4 Comments
User's avatar
William Drury's avatar

The post focuses on AI overpaying claims, but the mirror-image failure is already documented: UnitedHealth’s nH Predict tool, which plaintiffs say carries a 90% error rate (nine of ten denials reversed on appeal), stayed in use because only about 0.2% of policyholders ever appeal. That suggests a leakage checker is only half the audit, since an insurer’s incentives make wrongful denials far less likely to get flagged than overpayments. Causal attribution could help there too, because a denial pattern that survives only because nobody appeals looks very different from random process noise. I summarized the UnitedHealth case, including its lobbying and the 5% Medicare Advantage rate boost, in a related piece: https://wdrury.substack.com/p/wicked-shady-list-s-and-p-500-kickoff

Ari Joury's avatar

This is a very interesting article, William, thanks for sharing. I was turned off by the title, because it seemed one-sided. However, what you say about UnitedHealth is interesting to say the least. Indeed causal inference would be helpful there -- though I'd argue that for a model with 90% error rate, some more basic fixes are apt as first steps!

William Drury's avatar

Absolutely. Even a coin flip would potentially be better. But I wonder if UNH is actually incentivized to fix this, given that it is better for their bottom line? In the current pay-to-play regulatory environment they have paid, and so there might be no consequence to them mistreating their own customers. Love your 'stack, keep em coming!

Ari Joury's avatar

Certainly the incentives are skewed here, and I have no idea how to fix those... It's unlikely that regulators would change this any time soon; even if they wanted to it would take years to implement.

Thanks for the love!