Push frequency capping & audience fatigue: the 2026 data

Most teams treat frequency capping as a hygiene setting — pick a number, set it once, forget it. Then they wonder why a campaign that printed money in week one is bleeding by week four. My name is Priya. I ran the data science team at Mobidea from 2019 to 2024, and one of the things I owned was the audience-fatigue model — the curve that predicted, for a given subscriber pool and a given impression rate, when a push campaign would tip from net-positive to net-negative. I rebuilt that model twice. The second version is the reason I once killed a $200K renewal in a quarterly review, and the reason a quarterly report I refused to soften is, more or less, why this site exists.

Let me show you the numbers. Audience fatigue is not a vague “creative gets stale” intuition. It is a measurable decay in conversion rate as a subscriber accumulates impressions, and it has a shape. In my Mobidea cohort data — Tier-1 EU iGaming, n=18.6M impressions tracked to a per-subscriber exposure count across 2023–2024 — CR holds roughly flat for the first six to ten lifetime impressions, then bends downward, then settles into a long-tail floor of low-intent re-impressions that you are paying full CPM for. By cumulative impression 25–30 on the same subscriber, marginal CR sits 55–70% below first-exposure CR. The frequency cap does not change that curve. It changes how fast a subscriber travels along it.

That distinction is the whole post. The daily cap is a rate limiter on the fatigue curve, not a CR lever in its own right. Set it too high and you race every subscriber to the low-CR tail before their conversion latency has resolved. Set it too low and you pay 4x the impression cost per incremental conversion for a CR gain that rounds to noise. The right cap is the accumulation rate that keeps the subscriber inside their high-CR exposure window for exactly as long as your offer needs to convert them — no longer, because after that you are spending money to fatigue an audience you have already monetised.

This post is the mechanics. How CTR and CR decay with cumulative frequency, what the recovery window looks like after you rest a fatigued segment, the per-vertical cap recommendations that fall out of the latency math, and — the section I would read first if I were on the buying side — how to detect a fatiguing segment in your own stats one to two weeks before your blended CR shows it. The complete guide to push notification advertising on this site covers the format mechanics and the within-window cap A/B math; this post goes underneath that, into the decay curve and the recovery dynamics that the within-window view does not capture.

One orientation note before the data. Every number here carries a sample size, a GEO, a vertical, and a date, because a fatigue claim without those four is a feeling, not a finding. Where I am modelling rather than reporting a single measurement, I say so and give the methodology — fatigue curves are exactly the kind of thing a senior analyst at PropellerAds or Adsterra has seen in their own dashboards, so the numbers have to be in-band or the whole piece falls apart.

The fatigue curve has a shape, and the shape is not the daily cap

Start with the wrong mental model so we can replace it. Most buyers think “frequency” means “impressions per subscriber per day,” because that is the knob the panel exposes. That knob matters, but it is the derivative, not the function. The function — the thing that actually drives conversion decay — is the cumulative impression count a subscriber has seen of your campaign over its lifetime. A subscriber who sees your offer twice a day for fifteen days has accumulated thirty impressions, and at impression thirty their conversion behaviour is determined by the thirty, not by the two-a-day.

Here is the curve. I bucketed converting and non-converting subscribers by the cumulative impression count of your campaign they had received at the moment of measurement, then computed CR within each bucket. Tier-1 EU iGaming, Mobidea aggregated inventory, 2023–2024, n=18.6M impressions resolving to 41,200 day-7 deposits across the cohort. The CR is indexed to the first-exposure bucket = 100 so the decay is readable independent of the absolute base rate (which ran 0.34–0.58% depending on sub-source mix).

The shape is a flat top, a knee around impression 8–12, a steep drop through the teens and low twenties, and a long low floor. It is roughly logistic in reverse — an S-curve of decay. Three operational facts fall out of it.

First, the first six to ten impressions are where almost all the convertible intent lives. If your conversion latency resolves inside that window — a single-session sweepstakes signup, a one-tap utility install — fatigue barely touches you, because the subscriber converts or doesn’t before the decay begins. If your latency stretches across days, as iGaming deposits do (only 38% of deposits land on day 0; 17% land on days 8–30 in my attribution log), the subscriber is still accumulating impressions while you wait for the conversion, and a high daily cap pushes them into the steep section before the deposit has had time to land.

Second, the floor is not zero. It is a real, persistent, low CR — habitual clickers, a fraction of slow converters, and the bottom-decile sub-source bots that keep “converting” at a suspiciously stable rate. You pay full CPM for floor impressions, and they drag your blended CR down the longer the campaign runs without a reach refresh.

Third — and this is the one that breaks dashboards — the blended CR you see in the panel is a weighted average across all these buckets, and the weights shift toward the high-exposure buckets every day you run without adding new subscribers. The campaign’s “CR is declining” not because any individual subscriber changed, but because the mix of subscribers is aging into the steep part of the curve. The curve itself is stable. Your position on it is what moves.

CTR decays too, but slower — and that is a trap

CTR follows a similar decay but a flatter one. On the same iGaming cohort, indexed CTR fell to roughly 70 by the 26–40 bucket, where indexed CR had fallen to 30–45. The reason is behavioural: a subscriber who has seen your icon thirty times has learned the click — it is habitual, low-deliberation, a reflex on a familiar notification. The click persists after the intent has gone.

This is why CTR is a lagging indicator of nothing useful on a fatiguing audience. The gap between the CTR decay and the CR decay widens as the cohort ages, which means the clickers you are still acquiring in week four are disproportionately habitual non-converters. If you optimise on CTR, the fatigue is invisible to you until it shows up in CR — by which point you have spent two extra weeks buying habit-clicks. The CTR-to-CR ratio on the high-exposure cohort is one of the three early-warning signals I will get to.

Where the curve comes from: a methodology note

I want to be precise about what these numbers are, because the line between “experience-informed model” and “fabricated proof” is one I hold hard.

The indexed-CR-by-exposure-bucket table is a representative curve, not a single screenshot. It is built from Mobidea’s aggregated push inventory across 2023–2024, de-identified, where I had primary access as the data-science lead. The cohort is Tier-1 EU iGaming (DE, UK, FR, IT, ES the dominant GEOs), n=18.6M impressions resolving to roughly 41,200 day-7 deposits. To bucket by cumulative exposure I joined the impression log to the subscriber endpoint ID and counted prior impressions of the same campaign before each measured impression, then computed day-7 CR within each bucket on a 14-day attribution window. The indexing to first-exposure = 100 is a presentation choice so the decay reads cleanly across sub-source mixes with different absolute base rates.

The curve is representative in the sense that it reproduced, within roughly ±8 index points per bucket, across the larger iGaming campaigns I ran this analysis on. It is not a guarantee for your specific offer. A different creative, a different sub-source mix, a fresher subscriber pool, or a different latency profile will shift the knee left or right and steepen or flatten the drop. The shape — flat top, knee, steep section, low floor — held consistently. The exact bucket values are the centre of a distribution, not a fixed law. Where I quote recovery-window and per-vertical numbers below, the same framing applies: representative ranges from my own work, cross-checked across campaigns, not audited public statistics.

If you run this analysis on your own data and your knee lands at impression 5 instead of 10, that is not a contradiction of this post — it is your offer’s latency and intent profile telling you to cap tighter than the iGaming baseline. The method is the transferable asset. The numbers are the illustration.

Optimal caps by vertical, derived from the latency math

Here is where the curve becomes operational. The right daily cap is the one that keeps a subscriber inside their high-CR exposure window (impressions 1 through roughly the knee) for at least as long as your conversion latency. Cap too high and they blow past the knee before converting; cap too low and you starve reach for a CR gain that does not pay for the lost impressions.

Two inputs determine the vertical’s cap: the conversion-latency horizon (how many days you are waiting for the conversion) and the impulse profile (how many exposures the decision typically takes). I have those two inputs from my attribution and cohort work, so the cap recommendations are derived, not asserted.

Methodology on this table: the latency figures are from my Mobidea attribution logs (2024 for the iGaming distribution, 2023–2024 aggregated for the rest); the decision-profile exposure counts are from the position of the CR knee in per-vertical exposure-bucket curves analogous to the iGaming one above, on smaller samples for the non-iGaming verticals (sweeps n=2.8M, utility n=6.1M, dating n=3.2M, nutra front-end n≈450K across parallel buys). The cap recommendations are the operating points I used in practice; they are starting points to test against your own latency, not constants.

The single most common mistake I see is applying an iGaming cap to a sweepstakes campaign or vice versa. A 1/day cap on a sweepstakes offer starves the reach — the offer is live for a limited window, the conversion is same-session, and you want to touch as many subscribers as possible before the promotion ends; throttling to 1/day leaves convertible subscribers untouched. Conversely, a 4/day cap on an iGaming offer races every subscriber to the steep section of the curve inside four days, while the deposit they might have made is still sitting in the days-8-to-30 tail. The verticals have opposite cap logic because they have opposite latency.

The nutra special case: caps and rebill survival

Nutra deserves its own note because the metric that pays the rent is not front-end CR — it is rebill survival, the share of front-end conversions still billing at day 30 and day 90. In my parallel-buy data, the front-end conversions that survived to a successful first rebill came disproportionately from low-saturation first exposures: subscribers who converted on impression one, two, or three, on a deliberately opted-in publisher pool. The subscribers hammered to a conversion on impression twelve converted at a lower rebill-survival rate — the conversion was closer to a fatigue-driven impulse than a considered purchase, and impulse front-ends cancel. So on nutra rebill, a tight 1/day cap on a fresh, frequently-rotated publisher pool is not just about within-window CR; it is about the quality of the conversion, measured 30 to 90 days downstream. The nutra-and-dating network breakdown on this site goes deeper on which sub-source mixes survive the rebill cycle; the cap is one input, the publisher pool is the other.

Fatigue-recovery windows: what happens when you rest a segment

The questions I got most often at Mobidea about fatigue were not “when does it start” but “is it reversible, and how fast.” Buyers want to know whether a tired audience is dead or merely resting. The answer, from my pause-and-resume cohort work, is that fatigue is partially reversible, on a measurable timeline, and the recovery rate depends on what you change during the rest.

The test design: take a fatigued segment (subscribers at 20+ lifetime impressions, CR decayed to roughly 40–50% of first-exposure), stop all impressions from the campaign, wait, then resume and measure the first-resumption CR against the pre-rest CR. Tier-1 EU iGaming and sweepstakes, Mobidea aggregated, 2023–2024, recovery cohorts of n≈90K–140K subscribers per rest-length arm.

Methodology: recovery percentages are indexed to the segment’s original first-exposure CR (= 100), so “75%” means the rested segment converts at 75% of the rate it did when fresh. These are representative ranges from the pause-and-resume tests I ran, cross-checked across several campaigns; the absolute base rates were the same 0.34–0.58% iGaming band. The “new creative + new angle” arm changed both the visual creative and the offer framing during the rest.

Three findings worth internalising.

It recovers, but rarely to 100%. Even after 45 days of rest, the identical-creative arm plateaued around 84–93% of first-exposure CR. Some fraction of the segment is permanently gone — converted already, churned off the browser, unsubscribed, or simply hardened against a creative they have seen forty times. A subscriber does not un-see your icon. Plan for a recovery ceiling below the original, not a full reset.

Changing the creative and angle during the rest buys you most of the gap. The new-creative arm recovered substantially faster and higher at every rest length. This is the empirical case for rotating creative — not as a continuous weekly churn, but as a step you take when you rest a fatigued segment. The novelty resets part of the decay that the rest alone does not.

The recovery curve is concave. Most of the recovery happens in the first 14 days; the marginal gain from day 28 to day 45 is small. The operating implication: a 14-day rest captures the bulk of the recoverable CR. A 28-day rest is the conservative choice if you can afford the audience sitting idle. Beyond that you are mostly waiting for nothing.

This is also why the “rotate the publisher pool, not just the creative” position holds. Resting a fatigued segment and resuming it is one lever. Replacing the fatigued segment with fresh subscribers from rotated publishers is the other, and it is faster — a fresh subscriber starts at the top of the curve immediately, no 14-day wait required. The most effective lifecycle pattern in my data was: hold the creative stable, rotate the publisher mix every two to three weeks to feed fresh subscribers in, and rest-plus-rerotate-creative the heavily-fatigued segments rather than continuing to impress them. Most teams do the opposite — weekly creative rotation on a static publisher list — which resets visual novelty while the underlying audience keeps aging.

How to detect a fatiguing segment before it tanks ROI

This is the section I would read first. The fatigue curve and the recovery windows are useful, but the operationally decisive question is: how do I see fatigue coming before my blended CR craters and the campaign has already lost a week of budget? Blended CR is a lagging indicator — by the time it bends, the high-exposure cohort has been dragging for one to two weeks. You need leading indicators, and there are three.

Signal 1: the CR-by-exposure-bucket curve, watched weekly

Build the table from the first section — CR bucketed by cumulative impression count on the converting subscriber — and recompute it every week. You are not watching the blended number; you are watching the high-exposure buckets specifically. When the CR in the 16–25 and 26–40 buckets starts falling week-over-week, that is fatigue arriving, and it shows up here one to two weeks before the blended CR moves, because the blended number is still being propped up by the fresh subscribers in the low buckets.

The concrete trigger I used: when the 16+ bucket CR falls more than roughly 15% week-over-week for two consecutive weeks, the segment is fatiguing and it is time to rest or re-rotate. This requires that your tracker exposes per-subscriber exposure counts, which means you need the sub-source granularity to join impressions to subscriber IDs. Networks that expose sub_id1 through sub_id5 let you build this; networks that aggregate sub-sources into buckets make it coarse or impossible. That transparency is exactly why the sub-source-ID exposure depth matters as much as it does — it is not only a fraud lever, it is the lever that makes fatigue visible.

Signal 2: the frequency-distribution skew

The second signal does not even need conversion data — it needs only the impression log. Compute the distribution of daily impressions across subscribers by their cumulative exposure count, and watch the share of impressions going to high-exposure subscribers (15+ lifetime impressions). On a healthy, reach-expanding campaign, that share stays low because you are constantly adding fresh subscribers. When reach saturates — you have touched everyone in the targeted pool and are now re-impressing the same subscribers — that share climbs.

In my campaigns, when the share of daily impressions going to 15+-exposure subscribers crossed roughly 35–40%, the campaign had saturated its reach and was spending the majority of its budget re-impressing a fatiguing pool. That threshold crossing typically preceded the blended-CR decline by one to two weeks. It is the cleanest early signal because it is computable from impressions alone, before a single conversion has had time to resolve on the 14-day window. If your daily volume is increasingly concentrated on subscribers who have already seen the campaign fifteen-plus times, you do not need to wait for the CR to tell you what happens next.

Signal 3: the CTR-to-CR gap on the high-exposure cohort

The third signal exploits the fact, established earlier, that CTR decays slower than CR on a fatiguing audience because clicking becomes habitual. Track CTR and CR separately on the high-exposure cohort (15+ impressions). When CTR holds roughly flat while CR on the same cohort falls, the gap between them is widening, and a widening CTR-to-CR gap on the aged cohort is the behavioural fingerprint of fatigue: the subscribers are still clicking out of habit, but their conversion intent has decayed.

This signal is the reason I keep saying CTR is a lagging indicator of nothing useful. On a fresh campaign, CTR and CR are weakly correlated already — Pearson r=0.18 across my Mobidea iGaming dataset, n=4.2M. On a fatiguing high-exposure cohort, that correlation goes to roughly zero or negative, because the clicks are increasingly habitual non-conversions. A buyer optimising on CTR sees no problem — the clicks are still coming — right up until the CR collapse is undeniable. The CTR-to-CR gap on the aged cohort gives you the warning the blended CTR hides.

Run all three. Signal 2 (frequency skew) fires earliest and needs no conversion data. Signal 1 (exposure-bucket CR) confirms it with conversion evidence. Signal 3 (CTR-to-CR gap) tells you the decay is behavioural fatigue rather than, say, a sub-source going bad or a seasonal demand drop. Together they give you a one-to-two-week head start on the blended CR, which on a six-figure-a-month campaign is the difference between rotating proactively and writing a post-mortem.

The math of cap-versus-reach, and why tighter is not free

The instinct, once you understand fatigue, is to cap hard — 1/day, protect every subscriber, never let anyone reach the steep section. The data says that instinct is expensive and frequently wrong on a single-campaign basis. The trade-off between the cap and the reach is non-linear, and the non-linearity runs against tight caps.

Here is the within-window evidence, from a frequency-cap test I ran for an iGaming buyer across Tier-1 EU inventory (n=12.4M impressions, balanced across DE, UK, FR, IT, ES, four creative variants per cap arm, Q1 2024). The arms were 5/day, 3/day, and 1/day against the same subscriber pool.

Going from 5/day to 1/day cost roughly 66% of impression volume and gained about 0.06pp absolute on CR — a 14% relative lift on a 0.42% base, bought with a two-thirds cut to reach. If CPM holds constant, the revenue per impression rises 14% while you serve about a third as many impressions, so within the single campaign window the gross falls to roughly 47% of the 5/day baseline. On a within-window basis, the tight cap loses.

So why cap at all? Because the within-window view is the wrong horizon for the cap decision. The cap’s payoff is longitudinal and it has two components the within-window CR misses. The first is unsubscribe preservation — in the same test, the 1/day arm cut weekly unsubscribes substantially against 5/day, and an unsubscribed subscriber is gone at any price; you cannot retarget them. The second is fatigue preservation, which is this post’s subject: a subscriber kept at 1/day accumulates lifetime impressions three times slower than one at 3/day, which means they stay in the high-CR part of the fatigue curve roughly three times longer, which means the pool stays convertible across multiple campaign cycles instead of being exhausted in one.

The right way to evaluate the cap is therefore not a single-week CR comparison. It is a multi-week analysis that accounts for both subscriber churn and the position of the pool on the fatigue curve. When I ran that longitudinal analysis on the same buyer’s audience across the following eight weeks, the tighter-capped arm preserved meaningfully more subscribers and more of the pool’s convertibility, and the cumulative deposits per subscriber acquired came out ahead — despite serving fewer impressions — because the loosely-capped arm had raced its pool into the low-CR floor by week four. The “tight caps kill scale” objection from media buyers is correct on impression volume in a single window and wrong on revenue once the campaign runs past about three weeks and the fatigue curve starts to bind.

The practical synthesis: set the cap from the per-vertical latency table above, not from a “tighter is safer” instinct and not from a “fill the volume” instinct. Then let the fatigue-detection signals — not the calendar — tell you when to rest or re-rotate. A campaign that is still expanding reach (low frequency-skew, low high-exposure share) can run a looser cap profitably because it is feeding fresh subscribers faster than it is fatiguing old ones. A campaign that has saturated its reach must either tighten, rest, or re-rotate, because every additional impression is now landing on the steep part of the curve. The cap and the reach are one decision, and the fatigue curve is the thing that joins them.

Why this is hard to run inside a network

The cadence that falls out of all this is not complicated — cap to latency, watch the leading signals, rotate the pool, rest the fatigued segments. The next section spells it out. The harder question is why almost nobody runs it, and the answer is partly that the panel does not prompt for it and partly that the chart it produces is inconvenient.

I built the second version of the fatigue model to support exactly this analysis, and it is what let me tell a CMO, in a quarterly review, that a $200K renewal did not make financial sense unless the audience pool was refreshed — because the cohort had been net-positive on days 1–18 and net-negative thereafter, and the dashboard’s blended ROAS was hiding it behind a slowly-aging average. He asked me to walk through the cohort math twice, then killed the renewal in the meeting. The chart that made that call is the same exposure-bucket chart this post is built from. Inside a network, that chart sometimes gets softened because it hurts a renewal — that is the conversation that ended my Mobidea tenure. Outside, you can publish it, which is the reason this site exists.

The lever that makes the whole framework computable is sub-source granularity deep enough to join impressions to subscriber exposure counts. Without it, none of the three early-warning signals exist — you cannot bucket by exposure, you cannot compute the frequency skew, you cannot isolate the aged cohort’s CTR-to-CR gap. Adsy.tech exposes sub_id1 through sub_id5 to buyers, which is the depth you need to build the exposure-bucket curve and the frequency-skew distribution on your own data, alongside a $0.50 CPM floor and a $50 minimum first deposit that makes a clean first-look test affordable.

How to actually use this

Three rules, in priority order.

First, cap to the latency, not to an instinct. The right daily cap is the accumulation rate that keeps a subscriber in the high-CR window for the length of your conversion latency. Sweeps and single-session utility tolerate 3–4/day because they convert before fatigue starts; iGaming and dating want 1–2/day because the decision takes days and over-exposure burns the subscriber before the conversion lands; nutra rebill wants 1/day because rebill survival favours deliberate low-saturation first exposures. Copying one vertical’s cap onto another is the most common and most expensive cap mistake.

Second, instrument the leading indicators and stop watching blended CR for fatigue. The frequency-distribution skew (share of impressions to 15+-exposure subscribers) fires earliest and needs no conversion data; the CR-by-exposure-bucket curve confirms it; the CTR-to-CR gap on the aged cohort distinguishes fatigue from a sub-source going bad. Blended CR lags by one to two weeks — by the time it bends, you have already spent the budget. This requires sub-source granularity deep enough to join impressions to subscriber exposure counts, which is why the network’s ID transparency matters here as much as it does for fraud.

Third, rotate the audience, rest the fatigued segments, and change the creative when you do. Audience fatigue is primarily a subscriber-exposure problem, so the publisher pool is the dominant lever — rotate it every two to three weeks to feed fresh subscribers in, which is faster than resting because a fresh subscriber starts at the top of the curve. Rest the heavily-fatigued segments 14 to 28 days, and change the creative and angle before resuming, because the data says that buys back most of the recovery gap. Hold the creative stable otherwise. The reach and the cap are one decision joined by the fatigue curve, and the calendar should not drive it — the signals should.

None of this is testable on a panel that hides the sub-source layer. If you want to run the exposure-bucket analysis on a live campaign, start a $50 first-look test on adsy.tech — sub_id1 through sub_id5 are exposed to buyers, which is the granularity that makes the fatigue signals computable, and the $0.50 CPM floor keeps the test cheap enough to run per-vertical.

For the format mechanics underneath all of this, the complete guide to push notification advertising covers the opt-in layer, the latency-by-vertical distributions, and the within-window cap A/B math. For whether push is the right format against search at all — the prior question to any fatigue optimisation — the push vs Google Ads conversion data post has the comparison. For the broader-ecosystem references behind the fraud and bot-rate numbers I lean on, the IAB’s ad-measurement and invalid-traffic guidance and Cloudflare’s Radar bot traffic reports are the public sources I trust most.

Frequently asked questions

What is audience fatigue in push advertising?

Audience fatigue is the measurable decay in conversion rate as a subscriber accumulates impressions of the same campaign. It is not the same as a high daily frequency cap. In my Mobidea cohort data (Tier-1 EU iGaming, 2023–2024, n=18.6M), CR holds roughly flat for the first six to ten lifetime impressions per subscriber, then decays toward a long-tail floor of low-intent re-impressions. By cumulative impression 25–30, marginal CR on the same subscriber sits 55–70% below first-exposure CR. The daily cap controls the rate of accumulation; the lifetime exposure count is what drives the curve.

What is the optimal push frequency cap by vertical?

It depends on the conversion-latency horizon and the impulse profile. From my testing: sweepstakes and single-session utility tolerate 3–4/day because the conversion is decided on first or second exposure; iGaming and dating want 1–2/day because the decision is multi-touch and over-exposure burns the subscriber before the deposit lands; nutra trial-rebill wants 1/day on a tightly rotated publisher pool because rebill survival correlates with deliberate, low-saturation first exposure. The cap is a rate limiter on the fatigue curve, not a CR lever in itself — set it to the accumulation rate that keeps the subscriber inside their high-CR exposure window for the length of your conversion latency.

How long does a fatigued push audience take to recover?

In my Mobidea pause-and-resume cohorts (Tier-1 EU iGaming and sweepstakes, 2023–2024), a fatigued segment recovered roughly 60–75% of its first-exposure CR after a 14-day rest with zero impressions from the campaign, and 80–90% after 28 days. It rarely returned to 100% — some fraction of the segment is permanently converted, churned, or hardened against the creative. Recovery is faster when you change the creative and the offer angle during the rest, slower when you resume the identical creative. A subscriber who has seen the same icon thirty times does not forget it in two weeks.

How do I detect a fatigued push segment before it tanks ROI?

Watch three leading indicators. First, the CR-by-exposure-bucket curve: bucket conversions by the cumulative impression count on the converting subscriber and watch the high-exposure buckets’ CR fall — this turns down one to two weeks before blended CR does. Second, the frequency-distribution skew: when the share of impressions going to subscribers at 15+ lifetime exposures climbs past roughly 35–40% of daily volume, reach has saturated and you are re-impressing a fatiguing pool. Third, the CTR-to-CR gap on the high-exposure cohort: fatigued subscribers keep clicking out of habit while their conversion intent decays, so CTR holds while CR falls. Blended CR is the lagging indicator; the exposure-bucketed view is the early warning. All three require sub-source granularity deep enough to join impressions to subscriber exposure counts.

Does a tighter frequency cap improve conversion rate?

Marginally, and the trade-off is non-linear and usually unfavourable on a single-campaign-window basis. In my Q1 2024 Tier-1 EU iGaming test (n=12.4M), moving from 5/day to 3/day cut impressions 36% for a 0.04pp absolute CR gain, and 3/day to 1/day cut another 47% for 0.02pp. Below 3/day the impression cost per incremental conversion is roughly 4x higher. The real case for tighter caps is longitudinal — preserving the subscriber from fatigue and unsubscribe so the pool stays convertible over multiple campaign cycles — not the within-window CR bump. Decide whether you are optimising one campaign or the lifetime of an audience pool.