1 of 17

Case Study - Willyweather.com.au

Publift is AUNZ's fastest growing publisher consultancy representing more than 20 premium publishers. We specialise in remnant ad revenue optimisation and technical ad implementation. Publift help publishers grow their existing remnant ad revenue channels and introduce new products like native ads and programmatic solutions. The team are also specialists in ad server setup, training and ad operations workshops.

Colm: Thanks for the intro. Today, we are going to discuss a hands on approach to some of the strategies described earlier. I’m also joined by Tobin, my business partner & Will Bruce co-founder of Willyweather. Tobin is going to go more indepth into how we’ve optimised the account. Will is going to give us a publisher’s perspective.

Willyweather is Australia’s fastest growing weather site with over 2m UV’s per month.

2 of 17

Agenda

Main Problem Areas for Publishers
Calculating Actual eCPM
Passback Strategy
Optimizing Ad Exchange
Reporting
Results

3 of 17

4 of 17

Calculating Actual eCPM

Thanks Colm. To solve these issues you need to use the right settings in your ad server. In Willyweather’s case DFP. We needed to set the right price, FC & priority for each ad network every week.

Will how did you set network pricing in DFP previously?

Thanks Will. I’m going to talk through 3 ways you can increase your remnant ad revenue - 1. Setting the right prices, priorities and frequency caps for other networks. And managing performance changes from networks. 2. Using a better passback strategy. 3. Optimizing AdX performance. First of all I’m going to show you how to set the right price for each network by calculating the actual CPM you are getting from each network. Then we’ll cover how to monitor and adjust performance to react quickly to market and network changes.

Login to DFP and pull an order report to see how many impressions were sent to each network
The only two numbers that matter are DFP Total Imps & Net Revenue when it comes to assessing network performance.

5 of 17

Calculating Actual eCPM

3. Login to each network account, set the date range to the same period as the DFP report and take the paid views and Net revenue figures for that period. While we’re here take a look at Network C’s reported eCPM - it’s $2.45. This is a common mistake publishers make - they look at this number and prioritise each network based on this figure. But this figure doesn’t take into account rev share, fill rate, impression discrepancies between DFP and Network C and also passed back impressions which are less valuable.

Also notice how the received impressions varies from the amount of impressions sent from DFP. Industry standard is to expect up to 10% discrepancy between impressions sent from an ad server to impressions received by a network or direct campaign. In this case it was 8% of impressions sent that were not received. Worth noting if you are using a waterfall passback strategy which we’ll come to later.

6 of 17

Calculating Actual eCPM

4. Add these figures to a weekly tracking sheet where you can monitor the performance of each network and Total eCPM/revenue.

5. You can see column G shows the actual eCPM for Network C. You get this figure by dividing net revenue from Network C by DFP impressions sent to Network C. Note has this is very different to the eCPM displayed in the Network’s reporting UI. Network C’s report said $2.45. Our tracking sheet says $0.53. This is because Network C didn’t fill 72% of the impressions they were sent in January. 2.4M impressions were passed back or unfilled. They also lost 8% of impressions between DFP and their tag.

6. The actual CPM is roughly what you should be setting your DFP CPM to. Setting the DFP CPM about 10% higher than the Actual eCPM seems to be the sweetspot.

7. Use fill rates to adjust the frequency caps for each network. You can see in the weekly comments how CPMs and frequency caps have been adjusted in reaction to changing market conditions and revenues from Network C. It’s a good idea to track the changes you make each week in column H so you can closely monitor their impact. Talk through some changes on the sheet.

8. Use Price Priority to encourage network competition.

7 of 17

Passback Strategy

In the last section I showed you impressions sent from DFP & impressions filled by Network C but what happened to these impressions that weren’t filled by the network? You want to be able to account for every single impression that was generated on your website. Will do you remember what Willyweather were doing with passbacks?

Well typically most publishers create a waterfall structure where impressions are passed down to the next priority network each time they are not filled. AdX or AdSense is usually bottom of the chain because they have the highest fill rate. In the previous example we saw 8% of the Total impressions for Network C went missing between DFP and the network. Using a waterfall you’ll see upto 10% of impressions being lost at each passback stage. Not good. Add to this lost contextual information about the page, user information, time delays - the impression gets less valuable each time you pass it back. So you want to avoid passing back as much as possible. Getting any network you work with to have fill rates above 70% should minimise passbacks. We’ll discuss how to do this in the next section - using frequency caps. When this is not possible it’s best to passback to DFP so you can run the impression auction again and hopefully get a good price from AdX or AdSense.

So to create network passbacks to DFP, you create a new ad unit. Name it in a way you can easily track. Set the size and allow AdSense backfill. Save the ad unit.

8 of 17

Passback Strategy

9 of 17

Passback Strategy

10 of 17

Optimizing AdX

So the last 2 sections I covered improving network performance. Now I’m going to cover some ways you can increase your AdX revenue. Firstly I’ll talk about frequency capping other networks. Then we’ll talk about using programmatic deals.

Frequency caps. This is a bit complex so please bear with me while I use my notes here. We’ve tried to visualise the impact of using frequency caps. Along the bottom you have the impressions generated by a user. Most pages have 3 ad units per page so this shows the impressions generated on the first 7 pages a user looks at on the site. Will do you know how many pages per session your users have? 2-3. OK so your per session journey would end around impression number 9 on most sites. But if the same user comes back on the same day he will continue from impression number 10 onwards - because most frequency caps are per user per day.

The coloured bars show the bid that won the impression auction in DFP. Blue is AdX and red is Network A. The grey bar to the side shows the actual CPM paid by the winning bidder. For AdX the amounts are always the same thanks to Dynamic Allocation. For Network A there is a difference because the CPM you set in DFP is not a real-time price from Network A.

This is an example of a publisher not using frequency caps. They have set the CPM in DFP at $2 for Network A. Probably because this is the eCPM figure they saw in Network A’s account. All the red bars are set to $2.

Let me talk you through what happens as the user goes through the site. On the first page AdX recognises these are the most valuable impressions and is willing to pay a high CPM to win the impressions. Through Dynamic Allocation in DFP AdX takes these impressions and pays the exact amount it bid to win the impression. After the 5th impression Network A begins to win impressions. Because there is no frequency cap on Network A they continue to take all the rest of the impressions the user generates. This is because AdX is not willing to beat a $2 CPM for those impressions. The grey bars indicate the price that Network A actually paid for the impressions. You can see after the 14th impression Network A doesn’t actually have any more ads to show so those impressions go unfilled.

The total value of these 20 impressions was 2.4 cents.

11 of 17

Optimizing AdX

12 of 17

Optimizing AdX

13 of 17

Reporting

14 of 17

Reporting

15 of 17

Results & Analysis

16 of 17

Results & Analysis

17 of 17

colm@publift.com 0428042668

tobin@publift.com 0404477922

Contact: