Raise it to stand still.
E-commerce advertising is not a growth cost 38,990 audited firms → 71 e-commerce companies · 101 company-years · audit reports

01What was measured

Starting from the full set of 38,990 externally audited Korean companies, 71 carry an e-commerce industry code. Financials parsed for 67; removing the six that sell other people's goods and book only commission as revenue (§02) leaves 58 as the population for ad share. An audit report prints the current year beside the prior year, so for each company you can pull "how much the ad spend moved and how much sales moved that year." Open the prior year's report too and you get two years of change — 101 company-year observations. Cosmetics (121 companies) and food (90) were counted the same way (195 and 166). Twenty-five entities that sit under those industry codes but are not in those businesses — pharma subsidiaries, paint makers — were removed.

Advertising here means the advertising expense line only. Companies that push channel costs into sales promotion or sales commission are dealt with in §05. The two ratios in this piece are defined as ad share = advertising expense ÷ revenue × 100 and sales growth = (this year's revenue ÷ last year's − 1) × 100. Both are computed per company and reported as medians, not means: this population holds companies from 0.1 billion won of revenue to 400 billion, so a mean would be set by a handful of large ones.

02I had it wrong first — 5.43%

The first figure I got was a median ad share of 5.43%, more than a point above the 4.3% of cosmetics. I nearly opened with "this industry spends more on advertising." Opening a few of the top companies showed the advertising was not large — the revenue was written small.

If you sell someone else's goods you are an agent in accounting terms, so revenue is the commission, not the transaction value. The K Connect's income statement does not start with "revenue" at all but with operating income, 97.1% of it commission. The note reads: "where the entity transacts as an agent on behalf of the principal, only the commission amount is recognised as revenue." At Abetter, product revenue flipped from 47.3 billion won to 9.9 billion in a single year while service revenue went from 0.1 billion to 19.3 billion. Nothing about what it sells changed; it moved from gross to net. The same 1 billion won of advertising is 5% at a gross company and 30% at a net one. When the denominator differs, a ratio is not a comparison.

Scanning all 71 audit reports for the signals — "operating income", "commission revenue", "agent" — flagged 18 for a human to read. Two agents, three that book gross and net side by side, one whose real business is renting property. Take those six out and the median is 4.7%. Even that is not stated flatly: the value just below the median is in the 3.9s and the one just above is in the 5.4s, so one company entering or leaving moves it by 0.7 points. As a range: first quartile 1.7%, median 4.7%, third quartile 12.0%. Cosmetics: 1.6%, 4.3% and 11.7%. The advertising structure is effectively the same as cosmetics. "They spend unusually heavily" does not hold.

03The industry is shrinking

If ad share is the same, the next question is what that money buys. First, the stage.

Seven in ten shrank. The prior year (FY2024) was already at a median of −2.9% with 56% shrinking, and it deepened from there (the two-year comparison uses companies with advertising booked in both years, so the sample differs slightly from the table). Cosmetics was positive in both years (+4.8 → +5.9) and so was food (+0.5 → +3.8). This is not one bad year but two consecutive years of decline. Splitting by size gives the same picture — companies under 30 billion won in revenue sit in the −10s, and only the handful above 100 billion are barely positive. The "e-commerce is in the shrinking quadrant — high closure rate, almost no openings" finding from the earlier neighbourhood-commerce piece holds at audited-company scale too.

04You have to raise it to stand still

Each company's change in ad spend was split into four bands, with the median sales growth for that year. Two years pooled.

The three lines are shaped differently. Cosmetics grows when it spends — median +11.9% among those that raised, +0.2% among those that cut. Food moves less: +3.7% against −2.4%, a gap of 6.1 points. Brand and shelf carry sales more than advertising does. E-commerce runs −1.1% among those that raised and −10.9% among those that cut.

There is one thing to read here. The gap between raising and cutting is 9.7 points for e-commerce and 11.7 points for cosmetics. The 2.0-point difference has a bootstrap 95% interval of [−10.8, +11.3], squarely across zero — the sensitivity to advertising cannot be told apart. Spend more and e-commerce shrinks less by as much as cosmetics grows more. What differs is the height. The cosmetics line starts above zero and climbs to +12; the e-commerce line starts at −11 and reaches zero. A slope that cannot be told apart, sitting eleven points lower on the four-band average — and even among companies that raised spend alone, 13.1 points, 95% interval [+6.2, +19.6], well clear of zero.

Put plainly: this is an industry where the sales that survive without advertising melt by around ten percent a year, and the first slice of the ad budget goes to covering that. In cosmetics the same money lands on top of sales held up by shelf, repeat purchase and search. E-commerce sells only in the moment it buys traffic, and when it stops buying, the shelf space disappears. That is why the spend behaves less like an investment and more like rent. The name the data gives it is a maintenance cost.

The two-by-two is simpler still. "Spent more and shrank" is 27% in e-commerce, 14% in cosmetics, 22% in food. E-commerce is the only one of the three where "spent more and shrank" outnumbers "spent more and grew."

It is not identical every year. In FY2024 the ten e-commerce companies that raised spend by more than 20% grew +17.1%. In FY2025 the same band is negative. So this is not written as "raising it does nothing" — pooling the two years, it goes as far as "you have to raise it more than 20% to stand still." The cutting side lost ground in both years alike.

05What not to do with these numbers

This is not causation. Sales may have fallen first and the advertising cut followed — reverse causation. Three years of data give one look at it. The 26 companies that cut spend in the current year had a prior-year median growth of −2.3%; the 17 that raised it, −2.0%. The two groups were the same a year earlier and split into −11.4% and −2.1% in the current year. "They were already falling, so they cut" does not explain it. It is still not causation. What this piece claims is that there are industries that held their sales while cutting advertising, and e-commerce was not one of them. The contrast between industries is the finding, not a calculation of what one company loses by cutting.

The July 2024 settlement collapse. Sellers caught in the TMON and WeMakePrice failure are certainly in here. But audit report notes do not name counterparties — across 71 companies both names appear in one, and none attributes an impairment to them. It cannot be isolated from disclosures. What can be said is what is above: the decline began the year before. If the collapse mattered, it accelerated something already underway rather than causing it.

Costs move between accounts. Only the advertising expense line was counted, and some companies route what they pay channels through sales promotion or sales commission. A sales commission line appears in 38% of the e-commerce companies and 50% of cosmetics. Summing all three gives medians of 8.5% and 9.8% — the ordering flips from advertising alone (4.7% vs 4.3%). Counted either way, "e-commerce spends unusually much" does not hold — §02 stands. Individual companies can move.

The n differs. Ad share is stated on 58 companies; the change figures rest on 101 company-years. A company entering the audit regime this year has no prior report and drops out of the change table — and a newly entered company may well be in a growth phase, so that cell is empty. Only audited companies are here, so direct-to-consumer shops below a few billion won in revenue are absent by construction.

They are not called "own-brand" companies. The 58 book revenue gross; that does not make them brand owners. A retailer buying on its own account is also gross. This piece does not separate the two, and does not tell a brand story it has not separated.

The middle band is the worst place to be. The ten companies with an ad share between 10% and 20% show median growth of −13.9% and an operating margin of 0.8%. The seven above 20% grow +12.9%, but one in four is loss-making (29%) and the median margin is 1.9%. Buying growth costs a fifth of revenue, and after that there is nothing left. A company in between covers neither the maintenance share nor any increment. Ten companies is a place to check, not a finding.

06What changes because of it

Treat advertising as a maintenance cost rather than a growth cost and three things change.

First, aggregate ROAS stops meaning anything. Sales that would have happened without the advertising are mixed into the denominator. The question is not "revenue per won of ad spend" but "how many orders would not have happened without that ad". The answer is not in the total; it is at the order level — hold out a group, switch a region off, split search into branded and non-branded terms. Either way, every order needs a label saying whether it would have arrived anyway.

Second, the budgeting unit changes. If it is maintenance, you do not plan "how much this year" but "which orders on which channel are maintenance and which are increment." That is why the 10–20% band is the worst place to sit — put the budget near that level without knowing the size of the maintenance share and you cover neither it nor any increment. Measure the maintenance share first and either build above it or drop well below it and spend elsewhere. Do not sit in between.

Third, match denominators before comparing. The ad share of a company booking net is inflated. Convert to gross merchandise value or leave it out. In this industry what counts as revenue differs by company, and a ratio computed before settling that is not a comparison.

Turned into actions. The thresholds are read straight off the tables above; redraw them from your own data.

If you are setting an ad budget, these are the four questions.

07What this piece is for

The same advertising bought growth in cosmetics and standing still in e-commerce. The difference is not in the total. It is in whether each individual order would have arrived without the ad. What separates them is not the ad budget but the definition of the data — just as repeat-purchase rate cannot be computed until you decide what counts as one purchase, ad spend ends as a single total until you decide which orders are incremental, and the total, as shown above, separates nothing.

Most of our work looks like this. Far more often than the data being missing, the data that exists is not in a state that can answer. Setting that definition comes before the analysis, and it is usually the part that is actually hard.

Getting data into a state where it can answer is what we do.

(Source: audit reports of 38,990 externally audited Korean companies; 71 with an e-commerce industry code, FY2025, separate-entity basis for the most part. Financials parsed for 67; six agent, mixed or non-operating entities removed, leaving 58 for ad share. Change figures pool the current report's year-on-year with the prior report's, giving 101 company-years (cosmetics 195, food 166). Advertising is the advertising expense line. Comparison groups — 121 cosmetics and 90 food companies — come from the same master by the same method (25 flagged entities such as pharma subsidiaries and paint makers removed). No internal data used.)