DSO gaps are concentrated
The difference between a 47-day and a 38-day DSO is almost never general improvement. It is two or three large customers whose payment behaviour nobody has addressed because nobody has looked per customer.
Research · updated August 2026
Finance benchmarks are widely published and mostly useless, because they report medians across company sizes and industries that have almost nothing in common. These are narrower — US mid-market, segmented by revenue band — and they include the figures that make us look bad.
Send a few figures and we will place you against the band and tell you which gaps are worth closing.
Bar is the median for the $25–100M band; marker is the upper-quartile figure. Better is lower on every row except straight-through rate.
What we found
Each of these is cheap to measure in your own systems and most companies have never measured any of them.
The difference between a 47-day and a 38-day DSO is almost never general improvement. It is two or three large customers whose payment behaviour nobody has addressed because nobody has looked per customer.
Fully-loaded finance cost divided by invoices processed. The median is $4.80 and the upper quartile is $3.90. Most teams have never computed it and are surprised by the number.
8.4% median in vendor masters, and every team we tell this to believes theirs is lower. It is measurable in an afternoon and almost nobody has measured it.
2.4 finance staff per $10M revenue at the median, with a very wide spread driven by entity count and industry rather than by efficiency. Treat this one carefully.
Median 34% against upper quartile 58%. It is the metric with the most room in it and the one most directly addressable.
Almost every function we examine has at least one balance sheet account that has not been tied out in over a year, and usually nobody knows which.
A benchmark that aggregates a $12M agency with a $900M manufacturer produces a median that describes neither. Entity count, industry, and business model drive most of the variance in every figure here, and a benchmark that does not segment on them is reporting noise with a decimal point.
These are narrower — US mid-market, segmented into three revenue bands — and even so we would treat them as a rough position rather than a target. The useful comparison is against your own prior period, and the useful benchmark is the one that tells you which gaps are worth closing.
Company-level DSO is an average that hides its own explanation. A function at 47 days typically has most customers paying within terms and three or four large ones consistently at 65 or 70, and those few explain nearly all of the gap to the upper quartile.
That is actionable in a way the average is not. Collections effort spread evenly across the ledger is effort spent mostly on customers who were going to pay anyway.
We tell every prospect their vendor master probably has a duplicate rate around eight percent, and almost all of them believe theirs is cleaner. In sixty-one measurements, three were below three percent.
The cost is not the storage. It is fragmented spend that weakens negotiating position, duplicate payments that eventually occur, and inconsistent terms across records that should be one. All of it is measurable in an afternoon with a fuzzy match on name, tax ID, and bank account.
Finance headcount per revenue is the most commonly cited and least useful figure here. The spread is enormous and it is driven by entity count, transaction volume, and industry rather than by anything resembling efficiency.
A nine-entity group at 3.1 may be running a tighter function than a single-entity business at 1.8. We publish it because it gets asked for, with the caveat that we would not act on it without understanding what drives the difference.
Straight-through processing rate — the share of transactions completing without a person — has the widest gap between median and upper quartile of anything we measure, and it is the most directly addressable.
It is also the one that compounds. A function at 58% has fundamentally more capacity for the same headcount than one at 34%, and that capacity goes into analysis rather than into processing.
Sixty-one US mid-market finance functions measured between 2024 and 2026 during health checks and early engagement, across revenue bands of $10–25M, $25–100M, and $100M+. Headline figures on this page are for the $25–100M band unless stated.
Figures are computed from system data rather than reported by teams. Where a figure could not be computed reliably — cost per invoice requires a headcount allocation we could not always establish — that function is excluded from that metric rather than estimated.
The sample skews toward services businesses, software companies, and professional services, and under-represents distribution and manufacturing. It also skews toward functions with a known problem, since that is often why we were engaged.
No function is identifiable in the aggregates, and every metric here has at least fifteen contributing functions in the reported band.
Questions
Duplicate rate, cost per invoice, and straight-through rate take days and most teams have never computed any of them.