+9 votes
I've heard some mention of different Ad Ops strategies to optimize your Ad Waterfall and CPM floors but there still seems to be a serious lack of information on this topic. If I'm wrong and anyone knows some great sources for learning about this then I'd be happy to check it out. I've read all the Quantmar answers relating to this topic.

So assuming a beginner at Ad Ops was tasked with creating a manual waterfall for their new predominantly ad monetized game.

So say this beginner has constructed a basic Ad Waterfall based on the standard High, Middle and Low (Backfill) Tiers strategy where you give each of your say 3-5 networks a chance to bid on an impression at each of these 3 Tiers (or even 1 instance of Auto-CPM for each network?).

After this, where do you go to start optimizing CPM floors and number of Instances/Line Items in your waterfall?

I've heard recently of optimizing with Fill Rate, where if the fill rate is low (0.2%), kill the instance as it’s not contributing to your eCPM and harmful for latency cost. If the fill rate is high (10%), split the instance into two instances with a higher eCPM as the chances of you being able to generate more revenue per impression is higher. Then just aim for 1-5% Fill/Availability Rate for fixed eCPM instances. I've also recently heard 40 instances has proven to be an optimal balance between maximizing revenue and minimizing latency?

Is this the way to perform Ad Ops and fine-tune your CPM floors and number of instances?

Also how big of a role does A/B testing play in optimizing CPM floors? Can you simply A/B test all of your changes to your Waterfall and simply continue to apply the winner? I've heard it is hard to get reliable results on Waterfall A/B tests as the network's performance can fluctuate fairly significantly over the span of a test?

Would love to hear anyones thoughts on this or anything they'd like to share!
by (650 points)

1 Answer

+3 votes
In my experience having around 5 networks is optimal, where usually 2-3 networks deliver the best ecpm, while others fill the rest. Having less than 5 could be troublesome if you need to disable a network or two while figuring out issues (there's usually something causing crashes). If you have 2 strong networks (A and B) in the mix, I would set up 4 instances for both A and B - Premium, High, Medium, Backfill (mediation). If you get 10% fill rate on the highest target, you could try pushing it a bit, however I have usually seen that pushing too much affect overall revenue negatively, so I would be happy with 5-10% fill on a high ecpm. Remember to not only monitor ecpm and fill rate, but also ad revenue on user level. To avoid surprises you also need to be aware that some ad networks only have soft floors, not hard floors.

I would try with:

A Premium

B Premium

A High

B High

C High

A Medium


Personally I think 40 instances, as you mention, sounds like a lot of work, if that is for 1 app and 1 ad type (rewarded or interstitial)

I usually try to limit to 10-15 in waterfall to make it manageable, and have mediation pick up the rest. If you have 1 main market that brings you the biggest chunk of your ad revenue, focus on optimizing for that, and let mediation solve the other geos.

About A/B testing - haven't tried, as it required a lot of setup on each network last time I checked (multiplying instances), to make it work, so I didn't bother.  

No matter how you solve it - make sure to establish relations with each networks and make sure they send traffic your way and include you in the good campaigns.

Best of luck!
by (530 points)