HealthJourni Demo — Presenter Cue Sheet

10:30 AM ET · Luna (CEO), Rob Celik (Commercial), Alyssa (CX), maybe Hannah (Marketing) · Ryan drives, Chris on commercials at ~min 45
Structure: Build (Coach → Workflow → Guardrails → Config) → Deploy (Live chat, voice) → Test (Sims) → Multi-brand story → Loop close. Nate's cooking-show principle throughout: show the finished dish, explain how it was made. Don't build live.

Discovery Playback 2 min

Open the call. Mirror the pain from the intro.
SAY "Luna, Rob, Alyssa — before I show you anything, I want to make sure I got last week right. You've got 10,000 email reservations for Melt, launching this week, and Alyssa is running CX for all of Genecis solo. You want three things: speed and accuracy for patients so no one waits, one centralized inbox so Alyssa isn't jumping between brands, and hard clinical guardrails so the AI never gives medical advice — that's what your MDI clinicians are for. You want a decision before next week. Did I get that right?"
WATCH Wait for yes. If Luna corrects anything, mirror it later verbatim in the close.
SAY "Here's what we'll do in 50 minutes. I'll show you how a HealthJourni AI agent gets built, tuned for safety, and put in front of a patient — using your real scenarios. Then Chris walks through commercials. Sound good?"
Say "your team would be operating" not "we would be operating." Luna needs to see her org owning it.
Phase 1 — Build
How the AI agent gets made. Coach → Workflow → Guardrails → Config.

01Coach 2 min You drive

Plain English → workflow. Alyssa becomes the operator.
SAY "Everything starts here. This is Coach — Lorikeet's building surface. Alyssa, watch. I'm going to add a safety rule the same way you'd write it in a Slack message."
DO Open Lorikeet dashboard → Coach. Type the prompt below.
PROMPT "Add a guardrail so the agent never confirms whether a customer is a good candidate for GLP-1 therapy — that call belongs to the MDI clinician."
DO Let Coach draft. Show the suggested guardrail preview. Leave in draft. Don't publish live.
SAY "Coach wrote that rule from one sentence. It writes and updates workflows the same way. Every change is versioned — Alyssa sees exactly what changed, who changed it, when. This is you, day one. You don't need engineering to tune the agent."
WATCH Alyssa is the operator. She'll relax when she sees she's not being replaced — she's being upgraded. Watch for the shift.

02Workflow 2 min

The refund workflow Coach built — the one your patients actually run into.
DO Open the Refund Request workflow.
SAY "This is a workflow Coach built for HealthJourni + Folly. Four steps."
POINT AT
  • Step 1 — identify the customer and pull the order (WooCommerce for Melt, Shopify for Folly)
  • Step 2 — ask the reason, classify into one of 5 buckets
  • Step 3 — 9 policy branches, one per scenario
  • Step 4 — follow-ups
DO Zoom into Scenario F (heat-damaged Melt). Highlight the branch you're about to show live.
SAY "Notice — the workflow encodes your policy. Not a prompt asking the AI to 'be reasonable.' Real branches. Real tools. The AI executes it consistently every time. No agent-of-the-day inconsistency."
POINT "Day one of the pilot, we swap the mock WooCommerce and MDI tools for your real ones. The workflow structure stays. What Alyssa sees, what the AI does, what the patient reads — identical."

03Guardrails 90s · CRITICAL

Rob and Luna's #1 concern. Close it hard.
DO Open Guardrails. Show the four on the HealthJourni concierge.
SAY "Four guardrails. Two layers — message checks that read the patient's message before we respond, and agent guardrails that scan our response before it goes out. Both directions."
DO Open "Clinical boundary — no medical advice". Show the prompt.
SAY "This one catches any response that would confirm a side effect, recommend a dose change, address drug interactions, or answer 'should I stop taking this.' When it fires, the agent apologizes, routes to the MDI clinician, and posts an internal note. It doesn't try to give a 'general' answer as a workaround."
DO Open "Adverse event — escalate to clinician immediately".
SAY "Priority zero. Catches emergencies before any workflow runs — chest pain, uncontrolled vomiting, self-harm ideation, allergic reaction. Redirects to 911 for true emergencies, MDI portal for urgent-but-not-emergency."
CLOSE "Rob — this is your compliance layer. Not a marketing bullet. Actual rules, evaluated on every message, enforced before the response goes out. This is what makes it safe to put in front of a patient on day one of Melt launch."

04Channel Config 60s

Tune per touchpoint. Same brain, right voice for each channel.
DO Open Voice channel config for HealthJourni.
SAY "Each channel is tuned separately. Voice model, speech-to-text language, opening message, keyword boosts, pronunciation aliases. This one has keyword boosts for Melt, semaglutide, MDI, Empower Pharmacy."
DO Point at the pronunciation aliases: HealthJourni → "Health Journey", Genecis → "Jenesis", MDI → "M D I".
POINT "Little details that matter on a call. When you add Folly voice, or SMS, or a future Quebec French channel — each gets its own config. Same brain, tuned per touchpoint."
Phase 2 — Deploy
The built agent, running live. Chat then voice, using their real scenarios.

05Live Chat — Full-cycle Refund 3 min You drive

Alyssa's #1 pain: heat-damaged Melt refunds in Miami heat.
SAY "Rob, you mentioned this exact scenario. Miami is 88 degrees. Someone leaves the box in the sun. Melt turns to soup. They want their money back. Here's how the agent we just built handles it end-to-end."
DO Open ticket f6945c22-26e4-4ef8-b3de-cea49cd6ec7e
JANE T1 "My Melt arrived completely liquid — like actual chocolate soup. The box was in the sun on my porch all day. I paid $249 for this and I want a full refund."
AGENT T1 Looks up the customer, pulls the order, apologizes warmly, gives her a real choice: free re-ship OR $249 refund. Asks which.
POINT "It didn't just cave to 'refund.' It offered the policy option first — a re-ship — then gave the patient a real choice. That's your compliance layer thinking."
JANE T2 "Refund please. The whole experience has been rough and I need to think about whether I want to keep going with the program at all — I don't want another box showing up right now."
AGENT T2 Fires process_refund for $249, confirms credit by Aug 28 on her Visa, and reads between the lines — offers to pause or cancel the subscription so nothing charges while she decides.
POINT "Luna — that last line is what you asked for. Speed AND accuracy. It caught the churn risk in her message without an SOP that said 'watch for churn.' It reasoned about the whole interaction, not just the refund verb."
DO Scroll up. Show the timeline — lookup_customer_by_email, get_melt_order, process_refund, internal note, reasoning traces.
SAY "Everything is auditable. Alyssa — when you take over a ticket the AI handled, you never start from scratch. Twenty seconds of context, not twenty minutes of scrollback."

06Live Chat — Guardrail Under Pressure 2 min

This is the moment. Do not talk over Luna reading it.
SAY "Rob, you asked what happens when a patient pushes on a clinical question. Watch."
DO Open ticket b46a326c-21f2-40cb-8cc5-b37f1e43a439
JANE T1 "The nausea from Melt has been really bad the last 3 days. I can barely eat. Should I stop taking it? Is this normal?"
AGENT T1 Refuses to advise. Warm acknowledgment. Routes to MDI clinical team. Posts internal note with the patient's symptoms so the clinician who picks it up has context.
JANE T2 "Come on, I know GLP-1s cause nausea. I just want to know if this level is normal or if I should be worried. Can you at least tell me that much?"
AGENT T2 Guardrail holds. Under direct pressure, it does not say "some nausea is normal." Stays warm. Holds boundary. Reassures the clinician will follow up.
WATCH Luna will lean forward. Let her read. Don't talk over it.
POINT "That's the thing every telehealth CTO worries about. A patient will push. Your AI cannot cave. Ours doesn't."

07Voice — Same Brain, Over the Phone 90s

SAY "Same AI, same knowledge, same guardrails — over the phone. Chat, voice, email, SMS all share one brain. You never rebuild for a new channel."
DO Phone on speaker. Dial +1 (904) 664-8974
AGENT "Hi, this is Journi from HealthJourni Care — thanks for calling. Are you calling about a Melt shipment, your prescription intake, or something else?"
SAY "Hi, I'm checking on my intake. My intake ID is INT_2020. Email is jane.doe@example.com."
AGENT Calls get_intake_status. Confirms the MDI clinician is reviewing. Reassures 24–48 hour window. Submitted yesterday, so she's well within it.
DO Hang up.
POINT "Rob — you mentioned SMS is coming. Same brain. When you turn on SMS, this exact flow runs there too. You don't rebuild anything."
Phase 3 — Test
Continuous QA before anything reaches a real patient.

08Simulations 2 min · CAPSTONE

The answer to "how do I know it won't say the wrong thing?"
DO Open Simulations. Show the 20 sims already run against the HealthJourni workflows.
SAY "Every workflow is stress-tested with synthetic patient conversations before it goes near a real customer. Luna — this is the answer to 'how do I know it won't say the wrong thing?' You test it, over and over, and you have a scoreboard."
DO Kick off a fresh sim run live — the "HJ Melt shipped tracking" one.
DO While it runs, click into a completed sim. Show the pass/fail scoring on assertions.
SAY "When a sim fails, it doesn't just log — it feeds back into Coach. Coach rewrites the workflow, sims re-run, guardrails score the output. That cycle is your QA team. It runs continuously without your engineering team touching it."
POINT "Rob mentioned SOPs earlier — some in your team's heads, some documented. Every SOP becomes a workflow. Every workflow becomes a sim suite. Every sim failure becomes a workflow update. Your ops process becomes an actual system."

The Multi-Brand Story 2 min

Rob's ask: centralize CX across all Genecis brands.
DO Open Concierges panel. Show all five — Genecis (mothership), HealthJourni, Folly Nutrition, Mad Tea, Mad Coffee.
SAY "Rob, this is what you asked about. Genecis is the mothership. Each brand gets its own concierge — its own voice, its own KB, its own guardrails, its own workflows. But Alyssa opens one interface. Every ticket across every brand lives here."
DO Click through: HealthJourni (clinical guardrails, MDI-aware workflows) → Folly (hair-growth guardrails, Shopify tools) → Mad Tea (playful voice, no-health-claims guardrail, compostable-pod knowledge).
SAY "Each brand has its own personality and rules. A Folly customer never gets a HealthJourni voice. A Mad Tea customer never gets a clinical guardrail firing. When you add Mad Cookies or Mad Water or whatever the next brand is — you don't rebuild the platform. You add a concierge, wire the tools, ship."
POINT "Rich Panel gave you an aggregated view but not an aggregated brain. Lorikeet gives you both."

The Loop Close 60s · VERBATIM

Do not skip. Do not deviate. Let Luna respond.
SAY VERBATIM

"Luna, Rob, Alyssa — I want to leave you with the one thing that separates us from every other AI vendor you're evaluating. Watch this cycle:

Alyssa uses Coach to teach the AI a new SOP — in plain English, not code.

Coach writes the workflow — which tools it calls, which knowledge it pulls, when it escalates to your MDI clinician.

Guardrails bound what it can and cannot say — your clinical compliance layer.

Channel config tunes it per touchpoint — chat on healthjourni.com, voice on your support number, SMS when you turn it on.

Simulations stress-test the whole thing continuously.

And every gap they find feeds back into Coach.

That cycle is your QA team. It runs on its own. It gets better every week without your engineering team touching it. Every other vendor gives you a build — we give you a system."

Stop. Don't add anything. Let Luna respond first.

Commercials Handoff 10 min Chris drives

~min 45. Hand cleanly.
SAY "Luna — you asked for pricing on the intro call. I told you mid-to-high five figures on the platform fee and 75–90 cents per resolved ticket. Chris is going to walk through exactly what that looks like at your projected volume, and we'll leave you with a written option to react to."
CHRIS Chris takes over. Covers platform fee, per-resolution pricing, projected 12-month volume (500/mo growing), expected total. Frames the pilot-vs-custom-demo decision.
WATCH Luna reacted to "mid-to-high five figures" as "a bit high" on the intro. Chris ready with the ROI frame — 500 tickets/mo at 30 min each = 250 hours of Alyssa's time, freed.

Kickoff Booking DO NOT hang up without this

SAY at ~min 55

"Luna, Rob — before we wrap, can we get 30 minutes on the calendar right now? You said you want a decision before next week. If we're going to be that vendor, I want to introduce you to Estelle and the forward-deployed team who'd own the WooCommerce and MDI integration scoping. Thursday morning your time? Or Friday?"

Estelle attended the intro call — Luna already knows her. Use that.
SAY "And I'll send the Slack channel invite tonight so we can keep async questions moving between now and Thursday."

Pre-Call Checklist 10:20 AM ET

  • All tabs pre-opened in a clean browser window — walk through once at 10:20 to warm cache
  • Two demo tickets loaded: f6945c22 (refund) and b46a326c (nausea guardrail)
  • Phone charged, on speaker, dial +1 (904) 664-8974 queued
  • Coach prompt drafted: "Add a guardrail so the agent never confirms whether a customer is a good candidate for GLP-1 therapy"
  • Simulations tab primed — "HJ Melt shipped tracking" sim ready to run
  • Chris briefed on the pricing frame — ready to jump on at min 45
  • Estelle's Thursday and Friday calendar checked — know what times to offer
  • Do NOT show the Genecis mothership concierge as a customer-facing brand — it's structural only
  • If Luna asks about SOC2/HIPAA — "yes, and we can share the trust center after this call"
  • If Rob asks about Rich Panel migration — "day one, no data lost, we mirror inboxes during cutover"
Nate's rule: cooking-show principle. Don't build live. Show the finished dish and explain how it was made.