How Long Does a “New Collection” Stay New in Advertising?

An analysis of Meta Ad Library activity in Portugal’s jewelry market suggests there is no single answer: small launches are often one-off bursts, while larger campaigns keep introducing new-collection ads for weeks — and keep those ads live even longer.

When I started looking at jewelry advertising data, I had a fairly simple question: how long does a new collection remain “new” in advertising?

I expected the answer to be a number — perhaps a week, perhaps a month.

Instead, the data kept pointing back to campaign structure. Some collections appear in Meta’s Ad Library through one or two ad records. Others are supported by dozens of variants launched over several weeks.

Pooling those very different shapes into one average or median hides a lot of what is actually happening.

In the final dataset, more than half of the reconstructed collection events contained only one to three ad variants. Yet those small events represented less than 9% of all ad variants associated with identified collection campaigns.

The campaigns carrying most of the observed advertising activity looked very different.

Distribution of 65 reconstructed collection events. “Ad variants” refers to Meta Ad Library ad records associated with an identified collection event, not spend or impressions.

There is no single “new collection” lifetime

Across all reconstructed events, the pooled median newness-messaging window is zero days. At first glance, that sounds like an answer. It is not.

For a single-ad event, a zero-day window is inevitable: there is only one observed start date. Even with two or three ads, several records can begin on the same day, producing no measurable span between the first and last newness-message start.

Once there is meaningful ad-variant rotation, the picture changes.

• 4–10 ad variants: 19-day median messaging span

• 11+ ad variants: 15-day median messaging span

So the useful question is less “How many days does a collection stay new?” and more “What does the newness lifecycle look like for different kinds of collection campaigns?”

Median messaging and exposure windows by collection-event size. Messaging measures the span between the first and last observed starts of ads carrying new-collection messaging; exposure runs from the first such start to the latest observed delivery stop.

The advertising outlives the creative refresh

A second pattern is easier to interpret: ads often remain in delivery after the last observed start of an ad carrying new-collection messaging.

Across the 59 events with usable stop-date observations, the median exposure tail is 15 days.

But 15 days should not be read as a stable rule. The middle 50% of observed exposure-tail values spans roughly 8 to 38 days overall.

That is the more interesting point: the newness-message refresh and the advertising exposure lifecycle are not the same thing. A brand may stop introducing fresh new-collection ads while the associated ads remain active for days or weeks afterward.

Exposure-tail distributions for 59 eligible collection events. The overall median is 15 days; dots show bucket medians and horizontal ranges show the middle 50% of observations. Exposure tail measures the time from the last observed newness-message start to the latest observed delivery stop.

A few large campaigns account for most of the observed activity

The event-size split also changes how raw ad counts should be read.

Only 12 of the 65 reconstructed events fall into the 11+ ad category. Yet those 12 events account for almost three quarters of all identified ad variants in the final dataset.

One collection can therefore generate dozens of ad records while another appears only once.

For competitive intelligence, those are different behaviors. One is about launch frequency; the other is about ad-variant amplification and campaign intensity.

Treating them as the same metric would overstate the activity of brands that simply produce more variants around each collection.

New-collection activity peaks differently depending on how you measure it

The reconstructed events are not evenly distributed through the year, but the peak depends on the metric.

March contains the most reconstructed events with observed newness messaging: 11 events and 112 associated ad variants.

October contains fewer events — 9 — but the highest number of associated ad variants: 165.

January is another useful example: only two reconstructed events fall there, yet together they account for 127 associated ad variants.

This reinforces the same distinction as the event-size analysis: event count and ad-variant intensity are related, but they are not interchangeable.

Seasonal planning around the holiday period is one plausible explanation for the late-year concentration, but the dataset cannot establish why it occurs. The pattern is descriptive, not causal, and it should not be treated as a universal rule for the Portuguese jewelry market.

Monthly distribution of 63 reconstructed events with observed newness messaging. Event month is based on the first observed launch or reveal signal where available, otherwise the first observed newness-message date. March has the most events (11), while October has the most associated ad variants (165 across 9 events). The result describes this dataset rather than a representative estimate of the wider market.

Multi-brand retailer events skew longer — but the pattern is uneven

A more tentative pattern appears when multi-brand retailers are separated from brand-led advertisers.

Retailer events have higher medians on both measures:

• median exposure window: 53 days vs 28 days

• median ad variants per event: 4.0 vs 2.5

Those point estimates do not tell the whole story. The interquartile ranges overlap substantially, and the retailer sample contains only 15 events across four advertisers. The pattern also varies considerably from one retailer advertiser to another.

There is an observability difference too, but it is specific to the resolution method. Deterministic URL-based identity resolution covers 4.9% of retailer candidate ads versus 37.4% for other advertisers. Once semantic resolution is included, the same disadvantage does not remain.

The observed difference may therefore reflect campaign behavior, data observability, advertiser mix, or some combination of these factors. I would treat it as a directional signal worth following, not a causal conclusion.

Exposure-window and ad-variant distributions for 50 brand-led/other events and 15 multi-brand retailer events. Retailer events have higher medians (53 vs 28 exposure days; 4.0 vs 2.5 ad variants), but the interquartile ranges overlap and the retailer sample spans only four advertisers. Directional, not causal; the difference may reflect campaign behavior, data observability, advertiser mix, or a combination of these factors.

How the analysis was built

This analysis started with Meta Ad Library records classified as potentially related to new collections. From there, the goal was to reconstruct collection-level events rather than treat every ad record as a separate launch.

Reconstruction pipeline from 1,207 candidate ad records to 65 temporal collection events. Collection identity was resolved deterministically for 384 ads and through semantic review for 259, yielding 643 resolved ads and 51 collection identities; 564 unresolved or non-specific records were set aside. Related ads were then grouped by collection owner, identity and time.

The final reconstruction contains:

• 1,207 candidate ad records

• 384 deterministic identity resolutions

• 259 semantic identity resolutions

• 643 ads with a resolved specific collection identity

• 51 distinct collection identities

• 65 reconstructed temporal collection events

That means specific collection identity was resolved for about 53.3% of candidate ads. Ambiguous cases were deliberately left unresolved rather than forced into a collection identity.

What counts as a “new collection” ad?

Ad text was classified using multilingual rules covering explicit phrases such as “new collection”, “nova coleção”, “nueva colección”, “nouvelle collection” and “nuova collezione”, together with narrower launch and reveal signals. Generic collection mentions and weak discovery language were treated separately.

How collection identity was resolved

Collection identity was derived from landing-page URLs and collection slugs where possible, then supported by advertiser and brand context, explicit collection names in ad copy, and semantic review of unresolved groups. Generic categories, product features and ambiguous references were excluded.

How collection events were built

Ads referring to the same collection, advertiser/page and collection owner were grouped into temporal events. A 30-day inter-ad gap was used to separate distinct launch or reactivation events after inspecting the observed gap distribution.

Key metrics

• Messaging window — time between the first and last observed starts of ads carrying new-collection messaging.

• Exposure window — time from the first observed newness-message start to the latest observed delivery stop among those ads.

• Exposure tail — time between the last newness-message start and the latest observed delivery stop.

Limitations

This is observational research, not a census of every collection launch. The main limitations are:

• collection identity was resolved for about 53% of candidate ads;

• explicit launch or reveal dates are available for only a minority of events;

• Meta Ad Library records do not directly reveal campaign strategy or creative intent;

• no spend, ROI or sales data is used here;

• multi-brand retailer attribution is structurally more complex because advertiser, product brand and collection owner can differ;

• the analysis covers one historical cohort, so a future cohort is needed to test whether these patterns repeat over time.

What I take away from the data

The original question — “How long does a collection stay new?” — turned out to be too simple.

A small launch may appear as one ad or a same-day burst. A larger campaign may introduce new-collection ads over two or three weeks. And even after those fresh starts stop, the ads themselves may continue running for another couple of weeks — sometimes much longer.

For me, the more useful competitive-intelligence framework is to separate launch frequency, ad-variant intensity and campaign longevity.

A brand launching 40 ad variants around one collection is not necessarily launching more collections than a brand using five. It may simply be amplifying each launch much more aggressively.

The next meaningful test will be a genuinely new cohort of collection launches gathered over the coming months, not another rerun of the same historical ads.

I’m building these market-level views as part of QuantumExplore, with the goal of turning public advertising data into something more useful than ad-by-ad browsing. If you work in jewelry or fashion retail, I’d be interested to hear whether campaign-lifecycle benchmarks like these would be useful in competitive planning.