Sector
Healthcare
Predict the expensive event before it happens, from data the health system already has — and explain why, so a clinician can act on it. That second half is what most of this work is actually about.
§ 01 — how we approach it
Method before numbers.
Every healthcare engagement we have run has the same shape, and it is the shape rather than any individual accuracy figure that determines whether the work gets used.
- Risk stratification on data you already hold
- Identifying who is at risk from records a health system already collects — survey instruments, demographics, physical examination, claims — rather than from a new collection programme.
- Scoring without laboratory data
- A distinct and underrated mode: models that run on survey, demographic and physical-exam inputs alone. It means population-level screening without drawing blood, which changes what is operationally possible.
- Explainability as a contractual deliverable
- Not just a score. Which factors drive it, ranked, and what happens to the risk when one of them moves. In our health work this was written into the milestones, not offered as a courtesy.
- Population layer above the individual layer
- Individual risk is one product. Where disease is spreading, where exposure concentrates geographically, and what a health authority should do about it is another.
The data problem, honestly
Clinical risk modelling looks like a modelling problem and is mostly a data problem. Thousands of candidate attributes per patient. Severe class imbalance — the event you care about is, thankfully, rare. Missingness that is not random, because who gets tested is itself a clinical decision.
We would rather talk about that than about an accuracy number, because a model that scores well on a convenient sample and fails on the population is worse than no model — it is a confident wrong answer.
Grounded in the literature
The ensemble approach we apply to national-survey-class health data is published and independently cited — the methodology appears in BMC Medical Informatics and Decision Making and has been cited more than sixty times. We did not write that paper. We say so plainly, because a firm that blurs the line between "our method is published" and "we published it" should not be trusted with your clinical data either.
Where this is going
Our most recent healthcare work is an AI transformation assessment for a health-portal operator, delivered in 2025 with an academic collaborator. It is an assessment and a roadmap rather than a deployed model, and we would rather describe it that way than inflate it.
Tell us what decision you are trying to get right.
Not a discovery call about our capabilities. A conversation about the specific thing you need to predict, and whether the data you have can support it.