What is a typical advertising spend as a percentage of revenue?

Across three unrelated US industries measured from SEC filings for calendar year 2024, the median company spent between 2.01% and 3.30% of revenue on advertising. Services — Prepackaged Software (137 filers) had a median of 2.01%, Pharmaceutical Preparations (61 filers) 3.22%, and Services — Business Services NEC (40 filers) 3.30%. The medians are nearly identical despite the industries having almost nothing in common.

Is there a reliable advertising benchmark for my industry?

No, on this measurement. The spread inside each category was far larger than the difference between categories. The within-category coefficient of variation was 2.21 for prepackaged software, 6.10 for pharmaceuticals and 1.51 for business services; after removing filers whose revenue was effectively zero, the median CV was 1.51. A coefficient of variation above 1.0 means the standard deviation exceeds the mean, so the category average does not predict any individual company in it.

How much does Take-Two Interactive spend on advertising?

Take-Two reported $1,287,100,000 of advertising expense on $6,656,400,000 of revenue for the fiscal year ended March 31, 2026, an intensity of 19.34%. That is roughly six times the cross-industry median of 2% to 3.3%. Advertising stepped up 4.08 times in the fiscal year to March 2023 — from $297.3m to $1,212.5m — and has been flat to falling since. The Zynga acquisition closed in May 2022, inside that year.

Which companies spend the most on advertising relative to revenue?

Among companies CULT+MATH measured directly for calendar 2024 and the fiscal years ending in 2025 and 2026: Take-Two Interactive at 22.26% and then 19.34%, Maplebear (Instacart) at 13.53%, DoorDash at 12.12%, Electronic Arts at 4.70%, Uber at 4.32% and Lyft at 2.55%. The pattern is not industry. DoorDash and Instacart are four times typical because paid customer acquisition is their business model; Uber, which does the same deliveries alongside a rides business that does not buy its customers, sits near the median.

Where does advertising expense data come from in SEC filings?

The XBRL tag AdvertisingExpense, available through the SEC's free frames and companyconcept endpoints at data.sec.gov. For calendar year 2024, 1,633 US filers tagged it. Disclosure of this concept is voluntary, so the set is large but is not a census, and companies that disclose may differ systematically from those that do not.

How do you check an SEC XBRL figure is correct?

Pull the same company's own companyconcept history and reject any value whose period is tagged more than once with materially disagreeing numbers. In this study 12 of 250 rows were rejected: 11 for material conflicts and one for a name mismatch. One filer had the same period tagged as 5,144,579 and 5,144,579,000 and sat fifth in the raw ranking before validation removed it.

Does a company's industry predict how much it spends on advertising?

On this measurement, barely. The median is roughly the same across unrelated industries and the spread within each one is enormous, which makes the category simultaneously useless as a predictor and unnecessary as a benchmark. The variable that appears to matter is the business model — specifically whether customers are acquired through paid media — rather than the industry code.

Standing study · Marketing · Re-run on each filing season

Your category's advertising benchmark is noise. The cross-industry median is not.

We read every US public filer that disclosed advertising expense for calendar 2024 — 1,633 of them — and computed how much the number varies inside an industry against how much it varies between industries. The spread inside a category is far larger than the gap between categories, which means the peer benchmark on your slide does not predict your company. Every row, every rejection and every prior is published.

The typical company spends about 2% to 3.3% of revenue on advertising, and that is true whatever industry it is in.

Enterprise software, drug development and a miscellaneous business-services bucket returned medians of 2.01%, 3.22% and 3.30%. Three industries with almost nothing in common, landing within 1.3 percentage points of each other.

1,633 filers · CY2024 · SEC XBRL at data.sec.gov · $0 metered · last measured 2026-10-25 · next re-run when CY2025 frames populate

The registered question, written before any retrieval

Within a given industry category, does disclosed advertising intensity — advertising expense divided by revenue — cluster tightly enough that a peer band means anything?

Candidates and priors, recorded before any data was pulled
#	Candidate	Measured as	Prior	Outcome
C1	Intensity clusters tightly. Category is the dominant variable.	CV < 0.5	25%	Dead. Every category above 1.0
C2	Intensity spreads widely. Category tells you almost nothing.	CV ≥ 1.0	40%	This is the result
C3	A band exists but is broad.	0.5 ≤ CV < 1.0	25%	No
C4	Too few filers tag both concepts.	n < 15 per category	10%	No. 238 rows survived validation

C2 was the outcome I least wanted, and it was registered at the top of its band anyway, because it kills the saleable version of this work — the per-category benchmark chart — and replaces it with something harder to sell and more defensible.

The measurement

Within-category spread · CY2024 · categories chosen by filer count and named before any CV was computed
SIC	Category	n	Median	Mean	Std dev	CV
7372	Services — Prepackaged Software	137	2.01%	4.13%	9.13%	2.21
2834	Pharmaceutical Preparations	61	3.22%	28.22%	172.17%	6.10
7389	Services — Business Services, NEC	40	3.30%	7.64%	11.55%	1.51

A coefficient of variation above 1.0 means the standard deviation exceeds the mean. All three categories are above it. The median within-category CV is 2.21, against a registered threshold of 0.5 for the tight-cluster candidate.

Join rate: 1,404 of 1,633 filers, or 86.0%. The 229 that tagged advertising but no matching revenue concept are counted here rather than dropped silently.

Positive control, run before anything else. Procter & Gamble present at $9.6bn advertising on $84.0bn revenue. If a known large advertiser had not survived the join and the validation, every zero in the pass would have been meaningless.

The number that is worth more than the verdict

2.01%

median, enterprise software

3.22%

median, drug development

3.30%

median, business services

Three industries with nothing in common, and the typical company in each spends between 2% and 3.3% of revenue on advertising. Meanwhile the spread inside each category is enormous.

So the category is simultaneously useless and unnecessary: useless because it does not predict any individual company, unnecessary because the typical value is roughly the same everywhere.

The question a marketing leader should be asking is not what their category spends. The answer is about 2 to 3%, like everyone else. It is what does my business model spend, and have I chosen it deliberately.

What that looks like on named companies

Advertising intensity, read from each filer's own XBRL tags · multiple of the 2–3.3% cross-industry median
Company	Period	Intensity	Multiple
Take-Two Interactive	FY to Mar 2025	22.26%	~7x
Take-Two Interactive	FY to Mar 2026	19.34%	~6x
Maplebear (Instacart)	CY2024	13.53%	~4x
DoorDash	CY2024	12.12%	~4x
Electronic Arts	FY to Mar 2026	4.70%	~1.5x
Uber	CY2024	4.32%	~1.5x
Lyft	CY2024	2.55%	~1x

None of this is a category effect. DoorDash and Instacart sit at four times typical because paid customer acquisition is their business model. Uber, which runs the same deliveries alongside a rides business that does not buy its customers, sits near the median. Take-Two's step change came in the fiscal year the Zynga acquisition closed.

The Zynga explanation is a hypothesis. Advertising stepped 4.08 times in the fiscal year to March 2023 — from $297.3m to $1,212.5m — and the acquisition closed in May 2022, inside that year. The timing and the mechanism fit. The filing has not been read, and that stays marked wherever this appears.

The sensitivity test, run because one number demanded it

Pharma's CV of 6.10 with a 172% standard deviation is not a finding, it is a question. One filer answered it.

ONE WORLD PRODUCTS, INC. — advertising intensity 1,349% on revenue of $4,863.

Not $4.8 million. Four thousand, eight hundred and sixty-three dollars. The ratio is arithmetically correct and conceptually meaningless.

Dropping every row with intensity above 100% — the cases where the denominator is effectively zero
SIC	n after	Median	CV before	CV after
7372	137	2.01%	2.21	2.21
2834	60	3.22%	6.10	1.49
7389	40	3.30%	1.51	1.51

Median CV after the cut: 1.51. Still the wide-spread verdict. The result does not depend on how the outliers are treated, which is the only reason it is safe to state. The pharma CV does — it moved by a factor of four on one company — and anyone quoting 6.10 without that sentence is quoting an artifact.

The 100% cut is crude and we are saying so rather than dressing it up. Zapata Quantum sits at 97.6% on $3.9m of revenue and survives the cut with the same problem as the company that did not. A revenue floor would be the better instrument. It was not in the registered design, so it is not applied here.

What we threw away, and why

Every row entering the three categories was checked against the filer's own concept history. 12 of 250 were rejected — 11 for material conflicts, one for a name mismatch. All twelve are published because a validation step nobody can see is a claim rather than a check.

Rejected rows · the six that illustrate the categories of failure
Filer	What the conflict was	Why it matters
Universe Pharmaceuticals	same period tagged 5,144,579 and 5,144,579,000	Order of magnitude. Sat fifth in the raw ranking.
Eli Lilly	1,400,000,000 against 1,440,000,000, 2.9% apart	Above the 1% threshold and it cost a major filer. The threshold is a tuning question, stated rather than silently changed.
Backblaze	3.0m against 4.2m	A genuine disagreement, not rounding.
Standard Biotools	2.6m against 5.9m	A genuine disagreement, not rounding.
eGain	180k against 2.1m	A genuine disagreement, not rounding.
Reshape Lifesciences	frame says RESHAPE LIFESCIENCES, concept says VYOME HOLDINGS	A legitimate rename, rejected by the name check. The one false positive of the twelve.

Eli Lilly is the uncomfortable one. 1,400,000,000 against 1,440,000,000 — 2.9% apart, above the 1% threshold, and it cost a major filer. At 1% it rejects Lilly; at 5% it keeps Lilly and still catches every order-of-magnitude error. The registered design did not specify a number, so the one used is ours and should be set deliberately before the next pass.

What this study cannot tell you

Disclosure of this concept is voluntary. 1,633 filers is a lot and it is not a census. Companies that disclose advertising may differ systematically from those that do not, and this design cannot remove that selection effect.
One year. CY2024 only, plus named companies read at their own fiscal year ends. No trend claim.
SIC is a crude grouping. It is what the data carries. A category that holds a direct-to-consumer brand buying every customer and a wholesaler buying none is not a peer group, and that may be most of what the wide spread is measuring.
Nothing here reads a filing's prose. www.sec.gov is unreachable from the workspace that ran this, so every figure is a tagged fact and the Zynga explanation stays a hypothesis.
Advertising expense is an input. No row measures what it bought.

How this page changes

A standing study, not an article. Corrections are published in place and dated, never applied quietly.

Re-run schedule
Measurement	Cadence
Category CVs and medians	When CY2025 frames populate
Named-company intensities	On each company's 10-K
Validation rejections	Every re-run, republished in full

Change log
Date	What changed
2026-10-25	Published from the pass of 2026-10-24. Two treatment choices are ours rather than registered — the 100% cut and the 1% materiality threshold — and both say so on the page.

This study re-runs when the next year of filings lands.

Sound Decisions carries each re-run and each correction, with the registered question written before the data is pulled.

Subscribe to Sound Decisions