Banyan Tree Counseling
AI intake chatbot with crisis detection.
Off-hours intake that converts before the lead bounces. Paid ads driving traffic at midnight are no longer wasted budget — the chatbot pre-qualifies and books while everyone sleeps.
The problem
Banyan Tree (athenscounseling.com) is an Athens-area counseling practice. Their growth problem was specific:
- Most therapy inquiries come outside business hours — evenings, weekends, 2am
- A static contact form gets a 24–48 hour reply time
- By then, the prospect has already booked with someone else
- And they were running paid Google Ads at all hours, paying for clicks that bounced after staring at a contact form
What we built
An AI intake system on the front of the website:
- Cloudflare Worker chatbot powered by Claude — the prospect can have a real intake conversation immediately, any hour
- Georgia-license filter — the chatbot is trained to politely route out-of-state inquiries to appropriate referral resources (Banyan only practices in GA)
- Crisis detection — built-in flagging for self-harm or active crisis language, with immediate escalation to crisis hotlines
- Persona routing worker — separate edge worker that swaps homepage content based on the visitor’s apparent persona (parent of a teen, college student, executive, etc.)
- Booking integration — qualified prospects get a booking link before the conversation ends
Why this is a marketing win
Banyan’s paid ads now convert at any hour. The 24-hour-old “contact form lead” went away — every paid click either books a session, qualifies into a callback queue, or self-routes to a referral. Nobody bounces because they couldn’t get an answer.
The persona-routing worker quietly improved the conversion rate of the cold traffic landing pages by personalizing the language without anyone needing to manage A/B tests.
Tech stack notes
- Edge-deployed (Cloudflare Workers) so the chatbot responds in under 500ms even from East Asia — Banyan picks up some out-of-state Stanford parents looking for Georgia therapy for their college student
- Crisis detection is a non-trivial classification problem; we use Claude as a primary filter with conservative thresholds and human escalation
- Compliance: nothing the chatbot collects is treated as PHI until a human takes the case
Got a backend that's holding back your marketing?
We'll take a look. No obligation. We don't take every project — but if it's a fit, we'll tell you.