Every one of these is the same conviction in a different place: the evidence should reach the decision. Healthcare buyers don't have budgets for AI infrastructure — they have budgets for outcomes, and an outcome is what happens when the evidence arrives in time. Clinician time recovered, denials reversed, care gaps closed, patients engaged. The architecture is what makes them work; the outcomes are what we sell.
Healthcare finance allocates cost against whatever it can measure — square footage, bed count, salary dollars. None of them is what actually consumed the resource. Tracer allocates against clinical activity instead: prescriptions administered, nursing visits, patient days. Every allocated dollar traces back to the driver that placed it, and to the source query that produced the driver.
of allocated cost resolves to a driver with verified source coverage. A new cost centre no driver reaches blocks the run rather than being estimated; every other coverage gap is a finding somebody has to clear by name.
Claims work is split across queues that each know only their own corner — an appeal here, a resubmission there, a callback nobody chased. Every queue looks busy, so the work that gets picked is whatever is loudest, not whatever is costliest. Claims reads the documents a claim generates and ranks the open work across every queue on two axes at once: what a delay costs, and how long there is left to act. One record, read at the altitude of a single claim and at the altitude of the whole book.
record serving both the single claim and the whole book of work
A population query returns what the projection was built to carry. Patients whose qualifying evidence sat in a note, a scanned report, or a field that never mapped are absent — not excluded, just never carried across. Insight queries the source beneath the projection, so a new question doesn't mean a new pipeline.
Prior auth bots that work from claims data alone miss the clinical context that justifies the request. Auth works from the visit notes, the imaging reports, the history of failed alternatives — for higher first-pass approvals, stronger appeals, and better-justified denials when denials are correct.
Most patient-facing tools don't know who the patient is. They handle generic interactions and route everything else to a call center. Connect is designed to work from the patient's full record — visits, medications, recent labs, language preference, prior conversations — and give them a navigator that understands their care, deployed under your brand.
Connect does not give medical advice. It answers from the patient's own record and helps them navigate their care. Anything that calls for clinical judgement is directed to their medical provider.
Clinicians spend one to two hours a day on documentation. Existing scribing tools transcribe but don't reason. Assist reads the entire longitudinal record — prior conversations, social determinants, family history — and surface what matters at the moment of decision, while drafting documentation that needs only review.
Illustrative — a worked example, not live output.
Every Lumen solution runs on the same foundation: a preserved record that sits beneath the models rather than behind them. Projections are built over it and stay re-derivable, because the source they were drawn from was never discarded. Different surfaces, one source of truth within the organization it belongs to. Ingest, retrieval, and reasoning are built once and shared, so the second solution starts from a foundation that already exists rather than from a new integration. They are live to each other too: what one solution establishes reaches the others while the work is still running, not on the next batch. And the solutions on it are not only ours: partners build and sell their own, and customers build their own for themselves.
Curated views are how access limits and query performance get expressed. Continuity doesn't remove them — it puts something underneath them, so a view can be narrowed, rebuilt, or replaced without another migration.
Continuity within boundaries. The record arrives whole and stays inside the organization it belongs to; what is bounded is reach into it — legal and privacy obligations first, then role. What every customer shares is the platform. None of them shares the data.
Read the platform thesis →We work best with health systems, payers, and digital health companies that have a specific outcome they want to move and a willingness to do the integration work that makes AI useful. Tell us where you're stuck.