Why historical CPM data can mislead publisher monetization decisions

Introduction: the problem with looking at yesterday’s CPM

For publishers operating in today’s programmatic advertising ecosystem, CPM (cost per mille) is one of the most frequently monitored monetization metrics. It is often used to evaluate inventory quality, compare demand partners, establish floor prices, forecast advertising revenue, and determine whether monetization strategies are working.

However, there is a fundamental problem with relying too heavily on historical CPM data: the CPM a publisher earned yesterday does not necessarily represent what its inventory is worth today, or what it can earn tomorrow.

A publisher might have generated a $4.50 average CPM last month and decide to establish a similar floor price for the current month. On paper, this seems reasonable. After all, the inventory has historically attracted buyers at that price.

But what if advertiser demand has declined? What if traffic composition has changed? What if the previous month’s CPM was driven by a seasonal campaign, a particular geography, or a small number of high-value impressions? What if the current auction environment has fewer competing bidders?

In these situations, historical CPM becomes an unreliable reference point.

Programmatic advertising is a dynamic marketplace. Every impression represents a new opportunity, and its value depends on multiple factors that can change continuously. Advertiser budgets, audience characteristics, campaign objectives, device types, geographic distribution, auction competition, ad formats, viewability, and supply quality all influence how much buyers are willing to pay.

Historical CPM data captures the outcome of previous market conditions. It does not necessarily capture the current market’s willingness to pay.

This distinction is important because monetization decisions based on outdated or incomplete pricing signals can reduce revenue, increase unfilled inventory, create inefficient floor prices, and prevent publishers from capturing available demand.

The objective should not be to ignore historical CPM data. It should be to understand its limitations and use it alongside current auction signals, inventory performance, and demand-side behavior—as part of broader publisher yield optimization.

Let’s explore why historical CPM can be misleading, where publishers commonly make mistakes, and how a more data-driven monetization strategy can improve overall yield.

1. CPM is an outcome, not a permanent value of inventory

One of the most common misconceptions in publisher monetization is treating CPM as an intrinsic value of an ad impression.

For example, if a publisher consistently earns $5 CPM from its US desktop traffic, it may assume that the same inventory should continue to generate approximately $5 CPM.

But an impression does not have a permanently assigned market value.

Its price is determined by the advertiser’s willingness to pay at the time the impression becomes available, the competition for that impression, and the conditions under which the auction takes place.

Consider two impressions from the same publisher:

• Impression A is served to a US-based user visiting a high-intent financial content page during a major advertising campaign.
• Impression B is served to a user from the same geography visiting a general entertainment page during a period of reduced advertiser activity.

Although both impressions originate from the same publisher, their commercial value can be significantly different.

The difference can become even larger when factors such as device type, browser, audience characteristics, consent availability, ad placement, and viewability are considered.

Historical CPM combines the outcomes of thousands or millions of such individual auctions into an average.

That average is useful for understanding past performance, but it hides the underlying variation.

A historical average tells publishers what happened across a particular set of impressions. It does not guarantee that future impressions will have the same value.

This is why publishers should evaluate CPM as a changing market outcome rather than a fixed inventory benchmark.

2. Advertiser demand changes continuously

Programmatic advertising demand is not constant.

Advertisers regularly adjust their budgets, bidding strategies, campaign priorities, targeting parameters, and performance objectives. These changes directly influence the competition for publisher inventory.

For example, an advertiser may increase spending during a product launch and reduce its campaign budget after achieving its acquisition targets.

Similarly, a brand may temporarily increase its investment in a particular geography, audience segment, or device category.

When advertiser spending increases, more buyers may compete for available inventory. When spending declines, the same inventory may receive fewer bids or lower-priced bids.

Consider the following hypothetical example:

Period	Average CPM	Bid requests	Bid participation
January	$4.80	10M	72%
February	$4.50	10.2M	65%
March	$3.60	10.1M	48%
April	$3.20	10.3M	42%

The publisher’s traffic volume has remained relatively stable, but CPM and bid participation have declined.

If the publisher continues using January’s $4.80 CPM as its pricing benchmark, it may establish floors that are disconnected from current demand conditions.

The result could be fewer eligible bids, lower fill rates, and lost revenue opportunities.

The important lesson is that CPM should be interpreted alongside demand participation.

Publishers should monitor:

• Number of active bidders.
• Bid participation rate.
• Average and median bid prices.
• Bid response rate.
• Win rate.
• Auction clearing price.
• Unfilled impressions.
• Demand partner contribution.

A declining CPM does not always indicate deteriorating inventory quality. Sometimes it simply reflects a change in advertiser demand.

3. Seasonality can create false CPM benchmarks

Advertising budgets follow seasonal patterns.

Certain periods of the year attract significantly higher advertiser spending because of holidays, shopping events, product launches, financial cycles, and annual marketing plans.

Examples include Black Friday and Cyber Monday, Christmas and New Year campaigns, back-to-school advertising, Diwali and other regional shopping periods, major sporting events, and end-of-quarter and end-of-year budget utilization.

During these periods, publishers may experience higher competition for impressions and increased CPMs.

The problem begins when publishers use these seasonal peaks as standard pricing benchmarks for the months that follow.

Imagine a publisher that generated an average CPM of $6.00 during December because of strong holiday advertising demand.

In January, advertiser budgets reset and campaign spending normalizes. CPM drops to $3.80.

If the publisher continues enforcing a $5.50 floor based on December’s performance, it may reject a substantial portion of otherwise valuable demand.

The publisher might interpret this as a demand partner performance issue when the underlying reason is simply a seasonal market adjustment.

Why annual averages are not always enough

Using an annual average can also be misleading.

For instance, a publisher may compare November’s CPM with the average CPM of the previous twelve months. However, this comparison does not account for the fact that November may naturally have stronger demand than several other months.

A better approach is to compare:

1. The current period against the immediately preceding period.
2. The current period against the same period in the previous year.
3. The current period against a seasonally adjusted baseline.
4. The current period against comparable weekdays and traffic segments.

Seasonality should be incorporated into forecasting and pricing models rather than treated as an unexpected exception.

4. Traffic composition can change even when traffic volume does not

One of the most overlooked reasons historical CPM becomes misleading is a change in traffic composition.

A publisher might report almost identical pageviews month after month while the actual characteristics of its audience change considerably.

For example: US traffic increases while traffic from lower-demand geographies declines; desktop traffic shifts toward mobile; search traffic is replaced by social or referral traffic; high-intent article traffic decreases; more traffic comes from pages with lower viewability; returning users decline while new users increase.

All these changes can influence CPM.

Example: same traffic volume, different monetization value

Consider a publisher generating 10 million monthly pageviews.

Previous month:

Geography	Traffic share	Average CPM
US	50%	$5.00
UK	25%	$3.50
India	25%	$1.20

Current month:

Geography	Traffic share	Average CPM
US	25%	$5.00
UK	25%	$3.50
India	50%	$1.20

Assuming equal monetizable impression volume per traffic share, the weighted average CPM changes as follows:

Previous month: (50% × $5.00) + (25% × $3.50) + (25% × $1.20) = $3.675

Current month: (25% × $5.00) + (25% × $3.50) + (50% × $1.20) = $2.725

Illustrative calculation, assuming comparable impression volume and other conditions.

The publisher has not lost traffic. Its monetization potential has changed because its audience distribution has changed.

If the publisher looks only at total pageviews and historical average CPM, it may incorrectly conclude that demand partners are underperforming.

This is why CPM analysis should always include geographic, device, traffic-source, placement, and inventory-level segmentation.

5. Historical CPM hides the difference between average and actual auction value

An average CPM can be deceptive because it combines impressions with very different auction outcomes.

Consider a publisher that generated the following results:

Impression group	Impressions	Average CPM
Group A	100,000	$8.00
Group B	500,000	$3.00
Group C	1,400,000	$0.80

The overall average CPM is approximately $1.91 ((100,000 × $8) + (500,000 × $3) + (1,400,000 × $0.80)) ÷ 2,000,000.

However, this figure hides a significant difference between its inventory segments.

Group A attracts substantially higher-value demand, while Group C contributes most of the impression volume at a much lower CPM.

If the publisher uses $1.91 as a universal floor price, it may make two mistakes: set floors too low for Group A, potentially leaving revenue on the table; set floors too high for Group C, unnecessarily reducing fill.

The solution is to establish pricing strategies based on meaningful inventory segments rather than applying one historical average across the entire publisher inventory.

Average CPM versus median CPM

Publishers should also understand the difference between average and median CPM.

Average CPM can be influenced by a small number of exceptionally high-value impressions.

For example, consider these five impression CPM values: $0.50, $0.70, $0.90, $1.00, $15.00. The average is $3.62, while the median is $0.90.

Using the average alone would significantly overstate the typical impression value.

Median CPM provides an additional perspective on the distribution of auction prices, particularly when inventory has highly variable pricing.

Neither metric should replace the other. Both help publishers understand the underlying distribution of demand.

6. A high historical CPM does not necessarily mean higher revenue

This is one of the most important distinctions in publisher monetization.

A high CPM does not automatically translate into higher total revenue.

Publishers sometimes increase floor prices because they want to improve CPM, without evaluating the impact on fill rate and total monetized impressions.

Consider two monetization scenarios:

Metric	Scenario A	Scenario B
Available impressions	1,000,000	1,000,000
Fill rate	85%	45%
Filled impressions	850,000	450,000
Average CPM	$2.50	$4.00
Estimated revenue	$2,125	$1,800

Scenario B generates a 60% higher CPM than Scenario A. However, Scenario A generates more total revenue because it monetizes a significantly larger number of impressions.

Revenue = (Impressions × CPM) ÷ 1,000

This example demonstrates why publishers should not optimize CPM in isolation. The real objective is to maximize total revenue from eligible inventory while maintaining acceptable demand quality and user experience—as covered in increasing ad revenue without adding more ads.

A floor price that increases average CPM but causes a substantial reduction in fill may negatively affect overall yield. Teams tracking revenue per thousand after fill usually report that as eCPM.

7. Floor prices based on historical CPM can restrict auction competition

Floor pricing is one of the areas where outdated historical data can cause direct monetization losses.

Publishers often establish floor prices based on previous auction performance.

For example, if a particular inventory segment historically generated $3.00 CPM, the publisher might establish a $2.50 minimum floor to protect its revenue.

This strategy may work under stable market conditions.

However, programmatic auctions operate on current demand, and buyer participation can change significantly.

If the current market is willing to pay $1.80 for a particular impression, a $2.50 floor can prevent that impression from being monetized.

The publisher may believe it is protecting inventory value, but it is actually restricting eligible demand.

Why static floors become inefficient

Static floors can create several problems:

1. They do not respond quickly to changing advertiser budgets.
2. They ignore differences between individual inventory segments.
3. They can reduce bid participation during weaker demand periods.
4. They may prevent smaller or emerging demand partners from competing.
5. They can increase unfilled impressions.
6. They may be based on historical averages that no longer represent current auction conditions.

The importance of dynamic floor pricing

A more effective strategy is to evaluate floor prices using multiple signals: recent bid distributions, current demand participation, historical clearing prices, fill rate, inventory type, geographic performance, device and placement characteristics, time of day and day of week, and advertiser demand trends.

A dynamic floor strategy should adapt to changing market conditions rather than assuming yesterday’s price is still appropriate. Practical guidance is in dynamic floors that do not starve fill.

It is also important to distinguish between a publisher’s target CPM and the minimum price applied to every individual auction. These are not necessarily the same thing.

8. Demand partner performance changes over time

Historical CPM is frequently used to evaluate and compare SSPs, ad exchanges, and other demand partners.

For example, a publisher might observe Partner A at $3.80 CPM, Partner B at $3.20, and Partner C at $2.90. Based on these results, Partner A may appear to be the strongest monetization partner.

However, this conclusion can be misleading without understanding how each partner performed under comparable conditions.

Partner A may have received more premium US desktop inventory, while Partner B received a larger share of mobile and lower-CPM geographic traffic. Partner A may also have performed exceptionally well during a particular seasonal campaign that is no longer active.

What publishers should compare instead

Demand partner evaluation should include bid response rate, bid participation, bid density, win rate, average winning CPM, net publisher revenue, fill contribution, timeout rate, latency, and revenue by geography, device, format, and inventory segment.

It is also essential to compare partners using equivalent reporting definitions. One partner’s reported CPM may represent gross bid value, another’s may represent gross revenue, and another’s may represent publisher-net revenue after applicable deductions.

Gross CPM versus net CPM

For publishers, the most meaningful commercial comparison is generally the actual net revenue received or accrued after relevant platform fees and revenue shares.

A partner generating $4.00 gross CPM with a 25% revenue share does not necessarily outperform a partner generating $3.40 gross CPM with a 10% revenue share. Illustratively: Partner A nets $3.00; Partner B nets $3.06. Partner B produces slightly higher publisher-net CPM despite reporting a lower gross CPM.

9. Ad formats have different demand and pricing dynamics

Another limitation of historical CPM is that publishers sometimes combine different ad formats into a single average.

Display, native, video, audio, connected TV (CTV), and digital out-of-home inventory do not operate under identical demand conditions.

Each format has different advertiser objectives, creative requirements, buying mechanisms, audience expectations, supply constraints, measurement capabilities, auction competition, and pricing models.

For example, video inventory may attract higher bids than standard display inventory because of its advertising format and campaign objectives. However, this does not mean every video impression is automatically more valuable than every display impression.

Video CPM can vary substantially depending on instream versus outstream placement, video duration, player size, completion rate, viewability, sound-on status, device type, geographic demand, and content category. On CTV, ad delivery path also matters—see SSAI vs client-side ad insertion.

A publisher should never use the historical CPM of one format as a direct benchmark for another. Instead, each format should have its own performance baseline and pricing strategy.

10. Viewability and inventory quality can change without obvious traffic changes

Not all impressions offer the same advertising opportunity.

Two publishers may generate the same number of impressions, but advertisers may value their inventory differently because of differences in viewability, placement, page engagement, content quality, and measurement signals.

An ad placed near the top of an actively viewed article may have a stronger opportunity to be seen. An ad placed far below the fold may have a lower probability of becoming viewable. A video placement with strong completion performance may attract different demand from one with frequent abandonment.

These differences influence the attractiveness of inventory to advertisers. On video and CTV, counting rules differ from display—see how viewability is counted on video and CTV.

Historical CPM can become misleading when a publisher’s inventory quality changes but its reporting continues to aggregate all impressions into one average.

Publishers should monitor viewability and related quality metrics alongside CPM, rather than assuming that stable impression volume means stable inventory value.

11. Latency, timeouts, and bid losses can distort CPM analysis

Technical performance is another important factor that is often overlooked during monetization analysis.

Programmatic auctions depend on timely bid responses. If a demand partner consistently responds after the auction timeout, its bids may never become eligible for the final auction outcome.

Consider a hypothetical example: a publisher sends 1 million bid requests to a demand partner. Seven hundred thousand receive a response; two hundred thousand are valid bids; one hundred fifty thousand arrive within the permitted auction window; fifty thousand arrive too late.

The publisher may see lower revenue and assume that the partner’s demand has weakened. But the actual issue could be response latency rather than advertiser willingness to pay.

Historical CPM alone cannot identify this problem. Tuning timeouts without starving demand is covered in Prebid timeouts that still let demand compete and when a JS tag is enough.

Publishers should distinguish between no-bid responses, empty responses, invalid bids, late responses, eligible bids, lost bids, winning bids, and unfilled impressions.

12. Privacy, consent, and identity signals can change buyer participation

Advertising technology continues to evolve around privacy requirements, consent frameworks, browser restrictions, and identity-related signals.

These changes can influence how advertisers evaluate and bid on publisher inventory.

Factors that may influence monetization include GDPR and applicable consent signals, US state privacy requirements, CMP implementation, GPP and other privacy-related strings, cookie availability, mobile advertising identifiers, app tracking restrictions, supply-chain transparency, ads.txt and app-ads.txt authorization, and sellers.json and SupplyChain object information. Bid requests carry many of these signals in OpenRTB; see GPP, TCF, and US state strings in the bid request and ads.txt, sellers.json, and schain.

A historical CPM benchmark cannot automatically account for these changes. Publishers should monitor the quality and availability of auction signals and identify whether a change in bid participation correlates with changes in privacy or identity-related inputs.

13. Small samples and outliers can produce misleading CPM benchmarks

Not all historical CPM data is statistically reliable.

A small inventory segment may generate only a few hundred impressions per day. If one or two unusually high-value campaigns win a large share of those impressions, the resulting average CPM can be significantly inflated.

For example, a publisher receives 1,000 impressions from a particular segment: 990 at $1 CPM and 10 at $50 CPM. The weighted average CPM is $1.49.

If the publisher uses this figure to establish future floors, it may overestimate the segment’s typical monetization value.

The problem becomes more significant when historical reports use very short reporting windows, small impression samples, inconsistent traffic segments, unfiltered outliers, incomplete revenue data, or different reporting time zones.

How to improve statistical reliability

Publishers should consider minimum impression thresholds before establishing pricing benchmarks; median CPM alongside average CPM; percentile analysis (P25, P50, P75, P90); rolling averages; outlier detection; confidence intervals where sample sizes permit; and separate reporting for low-volume inventory segments.

A pricing decision should not depend on a small number of exceptional auction outcomes.

14. Historical CPM does not reveal the opportunity cost of an unfilled impression

An unfilled impression is not simply an impression that generated zero revenue. It can also represent a missed opportunity to monetize available inventory.

When a publisher applies a floor price that is too high, eligible bids may be rejected. When a demand partner times out, its potential bid may never enter the auction. When a particular buyer does not support a format or geography, the impression may remain unmonetized despite the presence of other potential demand.

Historical CPM reports typically focus on impressions that were successfully monetized, depending on the report’s metric definition.

This creates an important analytical gap: the publisher may see a healthy CPM among filled impressions while overlooking a growing number of unfilled opportunities.

For this reason, CPM must be evaluated alongside eligible ad requests, matched requests, bid requests, bid responses, winning bids, served impressions, unfilled impressions, and fill rate.

Publishers should also be careful not to treat fill rate as a single universal metric. Its denominator must be clearly defined.

The objective is to understand how much of the available monetizable inventory is being captured and at what effective price.

15. Reporting delays and revenue adjustments can distort historical comparisons

Another reason historical CPM can be misleading is that reporting data is not always final immediately after an auction takes place.

Depending on the platform and reporting pipeline, publishers may encounter delayed impression reconciliation, late-arriving conversion or event data, revenue adjustments, invalid traffic deductions, partner-side reporting corrections, currency conversion differences, time-zone discrepancies, and gross-to-net revenue adjustments.

For example, a publisher may initially observe a $3.50 CPM for yesterday’s traffic. After reconciliation and revenue adjustments, the finalized value may be $3.20.

If a pricing model uses preliminary reporting data without accounting for adjustments, it may continuously learn from inaccurate historical values.

A robust monetization platform should distinguish between estimated, reported, reconciled, and finalized revenue. Transparency expectations for sellers are discussed in MRC auction transparency.

16. Why one historical CPM benchmark cannot work across all inventory

A publisher’s inventory is not a single homogeneous product.

It consists of multiple segments, each with different demand characteristics and monetization potential: US desktop display, US mobile web, UK video, Indian native, in-app banner, in-app interstitial, CTV inventory, and more.

Applying a single CPM benchmark across all these segments is unlikely to produce optimal results. The more diversified the publisher’s inventory, the more important segmentation becomes.

A better approach is to establish separate monetization baselines based on combinations of geography, device, environment, ad format, placement, content, traffic source, demand source, auction type (open auction, PMP, preferred deals), and time. Deal mechanics are compared in open auction vs PMP vs preferred deal vs programmatic guaranteed.

Dimension	Suggested segmentation
Geography	Country, region, and market tier
Device	Desktop, mobile, tablet, connected TV
Environment	Web, mobile app, CTV
Ad format	Display, native, video, audio
Placement	Above the fold, below the fold, in-content
Content	News, finance, entertainment, sports
Traffic source	Search, direct, social, referral
Demand source	SSP, exchange, direct demand
Auction type	Open auction, PMP, preferred deals
Time	Hour, weekday, month, season

The objective is not to create unnecessary complexity. It is to avoid treating fundamentally different inventory as if it had identical market value.

17. A better approach: combine historical data with real-time auction intelligence

Historical data is still valuable. It is essential for identifying trends, understanding recurring patterns, forecasting demand, and detecting unusual changes.

The problem is relying on it without considering current market conditions.

A more effective monetization strategy combines three different categories of information.

A. Historical performance data

Historical data helps publishers understand long-term CPM trends, seasonal demand patterns, geographic performance, device-level monetization, demand partner contribution, average and median winning prices, fill-rate trends, and revenue distribution. It provides the foundation for understanding what has happened.

B. Recent performance data

Recent data helps identify whether historical patterns are still relevant. Monitor last 24-hour, 3-day, 7-day, 14-day, and 30-day performance. Comparing multiple windows is generally more useful than relying on one reporting period.

C. Real-time auction signals

Real-time data provides insight into current buyer behavior: incoming bid prices, bid participation, number of unique bidders, bid response latency, no-bid rates, winning bid distribution, auction clearing prices, floor rejection rates, auction timeout rates, and demand partner availability.

These signals help publishers understand what buyers are willing to pay now, rather than assuming previous auction outcomes will continue. Auction mechanics in first-price auctions and bid shading also affect clearing prices.

The three-layer monetization model

Data layer	Primary purpose	Key question
Historical	Establish benchmarks	What happened previously?
Recent	Detect market changes	Is performance improving or declining?
Real-time	Evaluate current demand	What is the market willing to pay now?

Combining these three layers creates a more reliable foundation for pricing and yield optimization. Reducing intermediary cost without sacrificing demand quality ties to supply path optimization.

18. How publishers can build a more intelligent CPM strategy

Moving away from historical CPM dependency requires more than changing floor prices. It requires a systematic approach to monetization analytics.

1. Establish inventory-specific benchmarks — not one site-wide CPM for every segment.
2. Use rolling performance windows — compare the latest seven days against the preceding seven and the same period last year.
3. Track fill rate alongside CPM — any floor adjustment should be evaluated against total revenue.
4. Monitor bid participation and auction competition — diagnose from underlying auction data.
5. Analyze gross and net revenue separately — especially with multiple SSPs and exchanges.
6. Review floor prices regularly — avoid carrying forward historical floors by default.
7. Run controlled experiments — test floor strategies on comparable segments with control groups.
8. Monitor demand partner concentration — diversify when incremental net revenue justifies it.
9. Build anomaly detection into reporting — sudden CPM or fill changes should trigger investigation.
10. Evaluate revenue per opportunity, not just CPM — define the denominator consistently across systems.

Implementation paths vary by stack—Prebid, tag, or server-side OpenRTB integrations—but the measurement discipline is the same.

19. The role of machine learning in CPM and yield optimization

As programmatic advertising matures, publishers increasingly use machine learning to complement human yield management—not to replace judgment with a black box.

Models trained on recent bid distributions can suggest segment-level floors that respond faster than monthly spreadsheet updates. Anomaly detection can flag sudden drops in bid participation or spikes in unfilled inventory before they show up as a “bad month” in a historical CPM report.

ML works best when inputs are clean: consistent segment definitions, reconciled net revenue, and separation of technical losses (timeouts, invalid bids) from true demand weakness. Historical CPM remains a feature in these models, but as one signal among many—not the target the system chases in isolation.

Automation should follow reliable measurement, as outlined in where publishers should focus on yield.

Final takeaway

Historical CPM is a record of past auction outcomes. It is not a permanent price tag on inventory, a substitute for current bid participation, or a reliable sole input for floor pricing across diverse segments.

Publishers who combine historical trends, recent performance windows, and real-time auction signals—and who read CPM together with fill rate, net revenue, viewability, and latency—make monetization decisions that reflect how programmatic markets actually behave.

Use yesterday’s CPM to learn what happened. Use today’s auction data to decide what to do next.

About Amli Media

At Amli Media, publisher monetization is built around transparency, efficiency, and measurable value. The programmatic infrastructure connects publishers with multiple demand sources across display, native, video, audio, CTV, and other digital formats, and gives publishers visibility into how their inventory is monetized.

The objective is simple: help publishers make every valuable advertising opportunity count.

FAQ

Can yesterday’s CPM set today’s floor price?

Not reliably on its own. Historical CPM reflects past auction conditions. Floors should follow recent bid distributions, fill rate, and segment-level demand—not a single blended average from an earlier period.

Why can a higher CPM mean lower revenue?

Revenue equals monetized impressions times effective price. A floor that raises average CPM but cuts fill can reduce total revenue. Optimize for net yield, not the CPM on impressions that still sell.

How does seasonality distort CPM benchmarks?

Holiday and event spending can inflate CPM for a few weeks. Using that peak as a year-round floor rejects demand when budgets normalize. Compare like periods and seasonally adjusted baselines.

What should publishers track besides average CPM?

Bid participation, median clearing prices, fill rate, unfilled impressions, net revenue after fees, viewability, latency, and performance by geography, device, format, and demand partner.

When is historical CPM still useful?

For trend analysis, seasonality planning, partner comparisons on equivalent inventory, and forecasting—when combined with recent windows and real-time auction signals, not as the sole pricing input.

← All posts · Dynamic floors · Publisher yield optimization · What is eCPM? · For publishers