Insights · Trust
What "AI in the back office" actually means
When a mental health organization hears "we build with AI," the reaction is rarely excitement. It's a flinch. Clinicians have watched AI get pointed at the most sensitive parts of their work (notes, risk scores, triage) and they're right to be wary. So when we say our AI lives in the back office and never touches care, we owe you a precise account of what that means, not a reassuring slogan.
The line we don't cross
AI inside a Verbena system never triages a client, never generates a diagnosis, never scores someone's risk, and never makes or influences a decision about a person's care plan. Those are clinical judgments, made by licensed people who are accountable for them. We don't put a model in that chair, and we write the prohibition into every Business Associate Agreement so it isn't just a promise, it's a term.
Where it does help
The back office is full of work that is repetitive, deadline-driven, and entirely non-clinical. That's where AI earns its place:
- Moving data between systems so an intake isn't typed three times into three tools.
- Drafting internal reminders for staff to send (a nudge that a candidate's hours are off pace, or a no-show needs a follow-up call).
- Summarizing internal documents like policies or grant requirements so an operator finds the answer faster.
- Assembling reports from data that already exists, so board metrics and funder reports aren't rebuilt by hand each cycle.
In every one of these, a person stays in the loop and the output is operational, not clinical.
Why the distinction holds under pressure
It would be tempting, once AI is in the building, to let it drift toward clinical work. That's where the flashy demos are. We don't, for two reasons. First, it's the right call for clients: care decisions belong to clinicians. Second, it's the durable call for the field: the moment a vendor blurs that line, every clinician in the building stops trusting the software, and adoption dies. Keeping AI in the back office isn't a limitation we tolerate. It's the thing that lets the rest of the work be trusted.
And your data stays yours. We use AI tools that contractually exclude customer data from training. Your information is never used to improve a model, ours or anyone else's. That, too, is written into the BAA.
Want the full picture of how we handle privacy, security, and AI? Read our security & compliance commitments, or ask us a hard question directly.