One playbook, different results by location — behavioral health

Same brand, three different businesses

PRESENTATION — WHAT THE OPERATOR SEES

Same name over the door, same playbook, same spend per site — and the sites do not behave alike. One is full with a waitlist, one is half empty, one posts admissions that look excellent until you notice its average length of stay is four days. The group meeting compares admissions per site and gets nowhere, because admissions per site is not a comparable number across levels of care.

DIFFERENTIAL — LIKELY CAUSES, MOST LIKELY FIRST

01
The sites are not the same business — OTP versus non-OTP is a six-fold difference in census

An opioid treatment program carries a standing outpatient population; a residential programme turns over a small bed count. Comparing them on admissions, census or cost per admission is a category error, and it is the first segmentation question in this vertical — ahead of residential versus outpatient.

Median clients per facility at 29 March 2024: SUD facility with an opioid treatment program 185 (217 for total outpatient); SUD facility without an OTP 30 (34 total outpatient). Within OTP facilities by operation: private for-profit 200, private non-profit 164, local/county/community government 220, state government 131.

SAMHSA, N-SUMHSS 2024 Annual Detailed Tables, Tables SU12 and SU13

NOT THIS IF — Every site holds the same licence and the same level-of-care mix. Then the variance is operational, not structural.
02
Level-of-care mix sets length of stay, and length of stay sets everything downstream

A site with detox will always show more admissions and less census per admission than a site with long-term residential. Any metric with admissions in the denominator or the numerator will diverge for reasons that have nothing to do with performance.

Median length of stay at discharge, 2023: hospital inpatient detox 4 days, free-standing residential detox 5, MAOT detox 6, residential/rehab in a hospital 7, MAOT outpatient 10, short-term residential 21, MAOT residential 25, long-term residential 36, IOP 40, non-intensive outpatient 53. Overall median 23 days. Means run far above medians in every ambulatory category — non-intensive outpatient mean 123 against median 53.

SAMHSA TEDS 2023, Table E-4

NOT THIS IF — All sites deliver the same level of care and their ALOS distributions overlap. Then look at utilization and referral geography.
03
Occupancy variance of this size is normal in the segment, and the group mean hides it

The national distribution is genuinely wide. A group mean of 85% can be one site at 100% and one at 55%, and the group-level metric will look healthy while half the estate is failing.

SUD residential utilization, 29 March 2024, across 2,852 facilities reporting beds: 482 at 50% or less, 1,239 at 51–90%, 878 at 91–100%, 253 above 100%. Ownership spread is large — SUD hospital inpatient runs 78.3% at private for-profit facilities against 107.7% at private non-profits, and 205 of 812 hospital-inpatient facilities (25.2%) sit at 50% utilization or less. Mental health inpatient runs 97.0% nationally against SUD residential at 85.1%.

SAMHSA, N-SUMHSS 2024 Annual Detailed Tables, Tables SU15, SU16, MH6/MH13

NOT THIS IF — Every site sits inside the 51–90% band. Then the variance being discussed is smaller than the national noise floor and probably is not the real problem.
04
Site size is the variance — you are running a large site and a micro-site under one brand

Outpatient SUD is barbelled; residential is a long tail with almost no top. A group holding one large outpatient programme and two sub-15-client residential programmes is operating two different companies with one marketing plan.

Facility size at 29 March 2024: outpatient SUD (n=12,091) — under 15 clients 27.3%, 15–29 14.1%, 30–59 16.1%, 60–119 14.8%, 120+ 27.7%. Residential SUD (n=3,333) — under 15 clients 50.4%, 120+ 2.2% (73 facilities nationally). Hospital inpatient (n=1,138) — under 15 clients 63.0%.

SAMHSA, N-SUMHSS 2024 Annual Detailed Tables, Tables SU14a / SU14b / SU14c

NOT THIS IF — All sites sit in the same size band. Then compare intake protocol and referral mix instead.
05
Payer and referral geography differ by site, and neither is in the playbook

A site near a drug court, a site inside a hospital catchment and a site selling to families out of state have three different demand engines. Payer mix then determines what a full building is worth.

Acadia Healthcare FY2025 payer mix across 277 facilities in 40 states and Puerto Rico: Medicaid 57.7%, commercial 24.6%, Medicare 14.3%, other 3.4%; revenue by service line — acute inpatient psychiatric 55%, specialty treatment 17%, comprehensive treatment centers 17%, residential treatment centers 11%. In the publicly funded population, criminal justice/DUI accounts for 24.5% of admissions and self-referral for 49.7%.

Acadia Healthcare FY2025 Form 10-K; SAMHSA TEDS 2023, Table D-1

NOT THIS IF — Payer mix and referral mix are within a few points across sites and the variance survives normalisation.

HOW TO TELL THEM APART

How to tell these apart in your own numbers

Each of these is a measurement you can run yourself, without us.

01 · The sites are not the same business — OTP versus non-OTP is a six-fold difference in census

Before comparing anything, label every site: OTP or non-OTP; then level of care; then setting (residential, hospital inpatient, PHP/IOP, outpatient). Compare only inside a label — it is the first step in reporting a behavioral health group and the one most groups skip.

CONFIRMS IF

The sites that look like outliers are the ones with a different label. The variance was in the taxonomy, not the operation.

EXCLUDES IF

Sites carrying identical labels still diverge — that is the real finding, and everything below applies.

02 · Level-of-care mix sets length of stay, and length of stay sets everything downstream

Per site, per level of care: patient days divided by admissions, against the TEDS median for that level of care. Report admissions and ALOS together, never separately.

CONFIRMS IF

The high-admissions site is a detox or short-stay site and its ALOS matches its level of care. The metric was misleading; the site is fine.

EXCLUDES IF

Two sites at the same level of care show materially different ALOS. Then it is utilization review, discharge planning or AMA rate — the only published AMA anchor in the vertical is a single named operator moving from 27% to 19%, vendor-published and not a benchmark.

03 · Occupancy variance of this size is normal in the segment, and the group mean hides it

Designated-bed utilization per site on one reference day, on a single bed basis held constant across the group, placed in the SAMHSA bands.

CONFIRMS IF

Sites straddle the bands — one above 91%, one at or below 50%. Roughly one in six residential facilities nationally sits in that bottom band, so the weak site has company, but the group average is not describing either site.

EXCLUDES IF

All sites cluster inside 51–90%.

04 · Site size is the variance — you are running a large site and a micro-site under one brand

Place each site's census in the SAMHSA size bands for its setting. Use the substance use tables — SU14a/b/c — whose row percentages sum to exactly 100.0.

CONFIRMS IF

Sites land in different bands. A 120+ outpatient programme and a sub-15 residential programme need different plans, different staffing and different reporting.

EXCLUDES IF

All sites land in one band. Note there is no usable mental health equivalent of this table: SAMHSA's own published MH7, MH9 and MH11 row percentages are defective — one row reads 393.8% and the MH11 total sums to 176.6% — so no MH size distribution should be quoted by anyone, including us.

05 · Payer and referral geography differ by site, and neither is in the playbook

Per site: payer mix, and admissions attributed to named referral source categories — alumni, interventionist, hospital, EAP, court, self.

CONFIRMS IF

The mixes differ materially. The playbook was written for one site's demand engine and exported to sites that do not have it.

EXCLUDES IF

Mixes are comparable across sites and the performance gap persists — then it is execution, and it is visible in the intake funnel.

WHAT RESOLVES EACH

What resolves this, and how you will know it resolved

The sites are not the same business — OTP versus non-OTP is a six-fold difference in census	Rx 04 · marketing attribution →	One reporting taxonomy for the group: OTP status, level of care, setting, size band. Most multi-site variance arguments end the week this exists, because half of them were arithmetic.
Level-of-care mix sets length of stay, and length of stay sets everything downstream	Rx 04 · marketing attribution →	Admissions and ALOS reported as a pair, per level of care, against the TEDS medians. Admissions alone is not a performance metric in a business where one site's episode is 4 days and another's is 36.
Occupancy variance of this size is normal in the segment, and the group mean hides it	Rx 03 · patient conversion →	Fix the weak site's front door before adding group spend. Adding demand to a group where one site is above 91% and one is at 50% raises cost per admission at the full site and does nothing for the empty one unless routing between sites actually works.
Site size is the variance — you are running a large site and a micro-site under one brand	Rx 02 · medical practice websites →	Hub-and-location templating so a 120-client outpatient programme and a 12-bed residential programme can share a brand without sharing a page that describes neither accurately.
Payer and referral geography differ by site, and neither is in the playbook	Rx 01 · patient acquisition →	Demand mapping per catchment rather than one national plan. Where a site's engine is a court, a hospital or an EAP relationship, the work is relationship coverage and field visits — and no amount of paid media substitutes for it.

WHAT "RESOLVED" LOOKS LIKE — Spread in designated-bed utilization across sites at the same level of care

MEDIAN

85.1% national for SUD residential; 97.0% for mental health inpatient; 91.2% for SUD hospital inpatient

TOP DECILE

253 of 2,852 SUD residential facilities reporting beds (8.9%) run above 100% designated-bed utilization; a further 878 run 91–100%

TARGET

No site below the 51–90% band, and the group's weakest site inside the national band for its setting. The realistic ceiling for a group is the 91–100% band sustained across sites — 8.9% of individual facilities exceed it, and only because clients occupy non-designated beds. This is a distribution position, not a promise.

SAMHSA N-SUMHSS 2024, Tables SU15, SU16, MH6/MH13; SAMHSA TEDS 2023, Table E-4

HOW THIS DIFFERS BY SCALE

How this differs by scale

Single site	One site with multiple levels of care has the same problem in miniature: the detox line and the residential line are different businesses on one licence and should be reported separately from the first month.
Group	Three to ten sites is where this presents hardest, because the group is large enough to average and too small for the average to be meaningful. Report per site, per level of care, always — and expect one site's undocumented intake protocol to be the group's most valuable asset.
Platform	Platform: the public operators publish the taxonomy to copy. UHS reports licensed beds, available beds, patient days, average daily census, occupancy on both bed bases, admissions and length of stay as one table, same-facility and all-facility side by side. Acadia reports same-facility revenue, patient days, admissions, ALOS and revenue per patient day as percentage changes. If a platform cannot produce those tables per site, the variance conversation cannot be had at all.

OTHER PRESENTATIONS — BEHAVIORAL HEALTH

A differential narrows the field. It does not replace the examination — that is what the six weeks are for. Every figure above is an industry reference range, not a client's numbers; those stay sealed. Sources are set out at /sources.

APPLY — 6 / QUARTER →