Book a Free Call
Case Study · Therapy & Behavioral Health

Therapy intake intelligence. Built on clinical guardrails, not chatbot tricks.

An adaptive intake system serving therapy practices — with a self-learning safety net, no-hallucination guarantee, and a full audit trail behind every clinical decision.

EngagementProduction system, ongoing
SectorBehavioral health
StackClaude + Supabase + Next.js
40+Clients onboarded through adaptive intake
A+Average clinical accuracy across test runs
150+Self-learned safety rules in the live system
0Hallucinated demographics or facts
The Problem

Therapy intake was leaking the right clients to silence.

Rigid intake forms

Static fields that ignored the answer they just received. New clients abandoned mid-form. Clinicians inherited half-finished records.

Slow first contact

Inquiries waited days for a reply. By the time someone reached out, the prospective client had already booked elsewhere or lost momentum.

No safety net

High-risk language slipped through generic auto-replies. Practices had no system that read every message for clinical risk before it sat in an inbox.

Lost referrals

Inbound referrals from doctors and adjacent practices fell through cracks — no routing logic, no follow-up, no record of where they originated.

The System

Six layers, one intake experience.

Each layer is independently testable, independently auditable, and independently swappable. Models change. The system stays.

01

Adaptive Intake

The form branches by answer in real time. Every question is informed by everything the client has already said. The path through intake is shaped to that person, not a generic template.

02

Clinical Safety Layer

Every message is read against 150+ safety rules covering risk language, crisis indicators, and clinical edge cases. The system escalates immediately when the threshold is crossed.

03

No-Hallucination Guarantee

The system never fabricates demographics, history, or facts the client did not state. Built-in evals catch any drift before it reaches a clinician’s notes.

04

Self-Learning Loop

Each intake produces signal. The rule library grows automatically — the system today is meaningfully sharper than the system three months ago, with no manual retraining.

05

Routing & Hand-Off

Every lead is routed to the right clinician or queue with a written brief. The hand-off arrives in their inbox, not a separate dashboard they have to remember to check.

06

Auditable At Every Step

Every AI decision — every classification, escalation, and route — is logged with input, output, and reasoning. Months of audit trail are searchable in seconds.

The Outcome

Faster intake. Safer triage. A clinical bar that gets higher every month.

First-contact within minutes

Adaptive intake replies the moment a lead lands — with a personalized response, not a template. Practices stopped losing prospects to slow follow-up.

Risk caught before the inbox

Crisis language is escalated to the right person within seconds. No more waiting for a clinician to scroll through unread mail.

Self-improving without retraining

150+ safety rules grew from the system itself. Every intake teaches the next one. Nothing is hardcoded; everything is auditable.

Stack
Claude AISupabaseNext.jsVercelTypeScriptResendTwilioPostgres RLS

Want this for your practice?

A workflow audit maps your current intake end-to-end and shows where this kind of system would fit. One week. Written plan. Yours to keep.

Book an AI Workflow Audit