Ad spend rose 84%.
Spending on people rose 9% 325 consumer companies in five industries · SG&A schedules · audit reports

01What was measured

Every audit report has a schedule of selling, general and administrative expenses (SG&A): the money spent on selling and on running the company, line by line. Salaries, severance and benefits sit near the top; advertising, sales promotion and sales commissions sit somewhere below. The money a company spent on people and the money it sent outside in a year are in the same table, with a single total at the bottom.

This piece uses three terms. The people share is the part of SG&A that is salaries, severance and benefits. The outside share is the part that is advertising, sales promotion and sales commissions. When the outside share is larger than the people share, we call it flipped.

We read this table in 449 audit reports across cosmetics, food, beverages, health functional foods and e-commerce. We kept only tables whose lines add up to the table's own total, and whose total also matches the SG&A line on the income statement. 417 of 449 passed the first check, 413 the second. Keeping companies with sales of 3 billion won or more (403) that actually report a staff-cost line leaves 325: 299 on a separate basis and 26 consolidated. The median company has sales of 29.9 billion won and SG&A of 8.55 billion.

Putting severance and benefits into the people share was deliberate. Using salaries alone makes the people share smaller and flipped companies look more common. We chose the definition that makes the conclusion smaller.

02Inside the same box, the two push each other out

Of every 100 won of SG&A, the median people share is 28.4 won and the median outside share 14.1 won. Together they come to 52.3 won; the other 47.7 won is rent, depreciation, freight and taxes, money that cannot be moved either way. In other words, half of SG&A is already fixed, and people and outside split the other half. The larger the people share, the smaller the outside share, and the reverse.

The outside share was larger than the people share at 115 of 325 companies, 35.4%. At these companies, the people actually running marketing and sales are not inside the company but outside contractors. The record that work leaves behind (which ad brought which order, which coupon reached which customer) stays in the contractor's systems. The company gets a month-end report.

By industry it looks like this.

This table surprised me too. Cosmetics, food and health functional foods, the industries known for heavy advertising, still have a larger people share. So we split the companies again, by how much they advertise. At 46% of these companies, advertising is under 2% of sales; at only 19% is it above 10%. Heavy advertisers are a minority on this list.

Take only the 82 companies spending around 14% of sales on advertising, and the outside share is 39.2% against a people share of 17.2%, with 82% flipped. That is the picture we expected, and it is there. But those companies are one in four. The other 243 spend 1.4% of sales on advertising and sit at 32.3 people vs 8.2 outside. The brands you notice for their advertising and the companies that file audited accounts are different populations: the list holds more ingredient makers, contract manufacturers and distributors than brands, and in health functional foods, home-shopping commissions are often taken straight out of sales and never pass through SG&A. The industry medians are pulled down by that majority.

03The bigger the company, the more outside wins

As companies grow, the people share falls from 32.3 to 21.3 and the outside share rises to 22.0 above 100 billion won. Above 100 billion won, the two flip. With only 47 companies, we cannot pin down by how many points flipping increases; we state the direction only.

It is not the industry mix either. Counting again within the companies above 100 billion won, cosmetics, beverages and e-commerce are each more often flipped than their industry overall. Only food goes the other way. The flip among large companies is not just a matter of having more e-commerce firms.

04Ad spend rose 84%; spending on people rose 9%

The SG&A schedule shows this year and last year side by side. Two columns of the same table, so the basis is automatically the same. Two years lined up for 302 companies, and in 129 of them advertising rose by more than 20%.

In years of big ad increases, 71.3% of companies also spent more on people. They did not stand still; the problem is size. Only 36.4% raised people spending as much as advertising (by more than 20%). People spending rose 8.8%, barely keeping up with sales (7.2%), while advertising moved more than ten times as much.

Turn it around: at the 71 companies that cut advertising by more than 20%, people spending moved +0.3%. It did not fall. Advertising swings hard up and down; people move slowly, in one direction. Advertising can be bought this year and cut next year; people hired this year are still there next year. So urgent execution is bought from outside, and reading what it produces keeps getting postponed.

Where the extra money went is in the same table. Among the 129 companies that raised both, the extra spent outside was 1.38 times the extra spent on people: 58 of every 100 new won went outside. 52 raised outside only, 61 raised people only, and 60 cut both.

Flipped and non-flipped companies look different. The 115 flipped companies spend 9.25% of sales on advertising; the 210 others spend 1.31%. Operating margins are 3.59% and 5.20%, which looks like a gap, but at this sample size it cannot be told apart from chance. This piece does not say flipping cuts profit. It says only that the two groups are shaped differently.

05One line called fees paid, which the filings do not split

In section 01 we left fees paid out of the outside share. Put them in and the numbers change a lot.

One definition moves the median company across the 1× line. Fees paid appear at 321 of the 325 companies, make up 8.9% of SG&A, and are larger than advertising at 56.6% of them. The problem is that the filings do not say what is inside that line. It could be platform payment fees or agency fees; it could be tax, legal, logistics or security services. Tolls paid to other people's channels and the company's running costs share one name.

So whether these companies sit at 0.46× or 1.04× cannot be settled from public data. Not because the sample is small, but because the format does not split it. This piece uses only the narrow figure in the body. Inside a company it is different: open the fees-paid ledger by vendor once, and in a day you can tell which lines are channels and which are overhead. Not visible from outside and not looked at from inside are different problems, but they usually go together.

06What it took to get these numbers: "salary" appeared in four places

That is the result; the time went here. Search an audit report for "salary" and it turns up in four places: the SG&A schedule, the key-management compensation table, the registered-director compensation table, and the related-party transactions table. The same word each time, each with this year's and last year's amounts. The label alone cannot tell them apart.

The first extractor took the first place "salary" appeared. At one cosmetics company that gave 1.19 billion won: the compensation of a handful of executives. The salary line in the same document's SG&A schedule is 20.18 billion won, a factor of 17. At a beverage company, a bracketed value from the related-party table was picked up, and a company with 297.5 billion won in sales was read as paying 95 million won in salaries.

Fixing it took four more mistakes. Reading widely after the table title swallowed neighbouring notes into one table; reading a little past the total line dragged in the first line of the next table, and 182 perfectly good tables were thrown out for not adding up. Finding tables by title caught a different table with a similar name; finding them by the units marker dropped every company that lays out SG&A line by line in the income statement itself, which was close to half the sample.

The current method reads tables as tables. Every candidate block is split into line, this year and last year, and only blocks whose lines add up to their own total count as a table. That total is then checked once more against the SG&A line on the income statement. Pass both, and it is that company's SG&A schedule. Three companies picked at random and checked by eye against the original came out 0.000%, 0.000% and 0.096% off.

Partway through, we realised the work had the same shape as the argument. "How much is salary?" was not a question the data could not answer. The data was all there, and without someone deciding which box a number belongs to, the answer is off by a factor of 17. Most companies' internal ad data is in the same state.

07So what do you look at

This ratio comes from one audit report. For your own company, the internal income statement gives it in 5 minutes.

08What this piece is saying

Post 15 said advertising is a maintenance cost, not a growth cost; post 17 said advertising is not a valve that controls profit. Both set ad spend beside sales and profit. This piece sets ad spend beside salaries, because they are in the same table.

What companies buy is execution. Advertising, promotion and commissions run next week once you pay. Reading what that execution recorded and turning it into the next decision does not happen next week because you paid. That is people, and people compete with execution for the same box in SG&A. So "there's no one to look at the data" is not a confession of laziness; it is the result of a cost structure. And the bigger the company, the narrower that space gets.

There is one thing that can be done here. Not hiring more people, but bringing back inside the records that outsourced execution leaves behind. Moving ad accounts into the company's name, writing raw-data delivery into contracts, opening the fees-paid ledger by vendor: none of it needs headcount. It is the same kind of work as sorting out the four "salaries" in section 06.

Turning data that cannot be used into data that can: that is what we do.

Appendix — the doubts we tried, and numbers for checking

Did we count people too broadly? Drop severance and benefits and use salary alone, and the people share falls to 23.2% while flipped rises to 39.1%. Narrowing makes the conclusion bigger; the broad definition is the conservative one. Conversely, counting fees paid as outside gives 51.1% flipped (section 05).

Did mixing separate and consolidated accounts matter? The 299 separate-basis companies alone give a people share of 28.7%, outside 13.4%, flipped 34.8%. Same as the whole.

Do companies with no advertising at all pull the numbers? Removing the 24 with zero advertising leaves 301, with outside ÷ people at 0.52× and flipped at 38.2%. It goes up, not down.

Is the flip among large companies an industry effect? Within companies above 100 billion won, cosmetics 53% (17 companies), beverages 50% (6) and e-commerce 50% (6) are each above their industry overall; only food is lower at 23% (13). The counts are small, so we say only that industry alone does not explain it.

We first wrote a different sentence and dropped it. The first hypothesis was "as sales grow, outside spend follows sales but people spend does not." In a 65-company sample, a 1% rise in sales went with +0.50% for people and +1.63% for outside, a clean split. After fixing the table reader and widening to 325, it was +0.33% vs +0.28%, and the gap was gone. So section 04 uses medians, not that ratio.

Remaining limits. We cannot count heads; this piece counts money. The same 100 million won could be one person or three, and the people who read data share a box with warehouse staff. SG&A salaries cover sales and administrative staff only, so manufacturing staff are left out. With two years only, three-year paths are invisible. Of 449, 32 had no findable SG&A schedule, and of 403, 78 had the table but no staff-cost line. One in four does not answer this question.

For checking. The wobble ranges of the medians (4,000 resamples, 95%) are people share [26.5, 30.1], outside share [11.4, 17.6], outside ÷ people 0.46× [0.37, 0.56]. The range for 35.4% flipped (Wilson 95%) is [30.4, 40.7]. Flipped by revenue: 3–10 billion won 27% [13, 48], 10–30 billion 33% [25, 41], 100 billion+ 47% [33, 61]; they overlap. People spending +8.8% in years ads rose has a range of [3.4, 16.0]; the 1.38× among companies that raised both has [1.00, 2.33]. Operating margins of 3.59% [2.2, 4.4] for flipped and 5.20% [4.1, 6.1] for the rest overlap at the ends. Median revenue is 33.9 billion won for flipped companies and 25.7 billion for the rest; staff costs are 6.97% and 8.40% of sales. The rank correlation between people share and outside share is −0.569; with revenue, −0.201 for people share and +0.170 for outside share. The rank correlation between outside ÷ people and operating margin is −0.057, essentially 0. The 0.58× difference between leaving fees paid out and putting them in is the resolution public data has on this question.

(Note: source — audit reports of externally audited Korean companies on DART, 5 industries (cosmetics, food, beverages, health functional foods, e-commerce); SG&A schedules read from 449 reports. Lines summing to within 3% of the table total (417), total within 2% of income-statement SG&A (413), sales of 3 billion won or more (403), staff-cost line present (325). 299 separate, 26 consolidated, mostly FY2025. People share = (salary + severance + benefits) ÷ SG&A; outside share = (advertising + sales promotion + sales commission) ÷ SG&A. Two-year comparison uses this year's and last year's columns of the same table (302). No company names. Zero internal data.)