A counselor finishes a session at 5:40 on a Thursday. It ran long, because they always run long. She has one more family waiting, her own kid to pick up at six, and a parking situation. She writes the session on a Post-it, tells herself she will log it Monday, and leaves.
Monday there are eleven new referrals sitting in the inbox.
That session happened. A young person got real help from a trained professional. It changed something. And nine months later, when the county pulls the file, it did not happen.
Nobody is doing anything wrong
The first thing to understand is that this is not a discipline problem, a management problem, or a caring problem. Everybody in that building is working hard and most of them are underpaid.
It is structural, and the structure is not an accident. Look at what has happened to small human-services organizations over the last decade.
Reporting requirements went up. Forty-four percent of nonprofits report mid-contract changes from their government funders, including new reporting and compliance requirements with no additional reimbursement attached. The work got bigger. The check did not.
The money to handle it stayed thin. Administrative capacity comes out of indirect cost recovery, and indirect has been squeezed for years. The federal de minimis rate finally rose from 10 to 15 percent of modified total direct costs in October 2024, which helps. But county and foundation contracts routinely cap overhead well below what compliance actually costs, and indirect is where the person who would build the system would sit, if there were a person.
The people got scarcer. Nonprofit turnover runs around 19 percent against 12 percent in other sectors, and at organizations under $2 million it routinely exceeds 25 percent. In human services, frontline vacancy rates have topped 40 percent in some categories. And salary is not the top reason people leave. Fifty-nine percent say it is carrying too much responsibility with not enough support.
Put those three together honestly. More reporting. Not enough money to do it. Fewer people, leaving faster, each carrying more.
The reporting does not stop happening. It gets absorbed by frontline staff, at the end of the day, on a Post-it.
This is not a small problem hiding in small organizations
Before the composite, one number that puts the scale beyond argument.
That distinction matters and it is worth stating plainly, because CMS says so itself: an improper payment does not mean the money was misspent. It means the record did not prove it. Somebody almost certainly delivered that care. The file just could not carry it.
That is the same failure as the Post-it, at national scale, in one program, in one year.
Where it turns into money
Here is a composite. Every number is a modeled illustration, not a specific organization.
A juvenile diversion provider with a county contract. Three counselors. The program spec calls for eight individual sessions, eight group sessions, and six follow-up hours per participant. They serve 250 young people a year. That is roughly 6,000 service hours in a good year.
Assume 6 percent of delivered service never makes it into the record. There is no national statistic for this, which is a point I will come back to. Not fraud, not laziness. Post-its. A counselor who resigned in March. A form that went into a folder instead of the system. An intake that sat three days past the window.
Now the second-order cost, which is bigger.
The county evaluates renewal on documented outcomes. This provider actually served 250 youth. Their file supports 235. Their completion rate looks worse than it is, because a participant who finished session eight without session six recorded reads as incomplete.
They compete for renewal against a provider who serves fewer youth and documents every one of them. And they lose, on paper, while doing better work.
Third-order, and this is the one that ends organizations. Among the most persistent finding categories in this sector: inadequate documentation, and time-and-effort records not supported by after-the-fact activity reports reflecting work actually performed. A finding produces questioned costs. Questioned costs produce repayment demands. GAO linked $1.17 trillion of the $6.97 trillion in federal award spending between 2017 and 2021 to severe, persistent single audit findings.
The threshold moved. The risk did not.
In October 2024 the single audit threshold rose from $750,000 to $1,000,000 in federal spending. A lot of organizations read that as relief.
It is not relief. It is a transfer.
The federal government still holds your pass-through entity accountable for how it monitors you. When the statutory audit goes away, the county has to look harder at subrecipients, not less hard, because the county is now the one carrying the exposure.
If your federal spend is under a million, you did not get safer. You got a different reader, one who sees your numbers every quarter instead of once a year, and who decides whether you are renewed.
Why it lands so hard on small organizations
Because of the balance sheet.
Pull the 990 of almost any community-based provider in the $1M to $5M range and you find the same shape. Revenue roughly matches expenses. Assets are modest. Liabilities eat most of the assets. Net assets often represent two to six weeks of operating expense, against a sector benchmark of three to six months.
These are reimbursement businesses. Deliver the service, invoice the government, wait, get paid, make payroll. There is no buffer anywhere in the cycle.
Now add the thing everybody inside the sector knows and nobody outside it does.
For an organization holding three weeks of cash, a delayed or partially denied reimbursement is not a cash flow inconvenience. It is a decision about whether payroll clears.
So when a documentation gap turns into a rejected line item on an invoice, the consequence is not a spreadsheet correction. It is a phone call to a bank, or a founder personally covering payroll, or a program that quietly stops taking referrals.
That is why it never gets fixed. It is filed under administration, and administration is the line item nobody funds.
Why the obvious fixes do not work
"Buy case management software."
For a team of thirteen, a mid-tier platform runs roughly $830 to $2,000 a month plus a $5,000 to $15,000 implementation fee. Call it $15,000 to $39,000 in year one. Most vendors will not publish a price at all, which means a director cannot take a number to a board without first sitting through a sales call. Then the person who configured it leaves, because turnover, and nobody in the building knows how it works. An unused database is worse than a spreadsheet, because now it costs money and gives false comfort.
"Hire a compliance person."
With what. That salary lives in indirect. The de minimis rate rose to 15 percent in 2024, which helps at the margin, but most county and foundation contracts still cap overhead below what a compliance salary costs, and a partial salary does not produce a system.
"Train the staff harder."
They are not undertrained. They are over-assigned. Adding a documentation expectation to a caseworker already at 130 percent capacity does not produce documentation. It produces turnover, and turnover is what destroys institutional memory in the first place.
"We will build it internally over the summer."
Someone will, and they will do a decent job, and eighteen months later they will take a better-paying job and the system will die with them. That is not cynicism. That is the base rate.
What actually works
The principle is short.
Everything else follows from that.
Capture at the point of delivery, not at the end of the week. If a counselor has to remember, you have already lost. Two taps on a phone in the parking lot, or nothing.
Count automatically against the contract. If the spec says eight sessions, the system should know a participant is at five with three weeks left in the period. An alert before the shortfall is worth a hundred reports after it.
Timestamp on arrival. Every referral gets a clock the moment it enters the building. This single change eliminates the "it sat on someone's desk" failure permanently, and it is nearly free.
Design for the person who has not been hired yet. At 25 percent turnover, the real test of a system is whether a new caseworker can use it correctly on day three with no training. If it needs a champion, it dies when the champion leaves.
Generate the report, do not write it. If a human types the monthly numbers, the numbers are a summary of a summary, and every layer loses fidelity and gains error.
Keep an audit trail. Who recorded what, when, unchangeable. That is the difference between a convenience and a defense.
None of that requires enterprise software. Most of it is workflow design. The reason it does not get done is not cost or technology. It is that nobody inside the building has forty uninterrupted hours to think about it, because they are all delivering service.
Nobody knows the real number, including us
Here is something worth being honest about, because most people selling into this space are not.
There is no published national statistic for the share of delivered service that goes undocumented in American human-services nonprofits. Not from GAO, not from the Urban Institute, not from Candid, not from any federal agency. The Medicaid figure is the closest proxy and it is not nonprofit-specific.
The 6 percent in the composite above is an assumption, not a benchmark. Anyone who quotes you a sector-wide percentage for this is guessing and calling it research.
Which means the only way to know your number is to measure it, in your records, in your pipeline. That is unsatisfying and it is also the whole point.
The one question
If you run one of these organizations, there is a single diagnostic that tells you where you stand.
Ask whoever produces your funder reports to show you how they built last quarter's service-unit numbers.
If they open a spreadsheet and start explaining, you are carrying undocumented hours right now. You do not know how many. Neither do they.
That is not a failure. It is the default state of an entire sector that has been asked to report more with less, every year, for a decade.
But it is measurable. And once you measure it, it is fixable. It is also a great deal cheaper to find out now than to find out in a monitoring finding, or in a renewal you lost to an organization doing less good work than yours.