The real cost of first look

It’s becoming increasingly common for some ad networks to push hard to get first look on your impressions. Publishers will often be offered a premium price (either a floor or fixed CPM) for every impression the network takes, with the promise to pass back any un-taken impressions to the publishers other sources.

Such deals sound great, but we frequently find that they don’t perform quite as sold, resulting in lower revenue for the publisher. Knowing good deals from bad is essential to avoid being sold a dud, so let me step you through the basics

What is first look?

In this context we’re talking about giving a network first-look at impressions (Not to be confused with DFP first-look). This is simply giving the network first-refusal on each impression. The network gets to see every request before any other source. They serve ads into the ones they can fill at the agreed price and pass the remainder on to your other partners.

This approach is particularly popular with re-targeting networks such as Criteo or Conversant that can potentially pay much higher prices for impressions served to particularly valuable users. Retargeting networks work with advertisers to identify high-converting users such as those who abandoned shopping carts. These audiences are targeted through broader networks and exchanges too, but specialist retargeters have a higher proportion. That potentially means higher CPMs (although with lower fill).

Why do networks want first look?

Networks want first look so that they have a better chance of winning the impressions that are most valuable to them. If they have a user cookied as being one-click from ordering a new laptop then they want to serve the ad that will make that advertiser happy before any other network can. Many networks will also collect additional user information when they take that first look, mean that every request has value to them (whether they pass that value back on to the publisher or not).

This probably sounds great: Higher CPMs and you can just pass the impression to your usual sources if they don’t want it. There are some pitfalls though, so let’s look at those.

Pitfall 1: Settling for too little

At OKO we believe that too many publishers give up first-look for too little gain. Promising like “We’ll beat your CPMs by 20%” sound tempting, but are designed to deceive. If a network offers to beat your CPMs and guarantee to match your current fill rate, then brilliant: Get their tags trafficked fast. More frequently they are looking to buy your best impressions for little more than your average performance. Let look at a simplified example:

A publisher serves one million impressions. His average CPM is $1.03, but if we break that into just five price buckets, we can see that the actual price he receives for impressions varies. (In reality, prices vary much more than this).

Avg CPM	Impressions sold	Revenue
$0.25	100,000	$25.00
$0.50	250,000	$125.00
$1.00	300,000	$300.00
$1.50	250,000	$375.00
$2.00	100,000	$200.00

The publisher receives $1025 in total and has a CPM of $1.03, so a network offering a $1.30 CPM might seem like a great deal. However, if they fill just 35% of impressions, they might just buy the 350,000 impressions that would have been sold for $1.50 or more. In that case the result would look like this:

Avg CPM	Impressions sold	Revenue
$0.25	100,000	$25.00
$0.50	250,000	$125.00
$1.00	300,000	$300.00
$1.30	250,000	$325.00
$1.30	100,000	$130.00

The publisher would now be paid just $905 for the same impressions. They’d lose $120 despite being offered a higher CPM.  Impressions do have different value to different networks, but overall numbers do tend to follow this sort of pattern. Of course, change that $1.30 for $2 and suddenly the deal looks  more rosy.

Pitfall 2: Lost impressions

The above example suggests a perfect system where every impression is bought. In practice every additional step in a system losing some impressions. If you measure the number of impressions a partner passes on compared with the number sent to your passback tag there will always be variance. Not every impression gets passed back and you, the publisher, of course lose money with every one that it lost.

A low fill partner that is not passing back efficiently can cause significant lost revenue. It might be that the gain on the impressions that they take is enough to make up for that, but it is important to factor that in to your deals. At OKO we provide publishers with the means to value and compare the actual performance of entire chains of passback deals in order to bring transparency to the whole system. It can be done manually though, you just need to know the earnings and the impressions involved.

Pitfall 3 : Performance

Just as every additional step losing impressions, it also adds to load times. That can impact the speed of ad loading, but also impact user experience. Slow website tend to have lower visit-depth and fewer repeat visits, which means less opportunity to earn. Our platform now measures the performance impact of every new partner added, but you can do similar by using services like webpagetest to do spot checks.

The solution: Request CPM

To measure whether your first-look deal is working for you, you need to understand your request CPM. A publisher’s available ad requests are the commodity that they have to sell, and request CPM is a measure of what they are getting paid for that, so any first-look type deal should be increasing request CPM. Let’s look at a simple example:

A publisher works with NetworkA, who pay a fixed $2CPM for impressions they take. NetworkA tags are placed on page and the AdSense tags added to the NetworkA dashboard

• NetworkA tags are called 1,000,000 times
• They take 200,000 impressions
• NetworkA’s dashboard reports 800,000 passbacks
• The AdSense units is called 700,000 times
• AdSense earns $700

It’s important that we use different AdSense tags for each ad unit we create with Network A. This allows us to measure all earnings (NetworkA + AdSense) on a per unit basis and understand the performance. If we passed back from several units to a single AdSense tag this would not be possible.

In this example our request CPM is $1.30. We get to this figure as follows

• Total earnings are $1300
  • $700 from AdSense
  • $600 (2x$300) from Network A
• Total impressions are 1 million
  • Note: The passback numbers and adsense impression are ignored. We are interested in how many requests were made on page
• Ad Request CPM = $1.30
  • $1300 / 1,000,000 * 1,000

If we really wanted to understand the uplift we’d split test those requests and compare the $1.30 with the CPM that AdSense would give us if it took all impressions.

So, should you do it?

Emphatically, YES! First look is useful bargaining chip to help get the best deal from your top partners. They important points are to ensure that you are getting a fair price and then monitoring the actual incremental gain you are getting from the deal. Publishers using DFP can also hold back one card by enabling dynamic allocation, effectively giving AdX last look and any impressions.