job with these complex media landscapes, particularly CheckM8 and Solbright. But large publishers have such huge volumes they simply can’t risk using smaller ad servers that never operated at such a large scale.
As the market matures, the big ad-serving systems can’t handle all the emerging methods of buying media. That publishers just need to retool isn’t a productive suggestion. These are very big, expensive, complex systems. There’s no proven solution that fixes all the problems.
Let’s look at the volume and behavioral targeting problem from a fruit buyer’s perspective. Publishers sell lots of fruit. You want to buy apples. You don’t need bananas, grapes, cherries, mangoes, oranges, and so forth, just apples. Publishers, however, can’t automatically sort the fruit, so you have to buy lots of other fruits in addition to apples.
Behavioral targeting is like a sorting machine. It sorts the fruit into bins, so you get only what you want. But there’s the prediction and pricing problem: publishers don’t know ahead of time how much of each fruit they’ll have in the mix.
Price is affected by supply and demand, so the only way to price the fruit is by the total amount of fruit available. In reality, price should vary by many factors. For instance, Washington apples might cost more than Georgia apples due to quality and scarcity. Unfortunately, there isn’t an automated way to segment those buckets. It’s even more difficult to predict how many Washington apples will be in the total fruit mix next month.
A small vegetable market dealing in just a few bins per day might have some advantages over a huge fruit distribution center dealing in hundreds of thousands of bins. It takes less time to sort the load.
The good news is this isn’t an impossible problem to solve, merely a difficult one. And many good minds are working on it.
