Hire a health coach or train existing staff?

Who you hire matters less than what the role comes with

It usually starts at the desk after hours. A half-written job posting for a health coach on one screen, the schedule on the other, and a salary number you are not sure the panel can carry. In the middle sits a medical assistant already at capacity who would be good at this, given four uninterrupted hours.

So the question forms: should I hire a health coach or train existing staff?

It is aimed slightly off target. The research on non-specialist delivery never compared credentials, it compared structures. Programs that held their gains had a manual, a named supervisor, and calendar time nobody could take. Those that faded were missing one, whether the deliverer was a nurse, an assistant, or a specialist-supervised coach.

This is the decision that follows why patients don't follow treatment plans, and it sets your ceiling for the year.

Option one: hire a health coach

One person whose whole job is the between-visit work.

1. 01

   Cost shape

   A standing salary and benefits from day one, carried whether or not the panel fills the hours. Comfortable at scale, painful earlier.

2. 02

   Hiring or vendor risk

   Concentrated in one person, and credentials vary more than titles suggest. Only 247 professionals hold Board of Behavioral Sleep Medicine certification, 224 in psychology, 16 in medicine, 7 in nursing (McCrae et al., 2025).

3. 03

   Supervision load

   The highest of the three, and it lands on you. In the trial most cited as proof coaches can do this work, non-clinician sleep coaches treated 159 older veterans and produced large early gains, all under a behavioral sleep medicine specialist (Alessi et al., 2016).

4. 04

   Fidelity to protocol

   Entirely dependent on what you hand them. A coach with a manual follows a method. Without one they improvise.

5. 05

   Time to stand up

   The longest: posting, interviewing, credentialing, training, ramp. Six months to steady state is realistic.

Option two: train existing staff

The person is already in the building and trusted.

1. 01

   Cost shape

   The lowest headline cost and the highest hidden one. You are not buying hours, you are reassigning hours you already own.

2. 02

   Hiring or vendor risk

   The lowest of the three. Risk moves rather than disappears: a trained medical assistant is still an assistant, and clinical work wins any calendar conflict.

3. 03

   Supervision load

   Still yours, and harder, because you supervise one person across two roles with different failure modes.

4. 04

   Fidelity to protocol

   The evidence cuts both ways. In a 600-patient pragmatic trial randomized by clinic, medical assistant health coaching roughly doubled the rate of HbA1c improvement seen in usual care, though the authors noted flexible coaching frequency left some patients under-dosed (Fortmann et al., 2024). Another program funded two days a week of medical assistant effort plus outside certification and psychologist instruction, yet only 49% of patients completed 12 weeks and 45% of providers reported difficulty referring (Djuric et al., 2017).

5. 05

   Time to stand up

   The fastest on paper. In the HABIT trial, primary care nurses delivered brief sleep restriction therapy after roughly four hours of training, beat sleep hygiene advice, and were judged likely cost-effective (Kyle et al., 2023). Speed to launch and durability at month twelve are different variables, the core of how to improve patient adherence between visits.

Option three: embed an outside behavioral service line

Not outsourcing in the billing-vendor sense, but a delivery layer arriving with its own protocol, supervision, and contact cadence, which is what is telehealth coaching effective examines.

1. 01

   Cost shape

   Variable rather than standing, tied to patients enrolled instead of a salary line. Less to carry in a slow quarter, less upside at scale. The dollar model is handled separately, in our piece on the economics of an embedded service line.

2. 02

   Hiring or vendor risk

   No single point of failure in your building, but you inherit someone else's turnover and quality control. What to ask before signing is its own checklist, covered in our piece on evaluating a coaching partner.

3. 03

   Supervision load

   The lowest of the three, but only if supervision is written into the arrangement. If a partner cannot say who supervises their coaches and how often, the load returns to you.

4. 04

   Fidelity to protocol

   Highest when the protocol comes with the service, no better than a solo hire when it does not. Ask not whether they have a method but whether you can read it.

5. 05

   Time to stand up

   Weeks rather than months, because the manual, the supervisor, and the contact schedule already exist. You choose a structure rather than build one.

Why behavioral programs work at first and then fade

Behavioral sleep change runs on two mechanisms. Sleep restriction shortens time in bed to rebuild sleep pressure, the biological drive for sleep that builds the longer you are awake (Kyle et al., 2023). Stimulus control re-pairs the bed with sleep rather than with lying awake (Espie et al., 2025). Both run in weekly increments, with someone troubleshooting adherence as the plan meets real life.

Strip out supervision or sustained contact and the method degrades into advice patients already have.

The follow-up data shows this shape. That 159-patient coach-delivered trial produced a large improvement on the Insomnia Severity Index, the standard symptom questionnaire, right after treatment, a moderate effect at six months, and a small one at twelve. Primary care personnel given two days of training showed a large reduction at post-treatment and a small one at 18 months (McCrae et al., 2025).

Notice what decayed. Not the credential. The dose.

Four conditions decided every result, and none of them is a credential

Across these trials, the same four features separate the models that held from those that faded.

1. 01

   A written protocol

   Specific enough that two people delivering it would do roughly the same thing.

2. 02

   Ongoing supervision by someone with real depth

   Not an annual review. A regular case conversation with someone who knows the method well enough to catch drift.

3. 03

   Protected, non-interruptible contact time

   Time clinical overflow cannot take, the variable examined in does contact frequency improve patient outcomes.

4. 04

   Contact that survives past week eight

   Most decay appears in the six to twelve month window, after the program has declared success (McCrae et al., 2025).

Where the evidence is still contested

The field is not settled. In 2025 the journal SLEEP ran paired PRO and CON pieces on whether a range of providers should deliver CBT-I. One side argues manual-based skill is more a function of training than of clinical experience (Espie et al., 2025). The other reads the follow-up data as showing non-specialist delivery decay (McCrae et al., 2025). Neither disputes that structure does the work.

At Clementine, the embedded coaching service line is built around those four conditions rather than a headcount, so supervision and contact cadence arrive attached to the protocol.

Protect the time before you pick the person

Most practices do this in the opposite order. Before you post the job or take the vendor call, block the contact time on next quarter's schedule and mark it non-interruptible. Then name, on paper, who supervises the work and how often. If you cannot do both with the schedule you have, no hire will rescue it and no vendor will either.

What not to do when you hire or train for this

Do not treat a certification as a substitute for a protocol. A credential says someone finished a curriculum, not what they do with a patient who stopped filling in the diary.

Do not fund the role out of slack time. When hours are unprotected the dose becomes flexible, the limitation the medical assistant coaching authors flagged in their own positive trial.

Do not stop at week eight, where results look best and decay has not appeared. And never sign a partner who cannot name their supervisor.

Which of the three actually fits your schedule

Hire if the panel keeps one person busy, a supervisor with behavioral depth already works there, and you can wait six months. Train if a staff member has real capacity, funded and protected hours, and a manual to follow. Embed if the panel is real but not large, supervision bandwidth is scarce, and you need something this quarter.

None is safe in the abstract. The safe choice is the one you can still supply with a protocol, a supervisor, protected time, and month-nine contact a year from now.