Revenue grew. Margins collapsed anyway. Eight K-beauty D2C brands · ad spend vs. operating margin · public-data analysis

01What happened

Read the sector indicators and K-beauty is thriving. Cosmetics exports reached $11.4bn in 2025 (+12.2%), second in the world, and the trade surplus cleared $10.1bn for the first time. I expected the individual income statements to say something similar.

They did not.

Whether growth came in at +68% or −14%, margin moved the same way. The deterioration is indifferent to growth — that is what the table is for.

02"The whole sector is like this" is false

I heard "everyone is having a hard year" more than once, so I went to the aggregate first. It said the opposite. Across 82 cosmetics companies, operating margin improved by 0.38pp. Narrow the field to the eight brand companies and the median decline is 3.2pp.

A company that shed 16pp therefore lost five times the sector median. A gap that size was manufactured inside the company, not by the market.

More decisive still: under the same conditions, several companies moved the other way.

• APR — grew revenue 111% and raised operating margin 6.96pp.
• D'Alba Global — grew 68%, the same band as Company A above, and held margin at +0.18pp. Its investor materials name "marketing and cost-structure optimization in key regions" as a formal objective for 2026.
• Konny by Erin — a different category (infant D2C), the same structure. It cut advertising from 10.65% to 9.13% of revenue while growing 64% and lifting margin 9.17pp. Nine consecutive profitable years, no outside capital.

"High growth costs you margin" is not a law of this sector. It is a question of execution, company by company.

03So advertising is the culprit — half right

Of the six that declined, advertising drove four. The others do not fit. Manyo Factory in fact cut advertising 8.5%; its decline came from a higher cost ratio on falling revenue, plus one-off charges. Silicon2 runs a distribution-margin model where advertising is barely 1% of revenue.

Assume the ad-spend explanation up front and you will misread companies like these entirely. The diagnosis has to be rebuilt for each one. And even for the four where advertising did drive it, the precise question is not "spend less."

Until you can separate the two, raising the budget and cutting it are both bets.

04Why did no one measure it

This is the part I was actually curious about, since financial statements only report outcomes. So I read every open job posting these brands had, and opened their engineering blogs and storefront page source one by one.

• The pixels are all in place. GA4, Google Ads, Meta, TikTok, Kakao, Naver conversion tracking, review platforms, chat widgets, email automation. Tooling is not the constraint.
• What is missing is the layer that exercises judgment on top of it. Zero dedicated data hires was common, and where the function existed it stopped at BI and dashboards.
• incrementality, geo-lift, holdout, MMM. Searching the full set of postings returned zero.

Two things from the research stayed with me.

One company publishes its operating principles on its careers page, and one of them reads, word for word, "validate leading indicators quickly." It is a good sentence. But the work that performs the validation — registering a hypothesis in advance, setting a minimum detectable effect, ruling on significance — existed nowhere in the organization. The principle was in place; the method was not.

At another, dashboard upkeep was listed as a part-time role paying around ₩10,000 an hour. At a brand turning over hundreds of billions of won, that was the price set on the seat where the numbers get read.

It all converges on one point. The data is not missing. It accumulates, and no one reads it. The tools run in separate silos channel by channel, and the layer that joins them at the customer level — the one that answers whether the money came back — is absent entirely.

05What last-click ROAS cannot see

Most brands manage against last-click ROAS, and that metric has a structural hole. When someone who was going to buy anyway clicks an ad once and buys, the revenue is booked as advertising performance. The larger the brand and the higher the repeat rate, the wider the hole gets.

Incrementality asks something different: had this campaign not run, would this revenue exist?

Answering that requires a comparison group — a holdout, a geo-split, or at minimum a hypothesis and a decision rule fixed before the spend goes out. It is a problem of experimental design, not of reporting. Which is why no amount of dashboard work ever arrives at it.

06How far public data actually goes

Visible — advertising, sales commission and promotion expense in absolute terms and as a share of revenue; three-year trends; the excess over peers; shifts in SG&A composition; inventory growth against revenue growth; channel structure, meaning whether commission channels are outpacing the company or lagging it; the marketing stack; and whether a data function exists at all.

Not visible — incrementality itself, true contribution profit by channel, repeat purchase and LTV, sell-through by store. Those require internal data: ad accounts, storefront orders, channel settlements, retailer EPOS.

A diagnosis built from public sources is therefore a hypothesis about where to look, not a measurement of incremental return. An analysis that blurs that line collapses at the first serious question.

07So what should change

That is what the outside shows. Observation alone changes nothing for the reader, so here is the ruler we actually use in this situation.

Treat advertising as a loan with a repayment date, not a budget. A budget is finished when it is spent; a loan has to come back. That narrows the question to one thing — when does this money return?

Set the threshold against the repurchase cycle, never as an absolute number of months — the right cycle differs by category.

The 2× line has a plain justification: once payback runs past two repurchase cycles, you are spending the next round of advertising before the previous customer has bought again. From there it stops being a growth question and becomes a financing one.

08Four choices that decide the answer

Payback as a concept is not new. What moves the number is the choices made while computing it, and that is where the actual work sits.

• Cut it into monthly cohorts instead of pooling. Divide total spend by total new customers and last month's repeat revenue quietly covers this month's acquisition cost. Pooled figures almost always flatter.
• Subtract variable costs only. Returns, payment fees, fulfilment, channel commission — and stop there. Load fixed costs in and you are no longer comparing channels; you are restating company P&L, and the ranking between channels disappears.
• Push returns back into the cohort of the original order. Returns arrive late, so booking them in the month they occur makes recent cohorts look permanently better than older ones.
• Keep repeat revenue out of the denominator. Include it and the metric stops measuring acquisition efficiency and starts measuring brand strength. Those are different problems, and mixing them is how you miss the moment to cut.

None of this is computable from public filings; it needs ad accounts joined to the order ledger at customer level. But without that ruler, whether to spend more or less stays a matter of taste.

09What this piece is really saying

The ad-spend argument always plays out as "spend more / spend less," and that fight has no winner — both sides are arguing without knowing incrementality. Growth doesn't excuse a collapsing margin either; the same year, one company grew +111% while raising its margin. Treat ad spend as a loan with a due date and measure payback — the line sits at twice your repurchase cycle. And if that calculation isn't possible yet, that's the thing to fix before the budget number.