AI predictive models can't predict what they don't know: who has died

Most data records activity, not death. When an individual dies and their record just stops, an AI predictive model reads that silence as someone still alive. Veritas supplies confirmed deaths, deidentified and matched to your data with tokens, so your models train and validate on labels that match reality.

Talk to our team

45K+

independently verified sources feeding the Index

90%

of U.S. deaths captured within two weeks of occurrence

205M+

records available, with data back to 1935

Missed deaths are a labeling error in AI predictive models

Health, claims, and consumer data record activity, not deaths. When an individual dies outside a hospital, the record stops, and the model treats that individual as still alive. These gaps are hard to see in the data itself, so the error can stay in a model long after it ships.

Survival and risk models

An individual who died but was never recorded counts as a survivor. The model learns that a fatal risk profile is safer than it is.

Outcome and churn models

A death after discharge counts as not readmitted. A death in a customer base looks like churn. Either way, the model learns the wrong outcome.

Validation sets

If the holdout set has the same gaps, a model that misses deaths still scores well, and the error goes unnoticed.

Deployed models

Without current mortality data, a model in production keeps scoring individuals who have died, and every downstream decision carries the error forward.

Confirmed deaths at every stage of an AI predictive model's lifecycle

Training data

Label outcomes with confirmed deaths from 45,000+ sources, not proxy signals like inactivity or closed accounts. Delivered deidentified and matched to your data with tokens.

Validation and benchmarking

Test predictive model performance against verified mortality endpoints, with full provenance on every death. Add Cause of Death to separate all-cause from condition-specific outcomes.

Bias audits

Measure how missed deaths shift model performance across subgroups before the model ships.

Production monitoring

Ongoing Monitoring removes individuals who have died from the populations your models score, so deployed models stay accurate.

Trusted by teams at

Compliance

Collected, stored, and delivered through SOC2 Type II and HITRUST r2-certified workflows. Veritas is also a QECP-approved qualified entity for Medicare claims data and a GovRamp member.

See how much mortality your AI training data is missing