The Four Levels of Monetisation
Apr 8, 2022
Building a monetisation stack from the ground up can be a big challenge, especially for less-experienced teams; there’s a lot to learn, and there will always be new problems. Waterfalls present a particular challenge, just because there are so many different possible ways to set them up.
We can break the problem down into four stages:
Mediation and Networks - getting the basics in place
Optimisation - finding the optimal waterfall setup
Segmentation - finding user groups with distinct characteristics
Combination - bringing optimisation and segmentation together to maximise revenue
What makes being a monetisation analyst so hard is that while optimisation and segmentation are both good on their own, they conflict with each other. Optimising more segments takes more time - and that’s why automation is necessary.
1. Mediation and Networks
If you don’t know what ad networks or mediation partners are, you should check out our post here; but even if you do, you still have to choose between them!
At the time of writing, Applovin MAX and Unity LevelPlay are the clear leaders in mediation; Chartboost and Digital Turbine are still working on developing their own platforms, while Google AdMob seems to have neglected their mediation platform.
These days, it seems like large and small studios alike are all moving towards MAX or LevelPlay, which makes sense given the importance of these platforms in the corporate strategy of these big companies. It’s worth noting that some companies choose to build their own mediation platforms in-house - this can work, but it’s not for the faint-hearted.
Once you’ve chosen a mediation partner, the next step is ad network integrations. If you don’t have the right demand sources, you won’t be able to sell ad inventory, and you definitely won’t be able to get the price.
The first step is to talk to as many as possible; you can find list of the ad networks supported by MAX and LevelPlay, and you should probably aim for the largest companies first; that means Google and Facebook, but sorting through companies by number of LinkedIn employees is not a bad way to get started.
There’s a big difference between integrating a network in your mediation and getting it to work; there’s no point having a network if it’s not going to buy your ad inventory! For all sorts of reasons, it can be difficult to get networks to start buying, but you should be able to get 4-5 networks working for each of waterfalls and bidding. The quality of account management varies between ad networks, but you should be able to get a real person to help you understand what’s going on. At this stage, the details of your waterfall setup is less important than just getting a range of different demand sources in place.
Ok, you’ve got mediation set up and a few ad networks are buying - but are you getting the best price? This is where waterfall optimisation really comes in. This is a tricky topic, and there isn’t much good advice out there - but you can follow some good rules of thumb. The key with all of these is to introduce them with an AB test - if it increases revenue, it’s probably a good change!
You can probably have more placements than you think - some waterfalls have more than 100. You’ll sometimes hear people talk about ‘latency’ - the risk that an ad slot isn’t sold because an auction takes too long to complete. Each placement takes a certain amount of time to process, so having lots of high-CPM placements will cause waterfall auctions to take longer - but a reduction in impressions might be offset by an increase in CPM. In the end, the only metric that matters is revenue change. Note that if you’re using MAX or LevelPlay, AdMob currently impose a limit of three placements per waterfall; if you go above this limit, they will reduce their demand for your impressions by around 80%, until you remove the extra placements. While this might soon change, it’s important not to get on the wrong side of that rule!
Each placement should have the same revenue - this will never happen in reality, but it’s worth aiming for. Remove the placements which earn the lowest revenue, and add new placements above the highest-grossing ones.
Fill rate should increase as you go down the waterfall. Fill rate (impressions/attempts) represents the probability that an ad network buys at that price; in order for every placement to have the same revenue, low-CPM placements need higher fill rates than high-CPM ones.
Look for the best-performing network. It might be that one network has a much higher average revenue per placement - if so, it’s probably worth adding more placements from that network.
Think about your top placement - is it getting revenue? If your top placement is generating more revenue than the average placement, you probably need to add one above it; if it’s getting nothing, then you’re wasting time, increasing latency, and you should turn it off.
This optimisation takes time - time to do the analysis, and time to set up the AB test. But small improvements to ARPDAU allow for higher-CPI, higher-scale UA campaigns; if you can get a bit more out of your monetisation, you can spend much more and increase your revenue proportionally.
If users can be segmented into groups which share characteristics, then setting up different waterfalls for each segment will be more efficient than running one waterfall over all the users.
There are several ways to do these segmentations, but the most important are based on geo and privacy settings.
In MAX and LevelPlay, each geo is treated independently; countries can have their own waterfall setups. This makes sense because a wealthy country like the US will have a very different pattern of advertising demand to a poorer country - and so the ‘best’ waterfall setup will differ between countries.
In practice, there are too many countries to keep track of all of them; many companies use ‘country tiers’ as a middle ground, bucketing similar countries together, and the mediation platforms make this pretty easy. But this middle ground is imposed entirely by practicality - ideally, every geo would have its own custom waterfall setup. Users can also be segmented based on their privacy settings. On iOS, users are either LAT (limit ad tracking) or non-LAT; this choice determines whether they can be tracked between apps using an IDFA (identifier for advertisers). Non-LAT users are much more valuable to ad networks, since they can be targeted more effectively, and so, all things being equal, a non-LAT waterfall will have higher CPM placements than a LAT waterfall. Again, mediation platforms help you implement these segments, but they can also be done client-side by developers. For more detail on segmentation, check out blog posts like this one from MAX.
Optimisation and segmentation can both increase revenue. But each optimisation takes time, and segmentation just creates more waterfalls to optimise; analysts just can’t keep up!
Once you have the basics down, ad monetisation becomes a losing battle, a time-sink that feels like pushing water uphill. Compromises like country tiers, which sacrifice revenue for convenience, are a symptom of the problem - more segmentation makes it harder to optimise.
To really get the best out of waterfall auctions, you need to automate the optimisation across every segment, and then create as many segments as possible. Analysts working by hand are too slow, but computers can do it, applying human heuristics in milliseconds. With a tool like Waterfull, analysts can make optimisations across hundreds of geos and segments at the push of a button, gradually evolving towards the optimal waterfall setup to maximise revenue. This stuff shouldn’t be done by hand - it shouldn’t be a cottage industry, done according to individual preferences and secret recipes. With the automation that Waterfull provides, mobile games companies can afford to share their games with a wider audience, and finally give their monetisation analysts a break.