Someone searching "rehab near me" at 1am is at the highest intent they will ever be. Right now they meet a form or an answering service that takes a message. Operators say it every time: the leads that go unanswered are the weekend ones. The Lead Response Management study puts a five-minute first response at roughly nine times the conversion of waiting longer.

The agent picks up instantly on your website, WhatsApp, SMS, or email. It recognizes whether it is talking to the person seeking treatment, a family member, or a referring professional, and runs the right path.

The inquiries you already paid to generate stop dying between 6pm Friday and 9am Monday.

Operators say it on almost every call: until we know someone is a fit, the admissions team should not be spending time on them. Detox, residential, PHP, IOP, and outpatient are not interchangeable, and a mismatched inquiry burns capacity an admissible case needs.

Your intake questions run as a deterministic flow the agent invokes, in your order, nothing improvised: substances and last use, prior treatment, co-occurring conditions, and who is paying. The agent applies your routing rules to suggest which program to evaluate. It does not diagnose, does not recommend treatment, and does not tell anyone what care they need.

Your admissions team keeps every care decision and gets it with the screening already done.

"Will my insurance cover this" is the question that decides whether an inquiry becomes an admit, and it is usually answered a day later by someone else. The agent asks for the carrier, plan, member ID, group number, and the level of care being considered, right there in the conversation.

It hands that packet to your verification team with everything they need to run benefits, and writes it into the record so nobody re-asks. The agent never quotes a benefit determination and never promises coverage.

Your team opens the day with complete verification requests instead of chasing member IDs, and the VOB starts hours earlier.

Every conversation ends as a real record: the screening answers, the level of care under consideration, the desired admit date, the insurance packet, and the channel and campaign that produced it.

The agent writes that into your admissions CRM and EMR by API and webhook, scoped to your stack, puts lead temperature into a real reportable field rather than a free-text note, and alerts the coordinator on duty when the intent is hot.

Marketing can defend the spend with admits by source, and admissions can check bed availability against a request that is already complete.

The first thing an operator does in a demo is try to break it. Name a clinician who does not exist. The fear is precise: that the agent says something your compliance reviewer never approved.

It answers from your programs, policies, and locations, and nothing else. Outside that scope it declines and hands off. It does not diagnose, recommend treatment, or promise outcomes or coverage, which is what your LegitScript and paid-search reviewers look for.

Crisis handling is not left to the model. You define the triggers, the exact wording, the 988 instructions, and the escalation path, and the agent runs that flow as a deterministic tool, identically every time.

The conversion event is not a chat transcript. It is your admissions team on a call with someone who is ready. Every step in between leaks: a link to a scheduling page, a callback request, a form.

The agent holds the time inside the conversation. It reads live availability from your team's own Google Calendar or Outlook Calendar, distributes across coordinators by round robin, least busy, or priority order, and honors your buffers and working hours.

It qualifies while it schedules, carrying forward anything the person already said. When the time is set, a real calendar invite and a confirmation email go out, and the coordinator gets a notification with the whole conversation attached.

Marketing leaders are asked to defend six-figure paid budgets with call counts. The questions they cannot answer are the ones that matter: which campaign produced inquiries that reached an assessment, and where the agent came up empty.

One analytics home answers those. Conversion funnels from visitors through conversations to completed goals. Top lead sources by UTM source, campaign, and landing page, each with a conversion rate. Every number is clickable down to the conversation behind it.

A goal counts only when its trigger genuinely succeeded, so a completion means the record really landed in your CRM. Ranked issues point at where the agent had no answer. Export any view to CSV for the board deck.

Nobody in this industry launches something patient-facing on hope. There is always an internal round first: marketing pokes at it, clinical blesses it, legal wants the worst case. A handful of manual test chats prove almost nothing.

Before launch, Tars runs simulated people against your actual agent. Personas and goals drive multi-turn conversations, generated from your own content and real inquiry transcripts, including crisis language. Every transcript is scored two ways: deterministic code checks, and LLM-judge evaluators that assess whether the question was answered and the right tools were called.

You set the quality threshold the agent has to clear, and deploy once the scores meet it. After launch, real conversations feed the same loop.