Agents in Recovery

AI for CEOs, owners, and executives

Leaders don't need to pick every tool. They need to decide where AI is allowed, who approves it, and what never goes into it. The rest follows from those calls.

Who this is for

CEOs, founders, owners, executive directors, COOs, and CTOs of addiction treatment centers, mental health practices, and recovery programs. If you sign vendor contracts or answer to a board or investors, start here.

Your top challenges, and where AI can help

1. Staff time goes to paperwork and phones

The problem. Clinicians spend evenings on notes. Admissions reps juggle too many calls. Good people burn out on work that isn't care.

How an agent can help. Pick one bottleneck and pilot one tool against it, with a person reviewing every output. Documentation and call coverage are the usual starting points: see EHR-embedded note assistants and overflow call answering.

2. You don't know what AI staff already use

The problem. Someone may be pasting notes into a free chatbot right now. Most leaders find out after the fact.

How an agent can help. Run an AI tool inventory, then teach the rules with a short staff quiz. An agent can draft the survey and summarize answers. A person decides what stays.

3. Every vendor says HIPAA compliant

The problem. Pitches blur together, and claims are hard to check.

How an agent can help. Ask every vendor the same questions, in writing: BAA, subcontractors, data retention, model training, and what happens when you cancel. Track the answers in a vendor BAA tracker. Our vendor evaluation guide has the full list.

4. Calls go unanswered after hours

The problem. People reach out at night and on weekends. A voicemail can mean they call someone else, or no one.

How an agent can help. An after-hours voice agent can collect basics and hand off to on-call staff. Before it goes live, run a crisis phrase test suite so crisis statements always reach a person. Call quality review helps leaders coach the team.

5. Board and owner reports take too long

The problem. Pulling numbers by hand is slow, and spreadsheets with names travel further than they should.

How an agent can help. Use aggregate reports only: no-show patterns, denial patterns, occupancy and length of stay, and a small-count check before anything leaves the building.

6. CTOs inherit a tangle of data flows

The problem. EHR add-ons, phone systems, CRMs, and chat widgets each move patient data somewhere new.

How an agent can help. Map every data flow before adding a tool. A built-in EHR feature such as a chart summary may stay inside a system you already have a BAA for, but confirm where the data goes. HHS's risk analysis guidance expects you to assess risk when you bring in new technology. Also run a website tracking audit.

Three starter steps

  1. Run the AI tool inventory this month, and name one person who owns AI decisions.
  2. Write a one-page AI policy: who approves tools, what data can never go in, and who reviews outputs.
  3. Pick one pilot with a clear admin-time goal, human review on every output, and a fixed date to decide whether to keep it.

Risks and guardrails for this role

  • BAAs. Any vendor that creates, receives, stores, or transmits PHI for you is a business associate. See HHS's business associate guidance.
  • 42 CFR Part 2. If you run a Part 2 program, consent and redisclosure rules apply to AI tools that touch SUD records. The HHS Part 2 fact sheet covers the 2024 changes, with compliance required by February 16, 2026.
  • Referral payments. Any marketing or call-center deal that pays per lead, call, or admission raises issues under 18 U.S.C. § 220 (EKRA). Flat fees only, reviewed by counsel.
  • AI phone calls. The FCC confirmed in February 2024 that AI-generated voices count as artificial voices under the TCPA, so outbound AI calls generally need prior express consent.
  • Clinical decision tools. The Section 1557 rule at 45 CFR 92.210 requires covered entities to make reasonable efforts to identify and reduce discrimination risks from patient care decision support tools, including AI.
  • State AI laws. Illinois limits AI in therapy under the Wellness and Oversight for Psychological Resources Act. Check every state you serve.

Every use case below keeps a person in charge of the decision. Agents gather, draft, sort, and remind. People approve.

Matching use cases (14)

Frequently asked questions

Where should a treatment center CEO start with AI?

Start with an inventory of the AI tools staff already use and a one-page policy on what data can go into them. Then pick one administrative pilot, such as documentation support or after-hours call coverage, with a person reviewing every output.

Do we need a written AI policy?

A short one helps. It should say who approves AI tools, what information can never go into them, who reviews outputs, and how staff report problems. Have your compliance officer and counsel review it.

Is an EHR's built-in AI safer than a separate tool?

Not automatically. It may keep data inside a system already covered by your BAA, but some features send data to another subcontractor. Review it like any new tool and ask where the data goes.

Who should own AI at a small program?

One named person, often the compliance officer, operations lead, or CTO, with authority to approve or stop a tool. Clinical leadership should sign off on anything that touches care.