Five places treatment center performance breaks down.
Census, reporting, program targets, search visibility and admissions. Where money, time or opportunity slips away between a family's first search and billing.
A center can deliver good care and still lose ground in the handoffs: between the room and the census, between systems that do not talk, between a page and the family looking for it. These are the five we found at one five-center behavioral health network, and what we built for each.
Client kept anonymous. Charts on this page use invented numbers. No patient data on this site.
The map
One family's path, and where each leak sits on it.
Five areas where money, time or opportunity can be lost. Some are confirmed findings. Others are risks the new measurement is built to test. Each section says which.
Leak 1 · Care to billing
Care delivered. Census incomplete.
- The problem
- A service happens in the room, gets documented in the EMR, and then has to be carried by hand into the billing census. Every handoff is a chance for a delivered day to go unbilled, or for a billed day to outrun its documentation.
- What we found
- One operations lead spent more than a day a week retyping attendance logs into the census. The logs carried no medical record number, only a name, so matching a service to its billing episode depended on memory.
- What we built Running
- The engine drafts the census from the raw logs and runs five billing checks on every line, holding anything it cannot confirm for a person. First real run: 637 entries, zero conflicts. The operations lead ran a parallel week against the manual census and signed off. Running today at 3 of 4 outpatient sites; the fourth is moving onto the shared template first.
- What would prove it worked
- Errors caught the same week, before claims are built. After a full billing cycle with rework tracked, fewer corrections and denials traced back to the census.
- The five checks, the 3-hour IOP rule and an estimator are in the billing census checks guide.
| Date | Record | Flag |
|---|---|---|
| 3/8 | MR-4417 | Attended, not in census |
| 3/8 | MR-2298 | Under 3 hours |
| 3/9 | MR-7741 | Billed, no signed note |
| 3/10 | MR-5093 | Matches |
Only the rows that need a person reach the list. Invented records, shaped like the real worklist.
Leak 2 · Reporting
Data everywhere. Answers late.
- The problem
- Census sits in the EMR, attendance in spreadsheets, billing in someone else's workbook. Every question from leadership becomes a request, and every answer is a week old by the time it arrives.
- What we found
- Five centers on four separate Kipu instances, plus site spreadsheets and the billing company's census. Seven people met on Monday and Tuesday to find discrepancies out loud, in a meeting.
- What we built Running
- A morning run reads every center's census from Kipu's API, read-only, and rebuilds the daily census, executive and clinical outcomes boards before leadership starts the day. Aggregates only, no patient names.
- What would prove it worked
- Leadership reads the same numbers on the same morning. Nobody rebuilds the weekly report by hand, and the Monday hunt gets shorter.
Nothing is written back to the EMR. The boards show aggregates, not patient names.
Leak 3 · Reporting
Programs below target. Pressure hidden.
- The problem
- A program can run under its break-even census for weeks before anyone sees it, because the number that would show it is assembled after the month closes. By then the decision is late.
- What we found
- Leadership had a break-even census for each center. The census those comparisons depend on was built by hand once a week, so seeing a center against its own target meant waiting.
- What we built Running
- Each center measured against the break-even targets leadership supplied, every morning, with an estimated value of care and a month-end projection. Estimated value of care is not collections or profit, and the boards say so. How days and dollars differ by plan is a separate calculation: payer mix analysis.
- What would prove it worked
- A center slipping under target is seen within days, not at month end, and the response is decided while it still changes the month.
Census against break-even
Month to date, by center. Aggregates only.
Cedar Park has run under break-even for three straight weeks. Written as a sentence, so nobody has to find it in a chart. Invented centers and numbers.
Leak 4 · Discovery
Useful pages. Poor discovery.
- The problem
- Families search Google, Maps and AI assistants before they call. A center can pay for good pages that search engines barely reach, and never know.
- What we found
- An audit of the four existing center sites: 51 location pages with no links pointing to them, 7 near-duplicate landing pages, 11 thank-you pages open to search engines, and 8 AI training crawlers blocked by a single setting. The sites themselves were fast and well built. These are technical findings, not measured traffic loss.
- What we built Audit done In build
- The audit is complete, every finding checked live. The website and SEO for the network's next center are in build, with search built in from the start.
- What would prove it worked
- Search Console impressions and clicks on the linked pages, before and after, and Business Profile actions per location. No ranking promised. How we measure search.
51 location pages, written and published, reachable only through the sitemap.
Leak 5 · Inquiry to admission
Marketing spend. Admissions unknown.
- The problem
- Ad platforms count clicks and form fills. Nobody can say what an admission costs by channel, so budget follows the cheapest form fill, not the families who actually start care.
- What we found
- Ad reporting stopped at clicks and form fills. Nothing connected an inquiry's source to whether that family was admitted.
- What we built Built In build
- A measurement engine that reports three costs per admission and the reasons inquiries drop, tested on invented data. Being set up for the network's next center, so every inquiry carries its source from the first day. Patient data never goes to an ad platform.
- What would prove it worked
- Cost per admission by channel, with inquiries of unknown source counted rather than guessed, and at least one budget decision made on admissions instead of form fills. How we count admissions.
One paid channel, followed from the first call to the admission. Invented numbers.
Where it stands
Five leaks, five different stages of proof.
At a five-center network, October 1, 2026
| Leak | Found | Built | Running on real dataLive | Result measuredResult |
|---|---|---|---|---|
| 1 · Census3 of 4 outpatient sites | ||||
| 2 · ReportingDaily boards, all five centers | ||||
| 3 · TargetsBreak-even census, daily | ||||
| 4 · DiscoveryAudit done, next center in build | ||||
| 5 · AdmissionsTested on invented data |
Not claimed: a recovered-dollar figure, traffic gains or admission gains. Those need a full cycle of real data after go-live. When there is a real number, it will appear here.
Found means we saw it in the client's own files or sites. Built means finished and tested. Running means it works on real data every day. Result measured is empty on purpose: no leak has a measured financial, traffic or admissions result yet.
The whole story, from billing spreadsheets to a connected view of five centers, is in the case study.
Questions
What operators ask first.
What are the five leaks?
Five places a behavioral health operator can lose money, time or opportunity: care delivered but missing from the billing census, data spread across systems so answers arrive late, programs running below target without anyone seeing it in time, useful web pages that families never find, and marketing spend that cannot be traced to admissions.
Are the findings from a real treatment network?
Yes. The findings come from work at one five-center behavioral health network, kept anonymous. The charts and records shown on this page are invented sample data shaped like the real ones. No patient data appears anywhere on the site.
Where are the IOP and PHP billing checks?
The five billing census checks, including the 3-hour IOP and 5-hour PHP thresholds, the correction window and a browser-only estimator, are in the billing census checks guide.
Which part of your operation still depends on someone joining the numbers by hand?
Start there. Book a free 20-minute call. Nothing to prepare, no data to send.