Platform · Patient Intelligence
Ask about your population. Get the patients.
A question in plain language becomes a defined cohort, with every patient carrying the values that put them on the list. Not a report somebody then has to work.
You see what it understood before anything runs.
Including the part it could not apply. A system that tells you what it did not understand is one you can trust with the part it did.
Population question
Asked
“Diabetic patients 18 to 75 with no A1c result in the last year”
Understood as — confirm before running
- Condition · Type 2 diabetes · diagnosed
- Age · 18 to 75
- Lab · HbA1c · not resulted within 12 months
Could not apply: “recently” — no timeframe was given, so nothing was assumed
AI interprets the question. Deterministic code runs it.
You see exactly what will run before it runs.
- Validated criteria
- No model-written queries
- No assumed thresholds
A quality score is a number. We give you the people behind it.
Quality
Colorectal cancer screening
- Open
517 patients overdue
Resolved to people, not rows — the same patient across five facilities counts once.
- Enrolled
Care gap program
The list becomes the roster the agents work, without an export in between.
- Worked
Explained · offered · objection captured
One gap per contact, in the patient's own terms, with a real reason recorded when they decline.
- Closed
Booked and written back
The appointment lands in your schedule and the outcome lands in your record, so the list reflects work done rather than attempts made.
Whichever scorecard you answer to, the work is the same: find who is open, reach them, close it, and prove it closed.
Three kinds of risk, and none of them is a black box.
Risk profile
Clinical risk
HighLikelihood of an adverse outcome
What produced it
- Four chronic conditions on the active problem list
- Inpatient stay within the last 90 days
- Two medication changes since the last visit
Payer risk
MediumExpected cost against the contract
What produced it
- Utilization above the attributed panel's median
- No primary care contact in the last six months
Documentation risk
HighComplexity carried but not captured
What produced it
- Three conditions documented last year, not re-documented this year
- Two conditions appear in notes but never on the problem list
A score nobody can interrogate is a number your clinicians will learn to ignore.
For technical reviewers: how Patient Intelligence works in depth
The model never writes a query against your database.
It fills in a form. That form is the only language it is allowed to speak, and our code — not the model — turns the filled-in form into a query.
- 01
One output channel
The question reaches the model with exactly one way to answer: a fixed set of fields. It cannot hand back a query, because no channel for one exists.
- 02
A closed vocabulary
Unknown fields are rejected outright. Conditions, labs and procedures come from fixed lists, and every number is bounded. Anything outside that fails before it runs.
- 03
Nothing is assumed
Ask for patients “over” a value without saying over what, and the system tells you it could not apply that part. It never invents the threshold.
- 04
You confirm what will run
The plain-English restatement is rebuilt from the validated criteria rather than from the model's own words. What you read is exactly what executes.
- 05
Only our code builds the query
Named regions are resolved by code, not guessed. Identity resolution runs before any count. No path anywhere executes a query the model wrote.
This is not only our safety argument. A health-system CIO specified the same design independently, before seeing ours: have the model emit a validated specification, never code that runs against live clinical data.
What keeps a number honest.
One definition, everywhere
The number on the card and the patients behind the click come from the same place. A figure you take into a payer meeting is the figure your care team is working.
The gap closes where it is found
A measure that ends in a dashboard has moved the problem. A measure that ends in a booked appointment has solved it.
Movement you can attribute
Every closure carries the contact that produced it. When a rate moves, you can say what moved it.
What produces a risk lens.
A weighted model over clinical indicators, psychosocial and social-needs factors, and area-level vulnerability — every one of them an input you can see, and a weighting your clinical leadership can read and challenge.
Clinical indicators
Condition count and combination, utilization, medication complexity, results outside the patient's own range, and what has changed since the last contact.
Social and psychosocial needs
Housing, food, transport, medication access and financial stability — captured on contacts, coded to the standard diagnosis codes for social determinants, and weighted alongside the clinical picture rather than filed separately from it.
Where the patient lives
Area-level vulnerability by locality, so two patients with identical charts and very different circumstances do not receive identical follow-up.
The third lens is the one finance cares about.
In a shared-savings or risk contract, complexity that is not documented is complexity you are not paid for. A patient whose record no longer reflects how sick they are makes the whole panel look healthier and cheaper than it is — and the correction is not a coding push, it is surfacing what your own record already contains, per patient, and letting a clinician decide.
Open one patient, and every line shows its work.
The review items on a patient brief are generated by explicit rules over that person's own records, never by a model summarizing them. Each one carries the record it came from.
A condition that dropped out of the record
Documented before, not re-documented in the current window — across every source you have connected, not only the chart in front of you.
An open care gap
The patient matches a measure's criteria, run through the identical query a cohort click runs. The card and the list can never disagree.
Evidence that contradicts the record
A lab result with no matching diagnosis. A risk flag with nothing documented behind it. A condition recorded in one source when the person has records in several.
It never tells a clinician what to find.
A review item says: check whether this condition remains active, clinically relevant, and appropriate to address today. It never says document this because it raises a score. A tool that surfaced only the conditions affecting reimbursement would be a revenue tool wearing a documentation-quality label, and its omissions would be as legible as its inclusions.
The brief sits on top of the patient profile.
The cohort becomes the roster.
This is the step teams recognize immediately, because it is the one that normally does not exist. Finding the patients and reaching them are usually two systems with a spreadsheet in between.
- 01
Ask the question, and confirm the criteria
- 02
Review the patients, and why each one matched
- 03
Enroll the cohort into a care program
- 04
The program's contacts run on their own cadence
Enrolling never starts a call or a text. A roster is not consent: an enrolled patient carries no consent, no call window and no workflow until your team sets them, and the scheduler's own gates still apply unchanged.
Population health
A question answered in the room instead of a ticket to an analyst — and the answer is patients, with the reason each one is on the list.
Risk adjustment and finance
Complexity your record already contains and no longer shows. Surfaced per patient, across every source, without telling a clinician what to write.
Quality
Gaps that end in a booked appointment rather than a dashboard, with the contact that closed each one attached.
One system, four views
The platform overviewBring a question your current tools cannot answer.
The fastest way to judge this is to ask it about a population your team already argues about internally, and see whether the answer survives the argument.
Talk to our team