CPM and ad frequency: when your audience stops converting

There is a scene that repeats in almost every account that has been running display campaigns for a few months. The ad started well: good volume, decent clicks, enough attributed sales to justify the budget. Six weeks passed and nobody touched it because “it works”. And yet that campaign’s ACoS climbed three points, the CPC ticked up a few cents and attributed sales fell, all without a single change on your end.

The most common explanation is not that the ad became bad. It is that the people you are showing it to have already seen it. Several times. In display advertising — and in the display side of marketplaces — the audience is a finite resource that depletes with use, and when it depletes the system does not warn you: it simply keeps buying more expensive impressions for a crowd that stopped reacting.

The two numbers that tell that story are CPM and frequency. Neither lives in the same place as ACoS, neither gets used as much as it deserves, and together they explain a slice of ad spend that many sellers write off without ever understanding it. This article explains what they measure, how they are calculated, where they appear on Amazon and MercadoLibre, and how to read the burned-audience signal before it takes a month of budget with it.

what CPM is and how it is calculated

CPM comes from cost per mille: cost per thousand impressions. It is the price you pay every time your ad is shown a thousand times, click or no click. The formula is spend ÷ impressions × 1,000. Spend 3,400 pesos while your ad shows 850,000 times and your CPM is 3,400 ÷ 850,000 × 1,000 = 4 pesos.

The difference from CPC is what you are buying, not how it is billed. With CPC you pay for proven attention: somebody clicked. With CPM you pay for presence: you appeared, noticed or not. That is why CPM is the natural unit of awareness campaigns and CPC the unit of demand-capture campaigns.

The interesting part is that CPM can also be calculated on campaigns billed per click, and there it becomes a diagnostic. If your Sponsored Products campaign bills on CPC, you can still divide its spend by its impressions to get the effective CPM: what it truly costs you to be present. That number only moves when your CTR or your CPC moves, which makes it a useful thermometer. The relationship between the three is direct: effective CPM = CPC × CTR × 1,000. An ad with a 9-peso CPC and a 0.4% CTR has an effective CPM of 9 × 0.004 × 1,000 = 36 pesos.

That identity is more useful than it looks. If your effective CPM rises while your CPC held steady, the cause is in CTR — that is, creative, price or relevance. If your effective CPM rises while your CTR held steady, the cause is in the auction: more competition came in, or your placement mix shifted. One division separates two problems that look identical in an ACoS report.

where CPM shows up on Amazon and MercadoLibre

On Amazon, the format explicitly billed by impressions is Sponsored Display with its vCPM option: cost per thousand viewable impressions. The difference from a plain CPM is the word “viewable”. According to Amazon Ads documentation, an impression counts as viewable when at least half of the ad was on screen for at least one second in the case of images, and two seconds for video. Amazon suggests starting from a reference bid of five dollars per vCPM, and the option has existed on the platform since 2022.

That nuance matters because it changes what you are buying. A classic CPM charges you for impressions served, including the ones nobody managed to see because the shopper scrolled straight past. A vCPM charges you only for those that cleared the viewability threshold. It is not a discount: it is a different unit, and comparing a CPM against a vCPM without adjusting is comparing apples to oranges.

On MercadoLibre, the developer documentation lists CPM among the metrics available for Display campaigns, alongside CTR, clicks and consumed budget, and describes that the advertiser can choose between a cost-per-click or a cost-per-impression model depending on their objective. For Product Ads, the documented metric set includes clicks, impressions, CTR, cost, CPC and ACoS, plus organic units and items. One recent change is worth noting: Meli’s documentation states that ad-level metrics endpoints were retired on 30 May 2026 and replaced by ad-group-level metrics endpoints. If you have a report built on the earlier structure, that is exactly the kind of change that breaks it silently.

As always with these platforms, it pays to confirm in the console and the official help before operating: both Amazon and MercadoLibre adjust metric names, bidding options and endpoints more often than anyone would like.

what frequency is and why it is harder to see here

Frequency is the average number of times the same person saw your ad in a period. It is calculated as impressions ÷ reach, where reach is the count of distinct people who saw it at least once. If 20,000 people saw your ad and you accumulated 70,000 impressions, your frequency is 3.5.

Here comes the uncomfortable part for a marketplace seller: marketplace platforms do not expose reach with the clarity that social networks do. On Meta or Google Display you open the panel and frequency is a column. On Amazon Ads and Mercado Ads, based on what is publicly documented, the metric you have at hand is the impression, not unique reach per campaign. Without reach there is no directly computable frequency.

That does not mean the phenomenon is absent; it means you have to measure it by its shadow. When an audience saturates, it leaves traces that are observable with what you do have:

• Stable or rising impressions with a steadily falling CTR. You appear as often or more often, and fewer people click. It is the clearest symptom.
• Rising effective CPM with no change in CPC. You are paying more to be present because your click rate deteriorated.
• Falling attributed sales with flat spend. The budget burns the same and returns less.
• Creative age. The simplest figure and the one nobody records: how many days has this ad been running unchanged?

None of the four signals proves saturation on its own. All four together, in the same campaign, in the same window, do. And the fourth is the one that helps most with the decision, because it is the only one you control.

what public measurements say about frequency

Provenance has to be very explicit here. The published benchmarks on frequency and ad fatigue come from programmatic display, Google Display, Meta and TikTok, measured mostly in United States and European markets. They are not Amazon Mexico or MercadoLibre Mexico data, and marketplace platforms behave differently: the shopper arrives with explicit purchase intent, not browsing content. Take them as an order of magnitude for how saturation behaves, not as thresholds applicable to your campaigns.

With that warning in place, 2026 compilations point to the following. In prospecting display, the recommendation is to cap at roughly five to seven impressions per person per week, with a reported CTR decline of about 18% for each additional impression above seven. In retargeting the recommended range drops to three to five per week, because fatigue accelerates faster when you show the same ad to someone who already visited you. For Google Display, similar ranges of three to five weekly impressions are cited, with CTR drops of 30% to 50% once exceeded while CPM keeps climbing.

On the social side the measurements are more dramatic: a reported CTR drop of about 45% after four repetitions on Reels-style placements, and CPM climbing 18% to 25% within a single week when the system insists on serving the same asset to someone who already scrolled past it twice. Those same sources propose a fatigue dashboard against a seven-day rolling baseline: a CTR drop of 15% or more, a CPM rise of 10% or more, and frequency above 3.5 on prospecting.

What is usable in all of this for a marketplace seller is not the numbers, which do not apply. It is the shape of the curve: saturation is not linear, it accelerates. The first repetitions cost little performance and the last ones cost a lot. Which is why the expensive mistake is not leaving an ad running one week too long, it is leaving it running a month too long.

Glossary: ACoS is ad spend divided by attributed sales; in a saturated audience it rises from the spend side rather than the creative side, which is why changing the bid does not fix it.

how to set your own threshold without the reach figure

If you cannot compute frequency, build a substitute. The recipe is boring and it works:

• Define a seven-day rolling baseline per campaign: mean CTR, mean effective CPM, mean attributed sales.
• Record each creative’s start date. Without that, no comparison has an axis.
• Set your own triggers from your account’s history, not from the ranges above: for instance, review whenever the seven-day CTR falls below that campaign’s historical lower quartile while effective CPM rises at the same time.
• Log what you did and when. Without a record, in three months you will not know whether the recovery came from the creative swap or from the season.

An example reading, with illustrative numbers. A display campaign starts at a 28-peso effective CPM and a 0.11% CTR. By week six the CTR runs at 0.07% and the effective CPM at 41 pesos, with the same daily budget and the same CPC. Monthly spend did not change, but the impressions you buy return 36% fewer clicks and each appearance costs 46% more. There is nothing to fix in the bid: what ran out is the audience. The action is not to bid differently, it is to change the asset or change the audience.

what to do when the audience is already burned

There are four exits, and they are not interchangeable.

Rotate the creative. The cheapest one and the first to try. Changing the main image, the crop or the message usually recovers part of the CTR because the system has something new to serve again. It is an effective patch with a limited shelf life: if the audience is small, the second rotation returns less than the first.

Widen the audience. If your targeting is very narrow, each person sees your ad many times by pure arithmetic. Opening the segment spreads the same impressions across more people and lowers real frequency without touching the budget.

Lower the budget. Counterintuitive but valid when the audience is intrinsically small — a niche, a small brand. If you buy fewer impressions per day against the same crowd, frequency falls on its own and cost per sale stabilizes. You sell a little less and earn more per sale.

Switch it off and rest it. The option almost nobody uses and that is sometimes correct. An ad paused for two or three weeks works better on return because the audience rotated. It carries an opportunity cost, so it is reserved for campaigns whose ACoS no longer closes under any arrangement.

Before choosing among the four, one step decides whether any of them is worth doing: check whether that product can absorb the current cost. A display campaign with a rising CPM can still be profitable on a wide-margin SKU and unsustainable on a thin one, even though both show exactly the same symptoms. Without the SKU’s margin beside it, the whole frequency diagnosis hangs in the air.

Glossary: real net margin is what remains after ALL costs — product, commissions, fees, shipping, VAT and advertising; it is what decides whether a rising CPM is still tolerable or no longer is.

reading CPM, frequency and margin in one place

The practical obstacle in all of the above is not conceptual: it is that the pieces live apart. Amazon’s impressions and spend are in its console, MercadoLibre’s are in Mercado Ads with its own metrics schema that just changed aggregation level, and the margin of the SKU you are advertising is in neither. Effective CPM gets computed by hand, the seven-day baseline gets assembled by hand, and the comparison against margin happens in a spreadsheet somebody updates when they can.

In iqseller that reconstruction stops being manual work. The Profitability module shows net margin per SKU with the COGS you loaded, commissions from the Amazon settlement and from MercadoLibre orders, FBA and Full fees, shipping, advertising, and the VAT and withholding breakdown, all organized along the Parent → Model → SKU tree. The Alerts module flags when a number leaves its normal range, which is exactly how a depleting audience reveals itself: not through a collapse but through a drift.

With that, the operational question changes. Instead of “why did this campaign’s ACoS go up?”, which has six possible causes, the question becomes “can this product still absorb what it costs to appear today?”. And that one has a numeric answer. When a campaign’s effective CPM and the net margin of the SKU it promotes are read on the same screen, the decision to rotate, widen, lower or switch off stops being intuition and becomes arithmetic.