A private therapy practice was losing vulnerable clients to slow response times. We built a three-lane system that responds to every inquiry instantly, follows up on stalled leads automatically, and generates AI-personalised care plans — all on local infrastructure, so client data never touches a third-party server.
Adhoc Therapy offers private mental health support. The nature of the work means most clients reach out at their lowest point — late at night, during a break at work, in a moment of courage they might not have again tomorrow. They submit a form, and then they wait.
The practice was overwhelmed. A real human reply took anywhere from a few hours to a full day. By then, the window had often closed. The client had either found another provider, talked themselves out of it, or simply lost the energy to follow through. And because there was no tracking system, stalled leads just disappeared — no one knew how many potential clients were slipping away every week.
A cold auto-responder wasn't the answer. What was needed was a system that responded with warmth, remembered every lead, and kept following up — without adding a single hour to the admin workload.
Clients reaching out during vulnerable moments couldn't wait 24 hours. By the time the practice responded, many had already moved on or lost the motivation to start therapy.
There was no central record of who had enquired, who had completed their intake, and who had gone quiet. Stalled leads simply vanished — no follow-up, no second chance.
Standard tools could send an immediate reply — but not one that acknowledged the client's specific situation. A generic "we'll be in touch" felt cold in a context that required warmth.
Client messages describe deeply personal struggles. Routing that data through cloud-based AI tools — OpenAI, Gemini, etc. — posed a real privacy risk the practice wasn't willing to take.
The system was architected in n8n across three independent lanes, each running on its own trigger and handling a distinct phase of the client journey. They share a single NocoDB database as the source of truth for every lead's status.
The moment a potential client submits the sign-up form, the system fires. It checks NocoDB for an existing record — if the person has reached out before, their record is updated; if they're new, a fresh record is created. Either way, a branded welcome email lands in their inbox within seconds, containing a secure, personalised link to the intake questionnaire. No human needed. No delay. The practice's front door is always open.
Life gets in the way. Clients often open the welcome email, intend to fill in the form, and then forget. The retention lane runs on a daily schedule, scanning NocoDB for anyone whose questionnaire link was sent but never completed. It routes them through a timed follow-up sequence — a gentle nudge at Day 3, and a respectful "closing your file" message at Day 7 if there's still no response. Every action is logged back to the CRM so nothing is sent twice.
When a client submits their intake questionnaire, the system reads their AI consent flag before doing anything else. The response path splits based on what the client chose.
Most AI-powered intake systems route client data through cloud APIs — OpenAI, Google, Anthropic. For a general business, that's acceptable. For a therapy practice, where a client might write "I feel so broken, I just lost my mum" in their intake form, it's a problem.
This system runs the intelligence layer entirely on local infrastructure. The Llama 3 model is served via Ollama — self-hosted, no external API calls, no data leaving the server. The AI reads the client's struggle, generates a response, and saves the care plan to NocoDB — all without a single byte of client data touching a third-party service.
The AI model runs on the practice's own server. There is no OpenAI account involved. No Gemini API. No third-party model provider. If the internet connection drops, the intelligence engine keeps working — because it doesn't need the internet. Client data stays on the machine it was submitted to, and nowhere else.
From the day the system went live, Adhoc Therapy's intake process ran without manual intervention.
| Metric | Before | After |
|---|---|---|
| Time to first client response | Hours to next business day | Under 60 seconds, 24/7 |
| Leads that went untracked | All of them — no CRM | Zero — every lead logged with status in NocoDB |
| Follow-up on stalled leads | Memory-dependent, often missed | Automated at Day 3 and Day 7, every time |
| Client data privacy | Unmanaged — data in email threads | 100% local — no third-party AI services used |
| Admin time spent on intake | Several hours per week | Near zero — edge cases only |
| Personalised response quality | Generic or delayed | AI-generated care plan within 60 seconds of form submission |
Every tool was chosen with privacy and reliability in mind — no unnecessary cloud dependencies, no data leakage risk.
The three-lane pattern — instant response, automated retention, private AI fulfillment — is not exclusive to therapy. Any practice where client data is sensitive and response time matters can deploy this exact framework.
Law firms handling sensitive cases can automate client onboarding, follow up on stalled consultations, and generate case summaries — all without routing confidential information through cloud AI providers.
GP surgeries, specialist clinics, and telehealth providers can capture patient inquiries instantly, track intake completion, and send personalised next-step guidance — 24/7, no receptionist required.
IFAs and wealth managers can automate prospect intake, follow up on leads who went quiet, and generate personalised financial health summaries — all within a compliant, locally-controlled environment.
If your practice is losing enquiries to slow follow-ups, or you need an intake system that respects client privacy without sacrificing speed — this is exactly what we build.
Book a Discovery Call// 30 minutes — No pitch — Just your situation and whether we can solve it