Floor Pricing Strategy for Mobile Apps: The Operator's Framework (2026)

Floor pricing strategy for mobile apps in 2026 has five levers: floor type (hard vs. soft), geo segmentation by demand tier, time-of-day and day-of-week patterns, per-format calibration, and recalibration cadence. Floors work differently in waterfall vs. bidding setups because the auction mechanics differ. The most common mistake is a uniform global floor that ignores geo demand depth. The second most common mistake is raising floors without monitoring fill rate at the per-network level. AI automation from AdMob, MAX, and CAS handles Tier 1 geos reasonably well but consistently under-optimizes for Tier 2 and Tier 3 markets where manual segmentation outperforms.

Where floors actually sit in the auction stack (waterfall vs. bidding mechanics)

Most operators who set floors in AdMob or MAX treat them as a single concept: a minimum CPM below which you will not fill. That framing is incomplete, and the gap between the framing and the reality is where floor strategy mistakes happen.

A floor in a waterfall and a floor in a bidding auction are enforced at different points by different parties, and that structural difference changes what the floor does to your revenue. In a waterfall, each tier has a CPM threshold. A network fills at that tier when its historical CPM clears the threshold. When it does not, it falls to a lower tier where a lower threshold applies. The floor is a position gate: misset the floor at one tier and the impression falls through. It does not vanish.

In a bidding auction, there is one floor for the entire auction. All bidders compete and any bid below the floor is rejected. When the floor is set above the clearing price, all bids are rejected and the impression goes unfilled with no fallback. The failure mode is structurally different. In a waterfall, a wrong floor at one tier costs you that tier's fill. In bidding, a wrong floor costs you the whole impression.

For publishers running GAM as their ad server alongside a mediation stack, Unified Pricing Rules add a third layer. A UPR floor applies uniformly across all demand types in the GAM auction. If you have a $1.50 UPR floor and a $1.20 mediation-level floor, the UPR wins. Any bid between $1.20 and $1.50 gets rejected even though your mediator would have accepted it. This conflict does not surface in standard mediation reporting. You see a fill rate drop and no clear cause.

AdMob's "Optimized" toggle is relevant here. When enabled, your manually set floor becomes a reservation minimum and AdMob's AXON model adjusts above it based on predicted demand. AXON's demand signal is deepest in Tier 1 geos. In Tier 2 and Tier 3, the model has less data to calibrate on. The toggle performs reasonably in US and Western Europe. It is less accurate where Google's demand data is thinner.

For the full structural comparison of waterfall and bidding mechanics, see Mediation Waterfall vs In-App Bidding.

Hard floors vs. soft floors (when each applies in-app)

The hard/soft floor distinction is real, but its practical significance in mobile mediation is narrower than web programmatic content suggests.

A hard floor rejects any bid below the minimum. If the floor is $2.00 and the highest bid is $1.95, the impression goes unfilled in bidding or falls to the next waterfall tier. Full stop. A soft floor is different: the auction accepts bids below the floor but attempts to shade winning bids upward toward the reserve price through the auction mechanism. In practice, most mobile mediation platforms do not support true soft floors at the mediation level. AdMob, AppLovin MAX, and LevelPlay all implement floors as hard minimums. Bids below the floor are rejected, not shaded. Soft floor behavior is more native to web programmatic auctions running on OpenRTB. When your mediation documentation says "CPM floor," treat it as a hard floor unless the documentation explicitly states otherwise.

Where the distinction does matter in-app: in GAM's Unified Pricing Rules. UPR floors function as hard floors across all demand types in the unified auction. For publishers running GAM alongside a mediation stack, understanding this is not optional.

Hard floors make sense when you have strong evidence of what a specific segment of your inventory is worth and enough volume that no-fill on underpaying impressions costs less than the CPM yield you preserve. Tier 1 rewarded video in the US is the clearest case. You have data, demand is deep, and the bidding competition is real.

Relax the floor approach when demand is thin and a no-fill costs more than accepting a slightly below-ideal CPM. Tier 2 and Tier 3 geos, early-stage apps with limited bid data, and low-CPM formats like banner and native all fit this profile.

AppLovin MAX applies floors simultaneously to both bidding and waterfall sources. A floor that is set too high cuts off both bidding and waterfall demand in one move, making the fill impact more severe than in a waterfall-only stack. LevelPlay floors apply uniformly across all bidding networks in the group without tier-specific control. MAX gives you more per-instance granularity than LevelPlay for this reason. See AppLovin MAX vs Unity LevelPlay for the full configuration comparison.

Geo-segmented floors (the lever that moves revenue at scale)

This is the section that competitor content skips. Geo segmentation is the highest-leverage floor adjustment available to most mobile publishers, and the failure mode when it is missing is one of the most common and most fixable revenue leaks in a mobile stack.

Advertisers value the same impression differently by geography. A US user generates advertiser demand at 3x to 8x the CPM of a Southeast Asian user on the same app with identical session behavior. A uniform global floor either leaves revenue on the table in Tier 1 geos (floor too low relative to demand depth) or destroys fill in Tier 2 and Tier 3 geos (floor too high for the competitive CPM range). There is no single floor number that is correct across both conditions at the same time.

The practical tier structure:

Tier 1 (US, UK, CA, AU, Western Europe): Demand depth is high, bidding competition is real, and floors can be set aggressively without material fill impact. Hard floor approach is defensible here. Use trailing 30-day average eCPM per format as your calibration basis and set floors at 60-70% of that average for rewarded video, 55-65% for interstitial.

Tier 2 (Scandinavia, Netherlands, Japan, South Korea, DACH): Strong demand but less depth than Tier 1. Set floors at 60-70% of Tier 1 equivalents for the same format. Monitor fill rate on any floor increase here more closely than in Tier 1. A floor change that holds fill in the US may not hold fill in Japan at the same CPM level.

Tier 3 (Southeast Asia, LATAM, MENA, Eastern Europe, India): Demand is thin and floor errors hurt fill more than they protect yield. Set floors conservatively at 40-50% of trailing average at most. Protect only your highest-value formats (rewarded video) with modest floors. Leave banners and interstitials relatively open. A fill rate that collapses in these geos after a floor change is almost always a floor strategy problem, not a network quality problem.

How to segment in practice: AdMob mediation groups support country-level floor configuration. MAX supports geo and region-level floor settings per waterfall instance and per bidding group. Three-tier geo segmentation (US, UK/EU, Rest of World) is the minimum viable structure for any app generating material revenue across multiple regions.

The copy-paste mistake is the most common floor error in multi-geo apps: setting the US floor and applying the same number to every other country. At a $3.00 US floor for rewarded video, you will have near-zero fill in Southeast Asia where competitive bids cluster around $0.50 to $1.50. This is not a network problem. It is a floor strategy problem. The diagnostic is simple: if fill rate for a format is materially lower in Tier 2/3 geos than in Tier 1 geos with the same floor applied, the floor is too high for those geos.

Bidlogic's 2023 analysis of per-country floor optimization found an average 7.79% ARPDAU increase in 74% of tested cases. The improvement came specifically from geo-specific calibration, not from a uniform floor change. That is the empirical case for segmentation.

If you are running a single floor across all geos, or if your non-US fill rates are significantly lower than your US fill rates on the same format, that is a floor segmentation problem and it is fixable. The free initial conversation is the right starting point. Book a call here.

Time-of-day and day-of-week patterns (real demand seasonality)

Advertiser demand is not uniform across the week or the day. Ad budgets are freshest on Monday and Tuesday, when weekly campaign budgets reset and competition is highest. Demand typically softens heading into the weekend for most advertiser categories. Within the day, US-targeted campaigns peak between 11am and 2pm Eastern and again between 7pm and 10pm Eastern, when user engagement and advertiser pacing overlap.

A floor set at a weekly average CPM is wrong for both peak and off-peak periods. During peak, your floor may be set too low. Advertisers are competing hard and would have bid above your floor anyway. During off-peak, your floor may be too high. Demand thins and impressions that would have filled at a lower floor go unfilled.

Most mobile mediation platforms do not support native time-of-day floor automation at the publisher level. AdMob's AXON optimization incorporates time-of-day signal in its floor model. MAX does not currently offer a native time-of-day floor control. The practical alternative for manual floor operators: set floors at a conservative weekly average and let bidding competition during peak push realized CPMs above the floor naturally. Do not set floors at peak-period CPMs. Those floors will hurt fill during off-peak periods and are not recoverable without a floor change.

Day-of-week floor sensitivity varies by advertiser category. Gaming apps with US-focused advertiser demand typically see a Thursday-Friday peak (end-of-week campaign pushes) and a Saturday-Sunday dip. Finance and shopping app inventory tends to peak Monday-Tuesday. Know which advertiser category your inventory primarily attracts before assuming the generic pattern applies.

Q4 seasonality is the single recalibration event most operators miss. Q4 (October through December) is the highest-demand period in mobile advertising, driven by holiday campaign budgets. Floors correctly calibrated for Q1 through Q3 may be 20-40% below where they should be in Q4. Raise Tier 1 floors in September in preparation. Drop them back in January after Q4 budgets deplete.

If you are running AXON-optimized floors, AXON's model incorporates seasonality and advertiser demand signals and adjusts automatically. You do not need to manually account for Q4. If you are on manual floors, the September raise is a high-priority calendar item. The post-ATT attribution environment has also shifted advertiser budget timing and eCPM curves for iOS inventory specifically. See SKAdNetwork 4.0 Conversion Value Setup for the broader context on how attribution changes affect iOS demand cycles.

The floor-too-high diagnostic (when floors strangle fill)

If your fill rate dropped after you raised floors and you are not certain whether the floor is the cause, run this diagnostic sequence before concluding anything.

The core mechanic: when a floor is raised above the clearing price in a bidding auction, all bids are rejected and the impression goes unfilled. When a floor is set above the CPM threshold at which a waterfall tier fills, the tier does not fill and the impression either falls to a lower tier or goes unfilled. Both failure modes look identical in aggregate: fill rate drops.

Step 1: Per-network fill rate vs. overall fill rate. If overall fill rate dropped but one specific network's fill rate dropped more than others, the floor may be too high for that network's competitive range. A fill rate drop that is uniform across all networks points to a global floor setting problem or a non-floor cause.

Step 2: Bid density below the floor. In GAM and some mediation platforms, pull bid distribution reports. If 60-70% of bids are clustering just below your current floor, the floor is above the competitive clearing price for a significant share of auctions. Reduce the floor to where the bid distribution has density. If raw bid data is not available, proxy it with fill rate vs. CPM correlation: a floor increase that produced a fill rate drop larger than the CPM increase would imply is overshooting.

Step 3: Geo-level fill rate decomposition. If overall fill rate is down but US fill rate held and Southeast Asia fill rate collapsed, the floor is too high for Tier 2/3 geos and the overall signal is being dragged down by those geos. The fix is geo segmentation, not a global floor reduction.

Step 4: Format-level isolation. A banner floor that is too high shows as a banner-specific fill rate drop. If rewarded video fill is holding and banner fill dropped after a floor change, the banner floor is the specific issue.

Step 5: SDK and adapter changes as a confound. A fill rate drop that occurs at the same time as a floor increase may not be caused by the floor. Check whether any SDK or adapter versions were updated in the same window. SDK and adapter version changes can introduce fill issues independently of floors, and conflating the two leads to wrong corrective action. Use the Mediation SDK Checker to confirm versions before attributing a fill drop to floors.

Once you have confirmed the floor is the cause: reduce the floor for the affected segment in increments of 10-15% at a time and monitor fill rate recovery per network over 7 days. Do not drop the floor to zero to recover fill and then attempt to raise it again quickly. Floor changes in bidding environments need 3-7 days for demand-side models to recalibrate to the new floor signal.

The ARPDAU test is the cleanest diagnostic overall. If eCPM went up 10% and fill rate dropped 15%, your ARPDAU went down. If eCPM went up 10% and fill rate dropped 8%, your ARPDAU likely held or improved slightly. eCPM in isolation does not tell you whether the floor change was worth it. ARPDAU does.

AI floor optimization (what AdMob, MAX, and CAS automation actually do)

Three major automation options exist for mobile floor optimization. Each deserves a plain description of what it does mechanically, not what the vendor says it does.

AdMob AXON optimization. When you enable the "Optimized" toggle in AdMob's CPM floor settings, your manually set floor becomes a reservation minimum and AXON adjusts above it in real time based on Google demand signals. AXON knows which advertisers are competing for your specific inventory, their bid density, and historical patterns by segment. The genuine advantage is that AXON has demand-side data visibility that you do not have from mediation report averages. It knows more about what advertisers will pay for your inventory than you do. The limitation: AXON is trained on Google demand, which is a subset of all demand competing for your inventory. For publishers where non-Google demand (AppLovin, Meta, Mintegral) makes up a large share of revenue, AXON's floor model is optimizing on partial signal. AXON also performs best in Tier 1 geos. In Tier 2/3, the model has less data and its floor recommendations are less accurate. Compare this to AdMob Mediation vs AppLovin MAX for the platform-level trade-offs.

AppLovin MAX auto-floors. MAX's automated floor recommendations are informed by AppLovin's own demand as a DSP. AppLovin has first-party signal on what advertisers are paying for MAX inventory. The same structural point applies: MAX floor automation optimizes on AppLovin's demand signal, which may not reflect the full competitive picture for publishers with a diverse demand mix. MAX automation works well when AppLovin demand is the dominant revenue driver. When other demand sources are primary, the optimization is working from an incomplete picture.

CAS.AI floor optimization. CAS's automated floor management is the most opaque of the three. The training data and optimization objective are not documented in operator-level detail publicly available as of 2026. Treat CAS floor automation as a starting point and validate against your ARPDAU data over 30-day periods. If ARPDAU improved after enabling it, the tool is working for your inventory mix. If ARPDAU held flat or dropped, manual segmentation will outperform it.

The structural limitation all three share: Every automated floor tool trains on sold impressions, not on unsold impressions or suppressed bids. A model that only sees CPM data for filled impressions is optimizing blind to inventory that went unfilled because the floor was too high. This is the same problem covered in the context of web programmatic optimization. Understanding how to evaluate whether an AI floor tool is showing real lift is covered in how to evaluate whether an AI floor tool is showing real lift. In mobile mediation, the problem is compounded because most mediators do not surface unsold impression data at the publisher level in standard reporting. The model cannot learn from revenue it never captured.

Where manual floors consistently outperform automation: geo-specific Tier 2/3 calibration, per-format floors for niche high-value inventory categories, and seasonal adjustments (the Q4 September raise that automation may lag by several weeks). In these specific cases, direct observation of your app's bid distribution and ARPDAU trends will outperform automated tools that are not granular enough for your specific inventory segment.

Calibration cadence (weekly, monthly, quarterly, and when to leave them alone)

Floors left unchanged for more than 90 days are almost certainly wrong for at least some segments of your inventory. Market demand shifts seasonally, new advertisers enter and exit, and competing apps change their supply patterns. A floor that was right in Q1 is not automatically right in Q3.

Weekly review: Rewarded video and interstitial floor prices in US, UK, CA, and AU. Any format where you recently made a floor change and are monitoring the fill rate response. All Tier 1 formats during Q4 (October through December).

Monthly review: Banner and native floor prices across all geos. Tier 2 geo floor prices for rewarded video and interstitial. Any format where fill rate changed more than 5 percentage points in either direction since the last review.

Quarterly review: Tier 3 geo floors. The question is not "should I raise this floor" but "is this floor currently costing me fill that matters?" Low-traffic ad units where weekly and monthly signal is too noisy to act on.

When to leave floors alone: If ARPDAU on a placement is stable or growing, fill rate is within a normal variance band (plus or minus 5%), and no external signal (Q4 season, major SDK update, new competing app in your category) has changed, do not touch the floor. Floor changes in bidding environments reset demand-side model calibration on your inventory. That reset has a short-term cost even when the new floor is the correct long-term level. Do not change floors speculatively.

The seasonal adjustment triggers: September 15 is the practical date to review all Tier 1 floors before Q4 demand ramps. January 15 is the date to review downward, as Q4 budgets deplete and January demand softens. These are not exact dates. They are the point after which waiting becomes costly.

The "floor drift" problem: publishers who have not changed floors in 6 to 12 months often have floors set against eCPM data from a different market environment. Post-ATT pricing normalization shifted mobile eCPM curves materially since 2022. Floors set in 2022 or 2023 may be too low for current Tier 1 demand or too high for Tier 2/3 geos where demand has thinned. The audit question is: when did you last change this floor, and what was your ARPDAU data basis for the current setting?

Use the AdMob Approval Checker for a structured audit of your AdMob mediation configuration. Use the Mediation SDK Checker to confirm your SDK versions are current before making floor changes. An SDK version issue can cause fills to fail independently of floors, and conflating the two leads to wrong decisions.

Floor strategy in bidding vs. waterfall setups (the structural difference)

The mental model most operators carry is waterfall-based: floors sit at each tier in the sequence, and a too-high floor at one tier just pushes the impression down the stack. That model breaks when you shift to a bidding-dominant stack, and most production stacks in 2026 are hybrid.

In a waterfall, a floor at one tier only affects that tier. Networks that cannot clear the tier floor fall to the next tier where a lower floor applies. The impression still has a path to fill. The waterfall is, structurally, a series of hard floors at each step in the sequence, with each tier acting as a sequential safety net.

In a bidding auction, one floor applies to the entire auction. All bidders compete. Bids below the floor are rejected. There is no lower tier to catch the impression. A single miscalibrated floor affects all demand sources simultaneously and the impression may go unfilled with no recovery path.

The hybrid case requires coordinating both layers. In a hybrid stack, the bidding auction floor should be set at or below the CPM of your highest waterfall tier in the same geo segment. If the bidding auction floor is above the top waterfall tier's CPM, then impressions that do not clear the bidding floor have no waterfall path to fill. You are cutting off a recovery mechanism that would have captured those impressions.

When all demand moves to bidding, floor strategy simplifies in one dimension (no tier-by-tier management) but becomes more consequential in another (a single floor miscalibration affects all demand). The number of floor decisions goes down. The stakes per decision go up.

Per-platform floor mechanics for operator reference:

• AdMob mediation: floor set per mediation group, per country group. Applies as a hard minimum to all demand in the group. AXON optimization available.
• AppLovin MAX: floor set per waterfall, per geo/region. Applies to both bidding and waterfall instances in the same stack. Per-instance floor control also available for granular management.
• LevelPlay (Unity): floor set per mediation group, applies uniformly to all bidding networks in the group. No waterfall tier-level floor control within the bidding group.
• GAM Unified Pricing Rules: applies across all demand types in GAM. Relevant primarily for publishers using GAM as the ad server alongside a mediation stack.

For the full structural comparison, see Mediation Waterfall vs In-App Bidding on how the auction mechanics differ between waterfall and bidding setups.

Per-format floor patterns (rewarded video, interstitial, banner, native, app open)

Setting the same floor across all formats is the fastest way to simultaneously leave money on the table in your highest-value formats and strangle fill in your lowest-value formats. The CPM ranges across formats do not overlap. The floor strategy that is correct for rewarded video is wrong for banner.

Rewarded video (US: $8-$18 eCPM at median). Set floors at 60-70% of your trailing 30-day average eCPM for this format in the US. A $12 average supports a floor of $7 to $8.50. Rewarded video is opt-in: a no-fill costs UX goodwill in addition to revenue. Be more conservative with floors here than the CPM data alone might suggest. For UK, AU, and CA, apply the same 60-70% logic against the local trailing average ($5-$12 range). For Tier 3, CPMs cluster around $0.50 to $2.00. Set floors at 40-50% of trailing average at most. Do not apply US rewarded video floors to Tier 3 traffic.

Interstitial (US: $4-$10 eCPM at median). More CPM variance than rewarded video due to format engagement differences. Floor at 55-65% of trailing average. Interstitials have higher fill tolerance than banners but lower than rewarded video. For Tier 3, CPMs run $0.30 to $1.50. Set low floors and prioritize fill over CPM maximization at this range.

Banner (300x250, adaptive) (US: $0.40-$1.20 eCPM at median). Banner CPMs are low and highly volatile. Setting a banner floor above $0.50 in most Tier 1 markets risks cutting fill for marginal CPM protection. The correct posture for banners is very conservative floors, or no floors at all for Tier 2/3. Focus optimization effort on rewarded video and interstitial first. Banner floor optimization returns less per hour of effort than any other format.

Native. Similar CPM range to banner, slightly higher engagement CPM in some categories. Floor logic follows the same pattern as banner: conservative floors, fill rate priority over CPM maximization.

App open (US: $2-$6 eCPM at median). App open is effectively a high-visibility interstitial. Floor logic is similar to interstitial, slightly more conservative given that app open is the first impression of the session. A no-fill on app open means the entire session starts with a missed impression.

The cross-format contamination mistake: publishers who manage all ad unit floors in a single mediation group with a single floor are applying the rewarded video floor (high) to banners (low CPM) and getting near-zero banner fill as a result. Keep format-level mediation groups and floor settings separate. If your mediator forces all formats into one group, that is a configuration constraint, not a floor strategy decision.

Format priority order for calibration effort: rewarded video first, then interstitial, then app open, then native, then banner. Spend floor calibration effort in that order.

Common floor strategy mistakes (and the specific fix for each)

If your floor setup covers more than one geo or more than one format, at least one of the following is active in your current configuration.

Mistake 1: Uniform global floors. One CPM floor applied to all countries and regions. The setup default in most mediation platforms is a single floor, and operators who are moving quickly do not segment. The cost: either Tier 1 floors set too low (leaving CPM on the table) or Tier 3 floors set too high (cutting fill). Usually both simultaneously. Fix: segment into three tiers at minimum (US, UK/EU, Rest of World) and set floors based on trailing eCPM data for each segment independently.

Mistake 2: Never recalibrating. Floors set during initial configuration and never changed. Floors are not surfaced in daily monetization reports. They live in mediation configuration settings, out of sight. Cost: misalignment with current market demand. Tier 1 floors set 12-18 months ago are often below current clearing prices. Q4 demand goes uncaptured because floors are still at Q2 levels. Fix: calendar a quarterly review minimum. Monthly for Tier 1 high-revenue placements. Use trailing 30-day average eCPM per format per geo as the calibration baseline.

Mistake 3: Copy-pasting floors between geos. Applying the same floor numbers across US, EU, and Southeast Asia because geo-segmented configuration is tedious. Cost: fill destruction in Tier 2/3 geos. A $3.00 US rewarded video floor applied to Vietnam or Indonesia produces near-zero fill. Publishers then conclude "this country just does not monetize" when the real cause is a misconfigured floor. Fix: set floors per geo tier. If your mediation platform limits country-level configuration, group by demand tier (Tier 1, Tier 2, Tier 3) and set three floor levels rather than one.

Mistake 4: Ignoring the per-network competitive baseline. Setting floors based on aggregate eCPM averages without checking whether specific networks are systematically bidding below the floor. Cost: a volume-filler network at $0.80 CPM may be contributing meaningful fill on impressions your $1.50 floor is rejecting. Or a network that averages $2.50 CPM may deliver $3.50 during peak periods but your $4.00 floor cuts it out of the auction. Fix: pull per-network eCPM and fill rate data separately for each demand source. A network with a fill rate that dropped significantly after your last floor change is the signal of a floor set above that network's competitive range.

Mistake 5: Using the same floor for waterfall and bidding sources. Setting a floor in a hybrid stack that applies equally to bidding and waterfall tiers without accounting for the structural difference. The floor interface in mediation platforms does not always make it obvious that a single setting applies differently to bidding sources and waterfall tiers. Cost: a floor high enough to protect rewarded video eCPM in bidding will often cut fill from waterfall sources that historically cleared at a lower CPM. The publisher sees fill rate drop and concludes the network is underperforming. The floor is the actual cause. Fix: in a hybrid stack, set the bidding auction floor at or below the top waterfall tier CPM so you are not cutting off the waterfall recovery path with the bidding floor.

If your floor setup covers more than one geo, more than one format, and includes both bidding and waterfall demand, you probably have at least one of the above mistakes active in your current configuration. The free initial conversation is the practical starting point for figuring out which one. Book a free 30-minute call.

Frequently Asked Questions

What is the difference between a hard floor and a soft floor in mobile advertising?

A hard floor is a firm minimum CPM: any bid below it is rejected and the impression either goes unfilled or falls to the next waterfall tier. A soft floor is a reserve price below which the auction engine tries to shade winning bids upward, but it accepts bids below the reserve if no bid reaches it. In practice, most mobile mediation platforms including AdMob, AppLovin MAX, and LevelPlay implement floors as hard minimums. Bids below the floor are rejected, not shaded. True soft floor behavior is more common in web programmatic auctions running on OpenRTB than in SDK-based in-app mediation. When your mediation documentation says CPM floor, treat it as a hard floor unless stated otherwise.

Should I use AI floor optimization or set floors manually?

Use AI optimization (AdMob AXON, AppLovin MAX automation) for Tier 1 geo segments where the mediator has deep enough demand data for the model to calibrate accurately. Set floors manually for Tier 2 and Tier 3 geos, where AI tools are trained on thinner demand data and consistently under-optimize. Manual floors also outperform automation for seasonal adjustments such as Q4 raises and January resets, where the model may lag the market by several weeks. The correct approach for most production stacks is hybrid: AI optimization for US and major Western Europe segments, manual segmented floors for everything else.

How often should I recalibrate my floor prices?

Monthly for Tier 1 high-revenue placements such as rewarded video and interstitial in US, UK, Canada, and Australia. Quarterly for secondary formats and Tier 2 geos. Two mandatory annual events: raise Tier 1 floors in mid-September before Q4 demand ramps, and review downward in mid-January after Q4 budgets deplete. Do not recalibrate more frequently than weekly on any placement. Floor changes in bidding environments reset demand-side model calibration and carry a short-term cost even when the new floor is the correct long-term level.

Why is my fill rate dropping after I raised floors?

When a floor is raised above the clearing price for a segment, bids in that segment are rejected and the impression goes unfilled. In a bidding auction there is no lower tier to catch the impression, so it simply goes unfilled. Diagnose by decomposing the fill rate drop by network, geo, and format before concluding the floor is the cause. A uniform fill rate drop across all networks typically points to a global floor issue. A drop for specific networks only may be a bid density problem or an SDK change that coincided with the floor change. The metric to track is ARPDAU, not eCPM in isolation. A floor increase that raised eCPM 10% while dropping fill 15% reduced net revenue even though the CPM number looks better.

Do I need different floors for waterfall and bidding setups?

Yes. In a waterfall, a floor at one tier only affects that tier and impressions that do not clear fall to the next tier. In a bidding auction, a floor that is not cleared means the impression goes unfilled with no fallback. In a hybrid stack, the bidding auction floor should be set at or below the CPM of your top waterfall tier so that impressions that do not clear the bidding floor still have a waterfall path to fill. Setting the bidding floor above your top waterfall tier CPM cuts off a fill path that would have captured those impressions.

What is a reasonable floor price for an interstitial in 2026?

For a US interstitial, a floor in the $2.50 to $4.00 range is defensible for most apps, based on median US interstitial eCPMs of $4 to $10. Set the floor at 55 to 65 percent of your trailing 30-day average eCPM for that placement in that geo. This protects against below-market bids while keeping the floor below the competitive clearing price for most auction cycles. For UK and major Western Europe, floors in the $1.50 to $3.00 range follow the same logic. For Southeast Asia, LATAM, and MENA, floors above $0.50 to $0.80 will destroy fill for most apps. There is no universal number. The correct floor is always anchored to your specific app's trailing demand data, not a generic benchmark.

When to bring someone in

Floor prices are not set-and-forget configuration. They are a revenue lever that requires the same attention as any other yield variable in your mediation stack. Getting them right across geos, formats, and auction types is the kind of structured work that pays for itself quickly.

If your floors have not been reviewed in more than six months, if you are running a single floor across all geos, or if you raised floors recently and ARPDAU did not improve, those are diagnostic signals, not edge cases. Most floor problems are fixable once the right questions are applied to the right data.

If you want to run a floor audit against your actual stack rather than against benchmarks, that is what the free initial conversation is for. Book a free 30-minute call.