Curie Medical Presentation Script

🧠 Free Recall ❓ Quiz 🔍 Cues 🔮 Simulation 🥷 ShadowBox 📐 Encode Cards

Presentation Script — Meridian Session (speak-ready)
Read this aloud once, then make it yours. Short sentences. Pauses do more work than words. Anything in [brackets] is a stage direction, not something you say. Time cues assume a 60-minute session: slides 1–4 ≈ 20 min, demo ≈ 13, slide 5 ≈ 4, buffer ≈ 8, debrief 15.

SLIDE 1 — Problem & OKR alignment (minutes 0–10, discovery)
Goal: they correct the table. Their words become your ammunition. Do not advance until every row is confirmed or edited.
"Thanks for the time — I know an hour with the three of you is expensive, so here's how I'd like to spend it. Ten minutes making sure I understand your business. Then instead of telling you what we'd do, I'm going to show you. And we'll finish with what the first thirty days would look like — so you can say yes or no to something concrete.

Your team shared a lot with us before today. Rather than play it back, we turned it into this table: three goals we believe you're carrying this year, what we think is blocking each one, and what's at stake. Some of this is going to be wrong. I'd rather find out now than after I've shown you a demo built on it. So — [pick up pen] — what did we get wrong?"

[Walk the table only if they're quiet. One breath per row:]
"Row one — keep the members you have. We heard renewals are getting harder, and the sticking point isn't your service. It's that every business review opens with an argument about whose numbers are right.
Row two — you want to sell something new. A premium intelligence tier. The blocker is that today, every answer waits about six weeks in an engineering queue.
Row three — your members signed up for EOM. That means they lose real money when a patient ends up in the hospital for something preventable. And right now, everyone finds out about those events three weeks after they happen, from claims.

One note on the dollar figures — those are our assumptions, not your numbers. Part of the thirty days is validating them together. So challenge them."

[Discovery questions — weave naturally, one per seat. Write their answers ON the slide.]
To the Business Leader: "Walk me through your last hard renewal. What did the member actually say?"
To the Business Leader: "How are your EOM members tracking against their benchmarks — and how do you know?"
To the CTO: "If an auditor asked who saw which member's data last quarter — how would you prove it today?"
To the VP of Engineering: "How many custom member feeds does your team keep alive? And what does on-call look like?"

To the room: "If you could put one new paid capability in front of members next quarter — what would it be?" [Whatever they answer, your demo is the answer. Say so later, in their words.]

Transition: "Okay. So the table holds — with your corrections. Let me show you why all three rows are actually the same problem." [Advance.]

SLIDE 2 — Current state (~2 minutes)
Goal: the architecture is the villain. Never their team.
"This is your world today, as we understand it. On the left, three systems. Your members' practice systems — we'll call it the EHR feed. The lab systems. And claims plus the patient surveys, coming in as exports.
Here's the thing — look at the numbers. The EHR feed says two thousand eight hundred forty-seven patients on active treatment. The labs know about twenty-three hundred. Claims are billing for three thousand one hundred and two. [Pause.] Same network. Three answers.
None of these systems is wrong. They're just never reconciled. So everything between these systems and a decision — [point at the middle] — is a person with a spreadsheet. That's the six-week queue. That's on-call for eighty custom feeds. [Look at VP Eng.] That lives in your backlog.
And when the audit comes — who saw what, across all those member agreements — today that's a manual reconstruction. [Look at CTO.]
And this — [point at contested QBRs] — is why the business review opens with a number fight. The argument is built into the plumbing. [Look at Business Leader.]
One more thing. EOM just made this expensive. Your members now lose money on hospital events this architecture reports three weeks too late."
Transition: "So the fix isn't a better spreadsheet. It's changing what sits in the middle."

SLIDE 3 — The lakehouse (~3 minutes)
Goal: the CTO hears their own strategy. One idea: one platform, one governance layer.
"Here's our answer, and I'll keep it simple. You don't have an AI problem. You have a foundation problem. Fix the foundation, and the AI — and the forecasting, and the self-service — get cheap.
This is the lakehouse. One platform. Along the top — the four kinds of work: getting data in, analytics, machine learning, and generative AI. At the bottom — all your data, in one open format. You're never locked in; if we ever fail you, your data is still yours, in a format anything can read.
And this band in the middle is the part I'd ask you to remember. One governance layer — Unity Catalog — that covers everything above it. Who can see what. Where every number came from. A full audit trail. Once, for the data, the models, and the AI.
The alternative is what most people do: stitch four or five tools together. A warehouse here, an ML platform there, an AI vendor on top. It can work. But the audit trail breaks at every seam — and in your world, with patient data across eighty member agreements, the seams are the risk.
Quick proof this isn't theory. A healthcare improvement company that serves two-thirds of U.S. providers runs this exact pattern. Their natural-language analytics went to production in three days. They measure their analysts at roughly ten times faster. And their data leader described turning on this governance layer as — his words — like flipping a light switch.
Last thing, for engineering: this is managed. Nobody's racking servers or babysitting clusters. The demo I'm about to show you — one person built it in an afternoon, and there wasn't a single infrastructure ticket."
Transition: "So what does that look like for Meridian specifically?" [Advance.]

SLIDE 4 — Future state + the handoff (~2 minutes)
Goal: orient in 90 seconds, then hand off to the demo — backwards.
"Left to right. Your members' data comes in — same three feeds, nothing gets ripped out, the sources stay where they are. Inside the platform, three things happen. Unify — three counts become one truth: one record per patient. Predict — a model watches every patient on treatment and flags who's likely to end up in the hospital in the next thirty days. Activate — the system drafts the response, using your own playbooks, with a citation on every line. And on the right, your members see it — in their reviews, in their workflow. Underneath all of it, that same governance layer. And the loop closes on its own: outcomes come back through claims, the model retrains, and you walk into every business review with proof.
Now — here's how I want to show it to you. Not left to right. Right to left. I'm going to start at the end — with one patient whose day actually changed — and then peel back the layers, one at a time, until you can see exactly why you'd trust it. Her name is Maria."
[Stop talking. Open the demo. Nothing on screen but the product until the demo ends. Demo script is a separate document.]
Re-entry after the demo: "That's the future state — you just watched it run right to left. Quick honesty before we talk next steps: what you saw runs on synthetic data — your real pilot feeds come in week two. The playbooks it cited are ones we seeded — your documents come in week four. And the model would be proven against your own last six months in week three. Which brings me to the ask." [Advance to slide 5.]

SLIDE 5 — Next steps (~4 minutes)
Goal: convert agreement into motion. Negotiate the success criteria — don't present them.
"Thirty days. On your data. Not eighteen months.
Week one, we do governance first — the data agreements, the access rules, and we pick two or three pilot members with you. Week two, one truth: your real feeds, one record per patient, and self-service questions live for your advisors. Week three, the model — and here's the part I like best: we test it against your own last six months. Events that already happened. So we prove it predicts, without waiting months for new events. Week four, the copilot runs on your playbooks, your clinical governance committee reviews the citations, and we put a readout in your hands you could take straight into a member business review.
Now the important part. I don't want to define success for you. If we're standing here on day thirty — which two or three things would make this real? Here's the menu: a patient count all three of you sign off on. The model clearing a bar we agree on, against your own history. Ten real member questions answered the same day they're asked. Or your pilot members' teams actually using it. [Pick up the pen.] Which two or three?
[Negotiate. Write them down. Aim for one revenue, one governance, one delivery — so each of you owns one.]
What we need from you is small. [To the Business Leader:] Sponsor it, and pick the pilot members. [To the CTO:] Your security review in week one, and sign-off on the governance setup. [To the VP of Engineering:] Two engineers, part-time. That's the whole ask.
And on the timeline — I'll say it plainly: the company I mentioned, serving two-thirds of U.S. providers, put this pattern into production in three days. Thirty days is conservative.
Day thirty, you get one of two things. Proof — against criteria you just picked. Or a cheap no. Either way, you'll know in a month what most organizations debate for a year."
[If asked "what happens after day 30":] "If the criteria hit, we scale from the pilot members to the network on the same foundation — nothing gets rebuilt. If they miss, you spent thirty days and learned it cheaply. That's the point of doing it this way."

OBJECTION NOTES — plain language, three beats each
Pattern for every objection: agree with the concern → answer in one or two plain sentences, pointing at something they saw → tie back to their goal → stop talking. Silence after an answer reads as confidence. Never answer a question they didn't ask.
COST — sounds like: "What's this going to cost us?" / "Platforms like this get expensive."
"Fair — and I'd push back on any big number too, so we designed this to not need one. The thirty days runs on pilot-scale, pay-as-you-go compute — think a small project budget, not a platform commitment. You only scale it after it clears the success criteria you just picked. And I'd put the cost next to the other number on the table: about six million of ARR at risk, and members losing real money on preventable events at roughly fifteen thousand each. The expensive option is the spreadsheet." [Stop.]
EFFORT — sounds like: "My team has no capacity." / "This feels like a big lift."
"That was true five years ago — it isn't now. Everything you saw is managed: nobody stands up servers, nothing gets ripped out, your sources stay exactly where they are. The demo was built by one person in an afternoon with zero infrastructure tickets. The ask on your side is two engineers, part-time, for thirty days. If it takes more than that, that's a signal — and we'd want to know it by day thirty, not month eighteen." [Stop.]
COMPLEXITY — sounds like: "This looks like a lot of moving parts." / "We've been burned by big data projects."
"I'd flip that. The complexity is what you have today — eighty custom feeds, three systems that don't agree, and a person with a spreadsheet in the middle of every decision. This replaces five tools with one platform and one governance layer. And notice what you watched: one trail, from a source row to a model to a cited action, with no seams. That trail is only simple because it's one system. Five vendors can't draw that line." [Point at the citation moment if it's still up.]
TRUST — two flavors. Answer the one they asked.
Trust the data — sounds like: "How do I know the numbers are right this time?" "You don't take my word for it — the system shows its work. You saw the quality gate: forty-one records quarantined, with the reasons listed — including the duplicate IDs that cause your number fights today. Every number traces to its source row; you clicked that lineage yourself. And week two of the pilot ends with a patient count all three of you sign off on. That signed number is the whole point." [Stop.]
Trust the AI — sounds like: "What if it makes something up?" / "I can't put AI in front of patients." "Neither can I — so it's designed not to decide anything. Two rules. One: it only writes from your own record and your own playbooks, and every line carries a citation — you watched me trace one from the plan back to the exact source row. If it can't cite it, it doesn't say it. Two: it never touches a clinical decision. It schedules, it calls, it arranges rides, it flags things for the care team. The judgment stays with Lakeside's clinicians — the system just makes sure they act today instead of in three weeks. And your clinical governance committee reviews the citations in week four, before any member sees it." [Stop.]
Quick one-liners (keep in your pocket)
PHI / security: "Every member feed is governed by its own agreement. Members see only their own patients, identified. Meridian sees the de-identified network. Every access is logged. And that's enforced in the platform — you saw the filtered view — not in a policy document."
Lock-in: "Open formats, bottom to top. If we fail you, your data leaves with you, readable by anything."
"Why not build it ourselves?": "You could — the question is whether eighty feeds and a governance layer is the best use of your engineers, versus the premium tier your members would pay for. We handle the plumbing; your team builds what's sellable."
A question you can't answer: "I'd be guessing, and with your business I won't. Let me bring the right specialist into a working session this week — what I can show you is where that answer would live." [Team-selling reads senior, not weak.]

Vendor-neutral Required Capabilities (spoken over the current-state slide)
Use these as the bridge out of slide 2 — after the pain walk, before Databricks is ever named. The setup line: "So set vendors aside for a second. Whatever sits in the middle of this picture has to be able to do six things. Tell me if any of these look wrong — or if I'm missing one." Getting nods here means slide 3 lands as "the thing you just asked for," which is the entire trick. Keep them spoken, not on the canvas — the slide stays clean, and requirements you say invite dialogue in a way bullets don't.
One patient, one record — the ability to bring the practice, lab, and claims feeds together into a single longitudinal record per patient, across all eighty member organizations, without ripping out any source system.
Provable access control — the ability to enforce, in the platform itself, that each member sees only its own patients while Meridian sees the de-identified network — with an audit trail you could hand a regulator, not a policy PDF.
See it coming — the ability to predict which patients are heading toward a preventable hospital visit before it happens, ranked so scarce nursing time goes to the right people first.
Your playbooks, put to work — the ability to turn Meridian's own protocols into patient-specific, fully-cited action plans — where the system drafts and humans decide, every time.
Answers without a queue — the ability for a non-technical advisor or practice administrator to ask a question in plain English and get a governed answer in the same meeting.
Proof on a loop — the ability for real outcomes to flow back in, retrain the models, and land as evidence at every QBR — with the trail from source row to answer unbroken across every one of those five capabilities.

Unstarred

Dunk (exclude from study)

Curie Medical Presentation Script

Review Questions

Todos

Sensemaker