two saas companies, both at $10M arr, both growing 80% year over year, both in the same category. one trades at $40M. the other trades at $120M. neither founder can fully explain why.
the difference is almost always gross margin, and the version of it on the second company's board deck is almost always wrong.
founders treat gross margin like a number that lives in the accounting system, set once at incorporation and forgotten. that's how a "reported 80% margin" company quietly runs at 62% for two years, raises a round at the wrong multiple, and finds out at exit when the buyer's diligence team builds the schedule properly.
what gross margin actually is
gross margin = (revenue − cost of goods sold) ÷ revenue.
the formula is simple. the fight is over what counts as cogs. for a 2026 saas company, the line items that belong in cogs are non-negotiable: hosting (AWS, GCP, fly), third-party apis (twilio, segment, intercom if it's wired into product), ai inference (openai, anthropic, replicate), payment processing (stripe fees), data costs (clearbit, snowflake usage tied to product features), and the portion of customer success salary attributable to onboarding, implementation, and renewal-blocking support.
that last one is where most founders break the math. cs is a salary line. salaries live in opex. so cs ends up in s&m or g&a, and reported gross margin looks like it does for a 2014 pure-saas business that didn't have a cs function. it isn't 2014 anymore.
these are 2026 multiples for arr. growth, retention, and category nuance move them within a band. they almost never move you across a band. a 62% margin company does not trade like an 80% margin company even if growth is identical, and the gap between the bands is where founders lose tens of millions in implied equity value without noticing.
the 80% that's actually 62%
walk through the math. a $10M arr company reports cogs of $1.2M aws + $400k inference + $400k stripe fees. total $2M. gross margin: 80%. clean.
now build it properly. customer success runs six people at $130k loaded — $780k. four are full-time on onboarding and renewal-blocking support. those four belong in cogs. $520k moves.
inference is metered usage. the founder priced it last year on openai's volume tier. usage tripled with the new model. the actual ai bill ran $700k, not $400k. another $300k moves.
implementation is sold as a "free onboarding service" — three weeks of engineering per enterprise customer, 14 a year. cost: $380k in eng time tied directly to revenue. cogs, whether it appears on an invoice or not.
new cogs: $2M + $520k + $300k + $380k = $3.2M. gross margin: 68%. growth, retention, headcount unchanged. the deck still says 80%.
the multiple impact
at a 10x multiple, the $10M arr company is worth $100M. at 5x, it's worth $50M. in this category, the gap between those two multiples is roughly the gap between an 80% margin and a 62% margin business. if the founder spent two years reporting the wrong number and priced the round against it, they sold $50M of equity to the wrong investor at the wrong price.
it gets worse on exit. acquirers don't accept founder math. they rebuild cogs from invoices. when the schedule comes back at 62%, the founder's offer drops from 8x to 5x, or the deal becomes 60% cash and 40% earnout. "we've adjusted for normalized cogs" is one of the most expensive sentences a founder can hear, and it's said about cs and inference attribution in roughly every saas exit since 2023.
two years of misclassified cogs is not an accounting error — it's an unpriced option you sold the next buyer.
why this gets misclassified
three reasons it happens to almost every saas company under $30M arr.
one. the accountant doesn't push back. most fractional cfos and accounting firms inherit the company's existing chart of accounts. if cs has always been in opex, it stays in opex. the firm is bookkeeping, not building the diligence schedule. that conversation happens once, in due diligence, when nobody can fix it retroactively.
two. the founder wants the high number. 80% margin looks like a clean saas business in a deck. 62% looks like a services business. the founder is pattern-matching to the comps they want to be valued like, not the comps they actually are.
three. operating ratios drift faster than the chart of accounts. when revenue mix shifts from 90% self-serve to 60% enterprise in eighteen months, implementation goes from zero to fifteen percent of every enterprise dollar, and the books take a year to catch up.
the cadence
monthly, alongside the close. the version of gross margin that lives on the dashboard should not be the "ideal" version — it should be the version a sophisticated acquirer would build. if those two numbers are different, the deck is selling a multiple the business can't defend.
what to watch: cogs as a percentage of revenue by component. when ai inference creeps from 4% to 9% over two quarters, that's a 5-point margin compression that nobody announced. when cs scales linearly with customers but stays in s&m, the multiple it implies is fiction.
how zift handles this
zift computes gross margin with cogs attributed by component — hosting, inference, payment fees, attributable cs, implementation — from your stripe + aws + payroll data, every fifteen minutes. on the first of the month you see the number, the delta from last month, the component that moved it, and how it compares to the multiple band your category trades in.
if you're a finance lead at a series A team running this across multiple products with different cost structures, zift handles that too.
most saas companies sell once. the cogs schedule gets built once. the only question is whether the founder built it two years early or the buyer builds it on the founder.