The outcomes teams like yours see with Tars

First sessions

Booked inside the conversation, on your clinicians' real calendars

Coverage answered

Plan, network status, and self-pay cost, on the spot

Clinician match

Specialty, modality, and availability in one exchange

After hours

The 11pm decision still becomes an appointment

Convert

Between deciding to get help and sitting in the first session, there are four places practices lose people.

Settle the coverage and cost question before they leave the page

Deciding was the hard part. The next question is smaller: do you take my insurance?
The agent works from the payer list, plans, and self-pay rates you give it.
It asks which plan they carry, then answers in network or out, and what a first session or an initial psychiatric evaluation costs.
Real eligibility checks route to your billing team, plan and question already captured.
Nobody leaves at the cost question without an answer, and the first session is still on the table.

Match the person to a clinician they will actually see

"Find a therapist" is four questions at once: availability, specialty, modality, and fit.
Your own clinician directory is the source: specialties, modalities, languages, telehealth or in person, current openings, and which prescribers take new patients.
The agent asks the few questions that narrow it, then shows the clinicians who genuinely fit and the times they have.
It never promises that a specific clinician is available, and it never answers a symptom description by naming one.
More people land on a clinician who will still be their clinician in six months.

Capture the decision at the exact hour it gets made

The decision arrives at 1am on a Sunday. Your front desk is closed and the form promises a callback.
Something is awake: the agent has the real conversation on the channel they arrived on.
It books the first session, or captures a qualified record with the plan, the preference, the urgency, and the source attached.
When someone is ready now, your team gets an alert with the full context.
Weekend and after-hours demand stops expiring into voicemail, in a category where people contact the next practice on the list.

Bring back the person who read everything and booked nothing

People ask two good questions and then stop, to check with a partner or look at the deductible. That is not a lost lead.
Channel is a property of each message, so a return visit reopens the same conversation.
The agent already knows what they asked, which plan they carry, and which clinician they were looking at.
You can follow up by email, SMS, or WhatsApp with the clinician's next opening or the cost they left to check.
The people closest to starting stop falling out of the funnel.

Platform

Four things you have to be sure of before an AI agent talks to people about their mental health.

Decide in advance exactly what the agent is allowed to say

You open the demo trying to break it. Name me a therapist for what I have.
Getting a clinician, a plan, or a symptom wrong is a legal problem.
The agent answers from your approved content and hands off anything outside it. No diagnosis, no treatment guidance.
Crisis response runs as a deterministic flow you write once: the 988 Suicide and Crisis Lifeline, intake stopped, a live person brought in, the same sequence at 3am as at 3pm.
Your clinical lead can read the whole crisis path in one sitting and sign it.

Turn the conversation into a session on a real calendar

Your conversion event is not a form submission. It is a first session on a specific clinician's calendar.
Booking happens inside the conversation, against live availability from your clinicians' own Google or Outlook calendars.
The agent books against your rules: session length, buffers, minimum notice, working hours, and the person's time zone.
Multiple clinicians can be assigned round robin, least busy, or in a priority order you set, and each gets the conversation attached.
Everyone gets a calendar invite, and the first session lands while the person is still resolved to go.

Trace every first session back to the campaign that paid for it

You know the visits and the form fills. Which campaigns produced people who actually started therapy is the gap.
One analytics home shows conversations, engagement, escalations, and conversion, each compared against the previous period.
Attribution breaks results down by campaign, UTM source, and landing page, so you see which spend produces booked first sessions, not conversation counts.
Bookings count only when the write to your CRM went through, and every number is clickable to the individual conversation.
Walk into a budget review with the first sessions and the campaigns that paid for them.

Prove the agent behaves before a single person sees it

Your clinical lead tries to break it, and legal asks what happens on the worst day.
Tars runs that review as a product step, so your reviewers sign off on what they watched the agent do.
Simulated people, driven by personas and goals from your content, hold multi-turn conversations with your actual agent.
The crisis path and the clinical boundary rules are scored on every version, by code checks and by an LLM judge.
Deploy when the scores clear your threshold, then keep improving on the questions people actually ask.

How Tars Agents Get Better

The healthcare acquisition flywheel

Putting an agent in front of people asking about care is not a click-a-button decision. Your clinical, legal, and marketing leads all have to sign off first. Tars closes the loop end to end. Train, test, deploy, learn, improve. More inquiries answered and more first appointments booked with every cycle.

Step 1: Train

Connect your service lines, locations and provider directory, accepted payers, pricing and policy pages, intake criteria, and past inquiry history. Set the guardrails and the exact wording for sensitive topics. The agent presents your care the way your best intake coordinator does, from your content and your rules.

Step 2: Test

Run simulated inquiries against the agent before launch, including the ones your reviewers are most worried about. Your escalation protocol and clinical-boundary rules become standing evaluators, so you see how the agent behaves on the hard cases before a single visitor does.

Step 3: Deploy

Go live on web, WhatsApp, SMS, and email when the scores clear your threshold, with booking and CRM write-back switched on from day one.

Step 4: Get Insights

See where inquiries drop off, which campaigns and landing pages produce booked appointments, and how conversion differs by location and service line. Every number opens down to the individual conversation behind it.

Step 5: Improve continuously

Close the gaps, re-test, and raise booked appointments month over month. Each cycle converts more of the traffic you are already paying for.

Create AI Agent

The healthcare acquisition flywheel: train, test, deploy, get insights, improve continuously

Frequently asked questions

Could it say something wrong about a clinician, a plan, or a symptom?

That is the right first question, and the answer is structural. The agent answers only from content you approve and declines when a question falls outside it. It does not diagnose, assess, or advise on treatment. On clinicians specifically, it presents who fits and who has availability from your own directory, and it never implies that a particular clinician is guaranteed or that they are the clinical answer to a condition. On coverage, it says what your payer list actually supports and routes anything needing a real eligibility check to your team. Before launch you run your own worst-case inputs at it, and those become standing tests it has to keep passing.

What happens if someone in the chat is in crisis?

A crisis protocol that you write runs, deterministically. You define the trigger language and the exact response: the agent surfaces the 988 Suicide and Crisis Lifeline and any resources you specify, stops asking intake questions, and escalates to a live person immediately with the full conversation attached. It runs the same sequence on web, WhatsApp, SMS, and email, and at 3am exactly as at 3pm. Your clinical lead signs off on one written protocol and knows what a visitor will be told.

Is Tars HIPAA compliant?

Yes, with a BAA available, alongside SOC 2 Type 2, ISO 27001, and GDPR. Because someone can disclose a condition or an intention on a public page, you also get role-based access, retention periods you control, an audit trail per conversation, and explicit control over what the agent can access and store.

Can it book onto our clinicians' actual calendars?

Yes. Booking happens inside the conversation, against live availability read from your clinicians' Google or Outlook calendars, with your rules for session length, buffers, minimum notice, booking window, working hours, and time zones. You choose how sessions distribute across clinicians: round robin, least busy, or a priority order. Attendees get a calendar invite and a confirmation, and the clinician or coordinator gets a notification with the conversation attached.

How do we know it is producing first sessions and not just chats?

Every conversation carries its source: the page it started on, the campaign and UTM parameters behind it, what the person qualified for, and whether a session was booked. Bookings count only when the booking actually went through. The dashboard breaks conversion down by campaign, source, and page, and every number is clickable down to the individual conversation, so you can read the ones that converted and the ones that did not. Export it and put it in the board deck.

We could build this ourselves. Why Tars?

Standing up a chat box on a model is the easy part, which is why so many practices already have one that nobody uses. The hard part is everything after: deterministic crisis and coverage flows your clinical and legal reviewers will sign, evaluation that shows how the agent behaves on your worst inputs before a patient sees it, one conversation that follows a person across web, email, WhatsApp, and SMS, audit trails and HIPAA posture, attribution into your CRM, and a person on your team taking over inside the same thread. That is what Tars has spent 8+ years on. You go live in weeks, not across several quarters of your own roadmap.

We are a group practice with a handful of clinicians. Is this overkill?

No. The agent is deployed against your clinician directory, your payer list, and your calendars, and it goes live on the pages where intent is highest before it goes anywhere else. Tars runs the implementation with you, including the guardrail setup and the pre-launch testing, so you are not the one supplying the prompt engineering. You keep the controls afterwards: edit the knowledge, refresh your content, adjust the protocol, read the dashboard.

Someone just decided to start therapy. Answer them before they close the tab.