Once you add a second ad network to your app, something has to decide which network gets each impression. That is the job of mediation, and there are two ways to do it: the older waterfall model and the newer bidding model. Most mediation platforms support both, often in the same setup. This post explains how each one works, what the trade-offs are for a small team, and how to switch without relying on guesswork.
How a waterfall works
A waterfall is an ordered list of ad sources. When your app requests an ad, the mediation platform asks the first source. If that source has no ad, the platform asks the second, and so on until someone fills or the list runs out.
The order is based on the price you expect from each source, usually its historical eCPM or a value you enter by hand. Many developers also create several entries for one network, each with a different price floor, so the network gets several chances at different prices:
Interstitial waterfall (illustrative, not real prices)
1. Network A - floor: high
2. Network B - floor: high
3. Network A - floor: medium
4. Network C - floor: medium
5. Network B - floor: low
6. Network A - no floor (backfill)The model is easy to understand, but it has three structural weaknesses:
- The order uses averages, not the value of this impression. A network lower in the list might have paid more for this particular user, but it never gets asked if a higher entry fills first.
- Calls are sequential. Every source that passes adds time before an ad is ready. Long waterfalls can mean the ad is not loaded when you want to show it.
- It needs maintenance. Prices shift by season, country and format. A waterfall tuned a few months ago is probably out of date, and re-tuning it is manual work.
How bidding works
With bidding (also called in-app bidding or header bidding), the mediation platform asks all participating networks at the same time what they would pay for this impression. Each network answers with a price, and the highest bid wins.
This fixes the waterfall's weaknesses fairly directly:
- Each impression is priced individually. Networks compete on what this user, in this context, is worth to them right now.
- Requests run in parallel. The auction has a single timeout instead of a chain of them.
- There is far less to tune. You have no floors to adjust and no order to reshuffle.
Bidding has costs too:
- You still ship adapters. Each bidding network usually needs its SDK and a mediation adapter in your app, which adds binary size, initialization work and update chores.
- Not every network bids on every platform. Support varies by mediation platform, ad format and sometimes account status. Check the current documentation for your platform; do not assume.
- You have less control. If a network performs oddly, you cannot move it down a list. Your main options are to remove it or contact support.
- The auction is less transparent. You see the result, but usually not the losing bids.
The hybrid setup most apps end up with
In practice the choice is rarely one or the other. A typical setup runs a bidding auction first, then compares the winning bid with the entries in a short waterfall of networks that do not support bidding. Whichever is expected to pay more serves the ad.
For a small app, a reasonable target looks like this:
- Every network that supports bidding is configured as a bidder.
- A short waterfall, a few entries at most, covers the networks that only work the traditional way.
- You do not maintain multi-floor ladders unless your own data shows they pay for the effort.
A long hand-tuned waterfall can sometimes earn a little more, but it costs you time every month. For a solo developer, a setup that needs almost no attention is usually the better choice.
Migrating from waterfall to bidding
If you already have a waterfall, do not switch everything in one release. Change one thing at a time so you can tell what caused any difference.
- Record a baseline. Before you change anything, write down per-format numbers for a stable period: ARPDAU, eCPM, fill rate, impressions per daily active user, and if your platform reports it, ad load latency.
- Update the SDKs and adapters. Bidding often needs newer adapter versions than waterfall does. Ship that update first and confirm nothing regressed.
- Complete each network's bidding setup. This usually means linking accounts or entering new placement IDs. Some networks use separate placements for bidding and waterfall.
- Check your app-ads.txt. Bidding can involve different seller entries than your waterfall did. If lines are missing, bidders may hold back spend without telling you. Our post on verifying app-ads.txt covers how to check.
- Start with one format. Interstitials or rewarded ads are good candidates because each impression is worth more, so differences show up sooner.
- Use the platform's A/B test if it has one. Splitting traffic between the old and new configuration at the same time is much more reliable than comparing before and after.
- Remove redundant waterfall entries. Once a network is bidding, its old waterfall entries usually just add latency.
Measuring whether it worked
The metric that matters is revenue per user, not eCPM. After a change, eCPM can rise while revenue falls if fewer impressions are served, and the reverse can also happen. Compare these numbers:
ARPDAU = ad revenue / daily active users
eCPM = ad revenue / impressions * 1000
fill rate = filled requests / total requests
imps per DAU = impressions / daily active usersA few rules keep the comparison honest:
- Compare like with like. Segment by format, platform and your top countries. A change in country mix can look like a mediation win or loss when it is neither.
- Run for whole weeks. Ad demand follows weekly cycles and shifts around month ends and holidays. A three-day test across a weekend tells you very little.
- Watch show rate, not only fill. If ads load faster under bidding, more of them are ready when the user reaches the ad moment. That appears as more impressions per user, not as higher eCPM.
- Reconcile the reports. The mediation platform's estimated revenue and each network's own dashboard will differ a little. Know the usual size of that gap so you do not mistake it for a real change.
If fill drops after the switch, the cause is often configuration rather than demand: a missing placement mapping, an adapter that did not initialize, or a consent signal that is not reaching the bidder. The checks in our guide to diagnosing low fill rates apply here as well.
Common mistakes
- Adding every available network. Each SDK adds size, startup work and privacy disclosures. A few well-configured bidders usually do better than a dozen neglected ones. Add a network only if you are willing to measure its contribution and remove it if it does not earn its place.
- Leaving the old waterfall in place. Dead entries slow down ad loading and clutter your reports.
- Not testing each adapter. Use your mediation platform's test or debug tooling to confirm every network can actually load and show an ad on a real device. A bidder that is misconfigured simply never wins, and no error tells you so.
- Forgetting consent plumbing. Each network needs the user's privacy choices passed through the mediation layer. A bidder without a valid consent signal may bid low or not at all.
- Judging too early. New bidders often need time to learn your inventory. Decide in advance how long the test runs, then stick to that window.
Conclusion
A waterfall asks networks one at a time in an order based on past averages. Bidding asks them all at once and takes the best price for the impression. For most indie apps, the sensible setup is bidding for every network that supports it, plus a short waterfall for the rest. Migrate one format at a time, judge the result on ARPDAU over full weeks, and remove any network or waterfall entry that is not contributing revenue.