The hidden factory is an office
ResearchThe claim floats around this industry that contract manufacturers carry an admin layer of around twenty percent. We went to the filings to test it before publishing anything, and the honest answer is stranger than the folklore: as a share of revenue the number is 2–6% — and as a share of what actually matters, it is enormous. On thin gross margins, selling and administration consume 40–50 cents of every gross-profit dollar at the big incumbents. A field note on where the twenty percent really lives, the forty-year-old paper that predicted all of it, and what is genuinely deletable.
2026-07-28 · Field notes · 6 min read · By Sondre Hegerland KristiansenSmall on revenue, half the profit pool
Eight contract manufacturers report selling, general and administrative expense as its own line, and their latest audited statements agree with each other: the median is 3.8% of revenue. Nobody's filings support "20% of revenue," and repeating it invites a one-search correction. But the revenue framing buries the story. Electronics manufacturing services run gross margins of 9–12%, so a thin-looking admin line devours the profit pool: 59 cents of every gross-profit dollar at Benchmark, 49 at Plexus, 42 at Jabil, 41 at Flex and Sanmina. The two leanest — Celestica at 2.1% of revenue and Fabrinet at 2.6% — also post among the industry's best operating margins, which is the causal arrow worth staring at. Two honesty notes that survive into every page this research touched: Hon Hai's quarter-trillion-dollar scale pushes its whole operating-expense line under three percent, and the Nordic EMS names — Kitron, Scanfil, GPV — publish by-nature income statements with no SG&A line at all, so no number is cited for them anywhere on this site. The full sourced table now lives on the economics page.
The twenty percent lives in the quote
So where does a widely repeated "around 20%" come from? From the place customers actually meet the number: the quote. A contract manufacturer writing about its own cost model describes a 25% margin as typical for low-to-mid volume — which is a 33% markup on direct cost. Materials carry their own stacked markups of 10–15% before assembly is priced at all. And the most clarifying figure in the whole research pass: in a published should-cost teardown of an EMS program, direct labor — the humans physically assembling the product — was just under two percent of program revenue, while materials ran 75–85%. Sit with that spread. Nearly everything a customer pays beyond parts is not assembly; it is overhead, indirect labor, and margin — quoting, program management, procurement transactions, expediting, order administration. The twenty percent is real. It is just a statement about the quote's loading, not the income statement — and either way, it is the part of the price that pays for humans doing transactions.
Miller and Vollmann called it in 1985
None of this is a new discovery. Harvard Business Review published "The Hidden Factory" forty years ago, and its argument was precisely this: in electronics especially, overhead is driven not by production volume but by transactions — ordering, expediting, scheduling, balancing supply and demand, quality documentation, engineering changes. There is a second, invisible factory inside every plant that manufactures paperwork, and it grows with the number of transactions, not the number of units. What has changed since 1985 is only one thing, and it is the thing this whole model is built on: transactions are now what software does best. Intake validates instead of a quoting team reading emails. The quote is computed instead of negotiated. The unit record reports instead of an account manager compiling status decks. What software cannot delete is also worth naming plainly: the quality system, compliance and audits, facilities, and the machines and people who fix them. Celestica and Fabrinet mark roughly where that floor sits — and their margins show the floor is a fine place to live.
The wedge is structural, not a discount
The honest version of the pitch is stronger than the folklore it replaces. "We are 20% cheaper" would be indefensible; what the filings actually support is sharper: the admin layer consumes 40–50% of every gross-profit dollar at the incumbents, the quote loading above direct cost runs 25–33% at the volumes where new products live, and both are made of transactions — the one input whose cost software collapses. That attacks the incumbent profit pool at its center, and whatever the OS deletes can move to price, to margin, or to both. One more honesty clause belongs on the record: eliminating all of SG&A frees only 2–6 points of revenue. The larger prize hides inside cost of goods sold, where the indirect-labor half of the hidden factory sits without a line item to its name — and no filing measures it. That is a claim the first hub will have to prove with its own numbers, which is exactly what the register is for.
Where this came from
Company figures from SEC XBRL income statements and the NOTE year-end report; each row is linked from the benchmark table on the economics page. Key references:
Flex — SEC filings · Jabil — SEC filings · Celestica — SEC filings · Fabrinet — SEC filings · NOTE AB — year-end report 2025 · Optimum Design — the cost-plus quote model · Zetter — what the EMS quote is not telling you · VentureOutsource — should-cost: direct labor under 2% · Gemba Academy — direct-to-indirect labor ratios · Miller & Vollmann — The Hidden Factory (HBR, 1985)
Add to this
Corrections, evidence, and disagreement are welcome — this is knowledge in the open. Anyone can read; sign in with GitHub only to post or react, and it appears here instantly.