For most of the last decade, growing a behavioral health organization meant the same two moves: hire more admissions reps and spend more on ads. That model is stalling. Reimbursements are flat or falling, out-of-network rates have been cut, and paid search keeps getting more expensive. A recent Dazos panel with David Farache, Dan Gemp of Gempire Advisory, and Robert Harrison of Recovery Unplugged laid out where AI in behavioral health is already changing the economics, and how to adopt it without creating compliance risk.
Why has the old growth model stopped working?
Every additional dollar now buys less than the one before it. David Farache noted that in-network reimbursements rarely rise enough to cover basic inflation, while out-of-network policies have seen steep rate cuts and paid keywords cost more each year. Meanwhile the manual operating model wastes the demand it generates. Robert Harrison said Recovery Unplugged ran 35 admission reps and still missed 20% of its inbound calls, which is roughly the industry average. Some centers he has reviewed miss more than half of their inbound traffic. When acquisition costs keep climbing and staff cannot keep pace with call volume, scaling by headcount alone reaches a ceiling.
Where does AI in behavioral health give a team its time back?
The most reliable gains come from removing manual work that never required a person. Harrison described using AI to cut about 15 minutes from each psychiatric evaluation. Across roughly 100 psych evals a week, with doctor time near $200 an hour, that reclaims meaningful cost every month. Recovery Unplugged also runs about 2,800 pre-assessments a month, and shifting 800 to 1,000 of those to AI saved reps 20 minutes of redundant intake each. The pattern is consistent: let AI handle the repetitive data capture, and let clinicians and reps spend their hours on what they were hired to do.
How does AI protect revenue during admissions?
Speed and accuracy in admissions decisions translate directly to collected revenue. Harrison explained that AI can run a verification of benefits and estimate expected reimbursement before a rep commits to an admission. That turns a high-stakes judgment call into a data-backed one. He was candid about the downside of getting it wrong: admit the wrong policy or miscalculate patient responsibility, and a month of care you expected to net $15,000 can settle at $2,000. Out-of-network cases can swing $50,000. Dan Gemp added that tying billing data back to marketing channel and admissions rep reveals which reps consistently admit the best-fit patients, so operators can route inquiries accordingly.
How does AI improve return on marketing spend?
Dan Gemp argued the first move is to stop wasting demand you already bought. His operating principle for the industry: a missed call is a wasted marketing dollar, so every organization should run a zero-missed-call environment. Agentic AI can answer overflow and after-hours inquiries, then call people back on a cadence that human reps rarely keep. Gemp also connected qualification to clinical outcomes. Better patient fit produces lower AMA rates and stronger long-term recovery, which now shows up as higher revenue per marketing dollar, tracked by channel. Harrison gave a real example: an expensive lead vendor delivered ten admits that were the wrong policies in the wrong state, and with the data in hand, his team pushed the vendor to reconfigure its campaigns.
What does responsible AI adoption require?
The risk profile in behavioral health is specific, and the panel treated it seriously. Dan Gemp raised a red flag about in-house “vibe coders,” where one analyst dropping billing data into a free ChatGPT session becomes a HIPAA violation and a possible data leak. His framing of build versus buy came down to exposure rather than ethics: you can build almost anything now, but a single public mistake becomes a fine and a competitive liability. David Farache reinforced that most new AI vendors have no healthcare experience, and that you cannot simply sign a BAA with Claude or OpenAI, because HIPAA-compliant AI has to be built for it. Harrison shared a cautionary story of a peer whose weekend-built website bot collected patient data with no BAA and no idea where it was stored. The safer default for most operators is to pass the liability to a vendor that builds for behavioral health, and to set a governance framework before deploying anything.
Where is this heading over the next few years?
The panel expects operations to become far more automated at the edges while people focus on judgment. David Farache described a near future of an autonomous CRM where reps enter no data at all and reports are produced from a sentence. Dan Gemp and Robert Harrison both pointed to a more transparent market, where a prospective patient can ask an AI about a treatment center and get an instant, sourced answer, which makes the content you publish about your organization more important than your website ever was.
See the full panel discussion on the Intelligence Advantage on-demand page, then book a session with the Dazos team to map AI to your own operation.
