Introduction: Why Waterfall Optimization Still Matters
Back in 2017, the concept of ad waterfall optimization was simple: sort your ad networks by eCPM and let them compete. The higher the bid, the earlier they appeared in your waterfall.
Fast forward to 2025, and things aren’t so simple anymore.
In-app bidding was supposed to kill waterfalls. Spoiler: it didn’t. Instead, we’re now living in a hybrid monetization era where bidding and waterfalls coexist—and optimizing them is harder than ever.
So what changed? And why do top-grossing studios like Supercell and Playrix still invest heavily in optimizing waterfalls in an era of automation?
The Old Rules (and Why They Broke)
For years, the playbook was clear:
- Rule #1: Rank ad networks by historical eCPM
- Rule #2: Adjust manually based on country and placement
- Rule #3: Keep testing floors and demand partners
This worked when there were 3–5 major ad networks. But in 2025, with real-time bidding, hundreds of demand sources, and mediation layers powered by machine learning, the old rules lead to:
- Auction latency → Lower fill rate
- Lost impressions → When networks reject the request
- Manual chaos → UA teams wasting hours for marginal gains
Related: What Every UA Manager Must Know About SKAN 4.0
The New Rules of Ad Waterfall Optimization
1. Stop Thinking “Waterfall-First”
Hybrid setups dominate. According to Tenjin’s 2025 Monetization Report, 85% of top-grossing mobile games use hybrid mediation—where bidding runs first, and waterfalls catch the leftovers.
That means:
- Use waterfalls only for networks that don’t support bidding
- Set realistic eCPM floors (too high → lost impressions, too low → revenue leak)
2. Segment by More Than Geo
In 2025, LTV-based waterfalls outperform static ones. Instead of grouping by country only, segment by:
- Player LTV tier (whales vs casual players)
- Session depth (early vs late sessions)
- Device type (flagship vs low-end phones)
For example, a high LTV user in Tier 1 might see a different ad stack than a casual player in Tier 3.
IronSource and MAX by AppLovin offer advanced segmentation features for this.
3. Automate or Die
Manual waterfall optimization is dead. Mediation platforms like MAX and ironSource now offer automated A/B testing and adaptive floors.
Pro Tip: Don’t turn off automation entirely—use it to test floors dynamically while you monitor performance.
Check out: Unity Ads Mediation Best Practices
4. Optimize for eCPM + Fill Rate + Latency
Old waterfalls optimized for eCPM only. The new metric? ARPDAU (Average Revenue Per Daily Active User).
Why? Because high eCPM with low fill = bad UX and lost revenue.
Formula to watch:ARPDAU = (Total Ad Revenue) ÷ (Daily Active Users)
5. Playable Ads Change the Game
Playable ads aren’t just for UA—they’ve moved into monetization. Networks offering playables often deliver higher engagement and eCPM. If your waterfall ignores these formats, you’re leaving money on the table.
Read: Playable Ads: The Future of UA and Monetization
The Future: Full Automation?
By 2027, experts predict AI-driven mediation layers that eliminate manual waterfalls altogether. But until then, smart hybrid setups will remain the norm.
Key Takeaways
- Waterfalls aren’t dead, but hybrid setups win in 2025
- Segment by LTV, device type, and session depth
- Optimize for ARPDAU, not just eCPM
- Use automation and A/B testing—manual-only is a revenue killer
Actionable Checklist
✔ Use bidding for top networks
✔ Build LTV-based segments for waterfalls
✔ Enable automated A/B testing for floors
✔ Track ARPDAU and latency, not just eCPM
Conclusion
Ad Waterfall Optimization in 2025 is no longer just about stacking networks; it’s about data-driven decisions, real-time bidding, and maximizing every impression. With the rise of in-app bidding, AI-driven mediation, and advanced analytics, publishers who adapt to these new rules will dominate the monetization game. The future belongs to those who embrace transparency, leverage automation, and continuously test their waterfall strategy. Stay ahead, stay optimized, and make every impression count.
