Notice: This blog is no longer updated. You may find a broken link or two

You can follow my new adventures @mikeonwine

Maximizing network revenue

September 13th, 2007

Many publishers either don’t have a strategy for maximizing network revenue or use aging technies such as daisy-chaining to

Out of all the publishers that I’ve talked to many don’t have a solid strategy for maximizing revenue from ad-networks. Many simply don’t understand how networks price, since most are black-boxes that don’t publish how they optimize and choose which ads to display. Yet, there are a couple factors that I find can be large drivers of revenue.

The more times an individual user sees an ad, the less likely he is to respond to it. Ok, seems obvious right? What you may not realize is exactly how quickly user response to an individual ad drops. The following graph is fictitious but representative of the normal response curve of a user on a single site to repeatedly seeing the same ad.


What the above shows is that if you are to maximize revenue you need to start thinking about users and not impressions. A user that’s been on your site for hours and has seen a hundred ads is far less valuable than someone who just logged-on.

Of course every ad-network will tell you that they have a large # of advertisers and deals and that you shouldn’t worry about such things — but lets not forget the Pareto Principle, also known as the 80/20 rule. A small percentage of top advertisers will generate the majority of revenue (and hence higher rates). What that means is that each ad-network will only have one or a couple high CPM ads.

Hence the effective CPM that you receive per user from a network declines as that network continues to see him over and over again. The larger the network the slower the decline, but each will look similar — here’s a rather rough sketch of what various network payout look like (again, numbers are hypothetical, but shape of graph will generally be correct).


Obviously daisy-chaining will not work in this situation as both the medium and large networks have paying ads for each individual ad-view. In the hypothetical example above, you would want an individual user to see the following sequence of ads to maximize revenue:

  1. Small
  2. Medium
  3. Large
  4. Medium
  5. Large
  6. Large
  7. Medium
  8. Small

Comparing the effective CPM of each network individually versus optimized together:


What you see is that you can vastly increase your CPMs by distributing your networks. Now — although these are fictional numbers — the concepts are real and they work. So how do you do it? Rather simple!

It’s impossible to allocate impressions on a per view basis as I did above so we must rely on a little bit of approximation. The way to do this is to setup multiple placements or zones with your ad-network and then to frequency cap them individually within your own adserver. The above could look something like this:


The key here is not to over-complicate. Sure, a Myspace, Facebook or Bebo may create hundreds of different placements each with different caps and priorities, but there are two reasons you shouldn’t

  1. It’s incredibly resource intensive to manage
  2. You don’t have enough inventory

Each placement needs to run a minimum amount of volume otherwise pulling out the effective CPM will be nearly impossible. A lot of pricing is based around user response to ads — eg CPC or CPA based pricing. Since clicks and conversions are rare events you need to have enough volume in each placement to get a predictable effective CPM. On CPC networks you can probably get away with a couple thousand impressions per placement per day but on CPA you’ll want to go closer to ten to twenty thousand.

There is some art here as you will have to update both the frequency caps and the pricing on your placements regularly. The first couple times chances are you’ll see your network cpms fluctuate as you play with the caps & inventory allocations but as you get a hang of it you should gain some serious lift.

Enough for today — next, how to effectively target network placements to maximize revenue.

Related Posts:

  • Foo

    Your list of the optimal ad sequence is wacky. It should read:

    1. Small
    2. Medium
    3. Large
    4. Large
    5. Large
    6. Large
    7. Large
    8. Large

  • Mike

    Not sure I agree?

    1. Small – $2
    2. Medium – $1.61
    3. Large – $1.25
    4. Medium – $1.15

  • Erick

    Great Article. Very accurate as well.

    We all enjoy reading your posts.


  • John

    Why is the small network assumed to have been able to get the highest first impression CPM? I would think the smaller the network, the less likely they would be to be able to sign high dollar CPM deals (usually brand deals which require large reach).

    - John

  • Mike

    Doesn’t particularly matter really — it’s certainly possible that the small network has one or two deals that pay a high rate.

  • Olov Andersson

    Having previously put a lot of thought into this subject, I have some observations. eCPM for an AdNetwork may vary with time. As you mentioned it takes time to accumulate performance data, that is also true internally for an AdNetwork.
    - When you split your placement impressions among several AdNets you will in effect lower all of their individual performances – unless you are a truly high volume site where it doesn’t matter anyway.
    - If they have decent optimization (admittedly non-trivial to judge from their marketing, atleast for the average publisher), they will be doing this on their ads in a refined way internally already. What you are saying is basically that they either don’t have enough ‘good’ ads to expose to the average visitor or, that their optimization sucks. This is rather bold :)

    I must thank you for a very interesting blog though.

  • Matt

    Mike, do you know if Right Media and other big Ad Exchanges/networks allow their advertisers to read client’s cookies. For example, if had placed a cookie on a user’s computer when that user visited their site, could a banner that runs read that user’s cookie (if the user happens to see the ad)? Or do that ad networks prevent their ads from doing this? Thanks for your insight!

  • advance zip repair

    realy a nice blog thanks for sharing