AI in healthcare analytics: what it does here, and what it does not.
A plain-language view for health plan, state, and health system leaders deciding what to believe and what to buy.
Where AI helps in analytics
- Data normalization. Mapping inconsistent codes, names, and formats from different systems into one structure, so financial and clinical data can be analyzed together.
- Pattern and outlier detection. Surfacing providers, members, or claims that differ from their peers, so a person can look closer.
- Risk scoring. Ranking members or accounts for follow-up, using stated, reviewable factors.
- Forecasting. Projecting volumes and spending from history, with the assumptions visible.
Where it does not belong
- Clinical decision-making. Our AI-assisted features support analysts. They are not used to make clinical decisions.
- Replacing validated measures. HEDIS and regulatory measures follow defined specifications, and our dashboards are built around reviewed logic, not free-form model output.
- Unreviewed conclusions. A flagged provider or claim is a lead, not a finding. A person reviews it.
AI-assisted, and also modeled by hand
Our platform uses AI, and we are specific about where. The AI-assisted features run on AWS Bedrock, inside our AWS environment, and help with tasks such as normalizing data, finding patterns, and scoring risk.
A great deal of the analytics is not AI at all. Much of it is modeled and coded by our team, based on four decades of analyzing hospital data: how measures are defined, what the thresholds are, and how forecasts are built. That keeps results repeatable and reviewable, and it means a person can explain where a number came from.
Eight questions to ask any healthcare AI vendor
What data does it learn from, and whose is it?
Is your data used to train models that serve other customers?
Where is the data stored and who can see it?
Ask for the hosting arrangement and the business associate agreement.
What exactly is the model deciding?
Distinguish a ranking that helps an analyst from an automated decision that affects a person.
Can I see why it flagged this?
If the reason cannot be explained to an auditor, treat the output with caution.
How is accuracy measured, and on whose data?
Ask for the method, not only the headline number.
Who reviews the output?
Look for a named human step before anything is acted on.
What happens when it is wrong?
Ask how errors are found, corrected, and communicated.
What certifications are held, and which are only planned?
Ask for the difference in writing.
How we handle data
We are HIPAA compliant. The environment where we load claims and clinical data is hosted on AWS using HIPAA-eligible services. Our team works with yours to decide how data is provided securely, under the right agreements, before protected health information moves. Work can begin on de-identified or synthetic data.
Status. SOC 2, HITRUST, and ISO 27001 certifications are being pursued. Read the full picture on the platform page.
Bring a use case and a hard question.
We will show what AI-assisted analytics can and cannot do with your kind of data.