Trust · Clinical design
The conversation adapts. The boundary does not.
Every LifeWatch program is designed, evaluated and governed before a patient hears it. The AI chooses its next question from what the patient says. It never chooses what it may ask, what it may resolve on its own, or what must reach a clinician.
Four things happen to a program before it reaches a patient.
This is how the clinical intelligence behind every contact is built and kept honest. It is a working method, not a marketing frame — the same four stages apply to the first program and to a one-line change in the tenth.
- 01
Research and evidence
What informs a program
Each program starts from established clinical guidance for the condition and from behavior-change science that is public and long-studied: what to follow, what to teach, which changes matter, and what the AI must never handle itself. The clinical content is written by clinicians, not generated or bought in.
- 02
Program design
What is designed into it
A multi-month curriculum, not a script: what each contact covers and in what order, the questions it may explore, the actions it may complete, how follow-up continues, and — in the same document — what must reach a person. The escalation rules live inside the program they belong to, so the two cannot drift apart.
- 03
Evaluation
How it is tested before it runs
The program's escalation behavior is scored against a fixed rubric using authored cases: the serious symptom a patient plays down, the ordinary complaint that must not be treated as an emergency, the caregiver answering the phone. Missed concern and unnecessary noise are both counted, and a change that trades one for the other does not ship.
- 04
Governance
How it is controlled once it runs
Every program is versioned. It is certified by our clinical team before it can reach a real patient — a gate enforced in the database, not in a process document — and your organization reads the whole program before go-live. A change re-earns certification, and you are told what changed before it reaches your patients.
The patient selects the branch. A clinician wrote every one of them.
This is what adapting inside a boundary means in practice. The contact arrives with an objective, listens, and moves — and each move it can make was authored and certified before the first call.
Heart failure program · month 3 · contact 2
This contact’s objective
Confirm Ruth knows what to do when her morning weight jumps, and check how the diuretic routine is holding.
What she said
“I skip the water pill on church days. I’m not going to be getting up out of the pew every twenty minutes.”
The branches the clinician wrote
She knows the routine
Advance. The next contact opens the sodium and fluid module, and this one ends early.
She is skipping doses by choice
TakenStay here. Work through the reason in her own words, note the pattern for her care team, and come back to it in three days.
Weight is up or symptoms are present
Stop teaching. Escalate to her care team with the transcript and the sentence that triggered it.
What Ruth says selects between branches a clinician wrote and certified. It cannot create a fourth.
Three days later the program comes back to the same point, because a curriculum does not drop a thing just because the first answer was no. Whether the routine held is what the next contact opens with.
For clinical reviewers: what is inside a program, and how a change is governed
What is inside one.
The method below is public, long-established practice. The programs built on it are not published — they are the clinical content our team authors, and the reason a LifeWatch contact is a conversation rather than a script.
Behavior change, not reminders
Programs are built on the transtheoretical model: people move through stages, from not yet considering a change through to maintaining one. Slipping back is part of the process, not a failure. Where a patient is in that arc changes what the next conversation should be.
Guiding rather than instructing
The conversational style is motivational interviewing: draw out the patient's own reasons for doing something rather than listing yours. Telling people what to do is the approach that has already not worked for them.
Written to be understood
Content is authored at an eighth-grade reading level, the standard in patient-facing medicine, and checked for understanding rather than assumed.
A curriculum, not a script
Each condition is a multi-month program of defined focuses and directives, with branch logic on what the patient tells you. Month two is not month one repeated.
One condition, done properly, then replicated
The construct is fixed and the clinical content swaps per diagnosis — so the second condition inherits everything the first one proved.
Certified before a patient hears it
A program is certified by our clinical team before it can run with real patients, and that gate is a database constraint rather than a step in a process document. Your organization reviews the certified program before go-live.
A changed program is a new program. It re-earns its certification.
Programs are versioned. Editing what one asks, or where its boundary sits, sends it back through evaluation and certification. Until then, the certified version keeps running, and your program owner is told what changed before it reaches a patient.
Program change · Heart failure follow-up
v2.4 · certified, running→v2.5 · in certification
- Authored
- Evaluated
- Certified · now
- Released
What certification is waiting on
- What changed
- Week-two call now asks whether a discharge medication was stopped at home. Two thresholds tightened.
- Authored by
- LifeWatch clinical team
- Evaluation
- Full escalation set re-run against the fixed rubric
- Your review
- Change summary sent to your program owner before release
- Until certified
- v2.4 keeps running unchanged
An uncertified program version cannot contact a patient. The gate is enforced in the database, not left to process.
Certified before it can run
An uncertified program version cannot attach to a real patient. The block is a database constraint, so it holds on a bad day exactly as it does on a good one.
Evaluated on every change
The full escalation set runs again against the fixed rubric before release. Nothing ships on the strength of having passed once.
Reviewable, before and after
Your organization reads the whole program before go-live — every question and every criterion. Every contact keeps its transcript afterwards.
Most of what goes wrong between visits is not a clinical failure.
The medicine was right. The patient did not fill the prescription, or filled it and stopped, or never understood what the dose change was for, or could not get to the appointment and did not want to say so.
A reminder does not fix any of that, which is why a program is a curriculum rather than a schedule of prompts. It goes at the thing that is actually in the way, checks whether the answer landed, and comes back to it later if it did not.
That is also why the same content works by phone and by text: the method is the conversation, not the channel.
Ask us how a program is designed.
If your clinical team wants to see the shape of one before deciding anything, that is a reasonable request and we will walk through it with them.
Talk to our team