Software spending benchmark

Recommendation. Plan sales and marketing at 33.2% of revenue (range 21% to 43%), R&D at 21.4% (15% to 27%) and G&A at 22.4% (19% to 28%). Each point is the median of the comparable public companies and each range is the middle half of them.

What the benchmark shows. Size decides G&A: the bigger the company, the smaller its G&A share (rank correlation −0.76). Below $5B, size does not decide S&M or R&D (rank correlations 0.01 and 0.18). What goes with S&M and R&D is growth (rank correlations 0.24 and 0.30). By thirds of companies by growth (median growth 1.5%, 14.5% and 25.8%), S&M runs 23.6%, 32.7% and 37.3% of revenue, and R&D 16.8%, 21.9% and 26.7%; G&A does not move with growth (17.0%, 15.7%, 16.5%). So the S&M and R&D lines should both follow the plan's growth target, and both points above assume growth in the low to mid teens.

Not used: the all-company medians. They are simpler, but they mix in the largest software companies, whose G&A share is about a third of a small company's: the all-company G&A median is 16.3%, against 22.4% for the smallest band, which would understate the plan's overhead by about 6 points of revenue. Nor only the most efficient companies (growth plus margin of at least 40%): under $1B there are 5 of them on GAAP figures and 10 before stock comp, too few to set a target.

Caveats. Public companies are bigger than a 300-person company: the smallest here, SOUNDTHINKING, INC., had $104M of revenue, so the plan's G&A is more likely to sit at the top of its range than the bottom. The figures are GAAP, so every line includes stock-based compensation, a median 11.8% of revenue under $1B; a private company that pays less in stock will show lower expense for the same people. The three medians come from different companies and do not add up to one real company: under $1B the median gross margin is 72.4% and the median GAAP operating margin -5.0%.

SEC's XBRL data is large and inconsistent: 3 different tags carry the top line, some filers split S&M from G&A and some do not, fiscal years end in every month, and each year is reported again, sometimes restated, in later 10-Ks. The raw values go into SQLite unchanged, and every rule below is a view on top of them.

CREATE VIEW v_fy AS
SELECT cik, MAX("end") AS fy_end
FROM v_revenue_annual
WHERE "end" BETWEEN '2025-01-01' AND '2026-01-31'
GROUP BY cik;

CREATE VIEW v_annual_all AS
SELECT f.*,
       CAST(julianday(f."end") - julianday(f.start) + 1 AS INTEGER) AS days
FROM fact f
WHERE f.unit = 'USD'                       -- why: ratios need dollar values; share counts and per-share values are other units
  AND f.form IN ('10-K', '10-K/A')         -- why: a full year is reported in the 10-K; 10-Q values are quarters or year-to-date
  AND f.start IS NOT NULL                  -- why: income-statement lines are durations; a value with no start is a balance
  AND julianday(f."end") - julianday(f.start) + 1 BETWEEN 350 AND 380;
                                           -- why: a 12-month period, allowing 52/53-week years; drops quarters a 10-K also tags

Of the 151 software filers, 27 are not in the benchmark: 17 because they file a 20-F or 40-F as foreign private issuers, not a 10-K, and 10 because they stopped reporting before filing a 10-K for a year ending January 2025 to January 2026 (each has a Form 15 deregistration on file). Every name and reason is in dropouts.csv. The 42 filers in SIC 7371 and 7374 are listed but not added: those codes mix software with IT services and payment processing, and picking the software ones by hand would put a judgment call into the sample.

A check found that two filers' SG&A tag already contains R&D, Paycom's and a smaller company's (both are in the check's output): gross profit less SG&A alone equals their operating income. Their combined SG&A share would have counted R&D twice, so a rule now leaves it out. The third company that check lists, SOUNDHOUND AI, INC., is real: a large gain on earn-out liabilities sits inside its operating expenses.

Amplitude tags no operating income at all, so it first fell out of every margin figure; the rule above now computes it from gross profit and total operating expenses. Paylocity's two revenue tags differ by 7.7%: the larger includes interest on client funds and is its reported total. A second derivation in Python, from the raw JSON and without SQL, reproduces every company's ratios and every quartile on this page exactly, and confirms revenue, S&M, G&A and R&D each sit in SEC's XBRL data under the accession number given.

Public companies are bigger than the company the targets are for. Only companies that filed a 10-K for a year ending January 2025 to January 2026 are in: the 10 that deregistered before filing one are not, while 7 that deregistered after their last 10-K are in, on that year (for example Couchbase, on its year to January 2025). One year is a snapshot; growth and margins in 2025 reflect that year's market. companyfacts carries no breakdown by function, so stock comp cannot be taken out of S&M, R&D and G&A one by one, only out of operating income as a whole. SIC codes are chosen by the filer and rarely updated, so the group includes a few companies, such as 3D Systems, whose main business is not software.

Next: run the same SQL on the prior two years to see whether the patterns hold, and add free cash flow margin, the more common partner to growth in the Rule of 40.

Source: SEC EDGAR, data.sec.gov XBRL frames, companyfacts and submissions APIs, fetched 2026-10-24 (UTC). Values are as filed in each company's 10-K. Independent study of public filings; not investment advice.