Prior authorization exists because someone decided that a subset of expensive care should be reviewed before it happens rather than paid for afterward. Whatever one thinks of the premise, the implementation has been a fax-and-phone-call process consuming an enormous number of clinical hours, and it was an obvious candidate for software on both sides of the transaction.
Both sides duly deployed it. The result is one of the cleaner natural experiments in health administration, and the finding is uncomfortable: the process is dramatically faster per request and takes about as long as it ever did from the patient's point of view.
Automating a queue does not shorten it
The payer side automated review. Requests that meet clear criteria are approved without human involvement, which is genuinely better — a decision that took four days now takes minutes for the substantial share of cases that are unambiguous.
The provider side automated submission. Practices installed systems that identify which orders require authorization, assemble the documentation and file it, removing the friction that previously limited how many requests a clinic could physically produce. Volume rose accordingly, and the rise was not a distortion. Those were requests that should have been filed and previously were not, because nobody had the hours.
The requests that automation cannot resolve are the ones that mattered all along. Straightforward cases were never the burden; the burden is the case that does not fit a criterion, requires clinical narrative and ends in a peer-to-peer conversation between two physicians trying to find a mutual half hour. Automation increased the absolute number of those cases by increasing total volume, and the peer-to-peer call is not automatable by construction — it exists precisely because the automated path failed.
So the denials and appeals queue grew while the approvals queue shrank, and staffing moved rather than reduced. Practices report reassigning people from filing to appeals, which is more skilled work at higher cost, and several describe net administrative spending that went up. That expense flows through to premiums eventually, joining the broader repricing of insurance that is behaving like a second inflation.
The clinical cost lands where staffing is already thinnest. Primary care carries a disproportionate share of authorization work relative to its revenue, which adds to an arithmetic problem the specialty was already losing. Practices with the least administrative capacity generate the fewest requests, which reads in the aggregate data as lower utilisation and is in fact rationing by paperwork.
What would actually change the outcome is agreement on the criteria rather than faster processing of disagreement. Gold-carding arrangements, which exempt clinicians with consistent approval records from review entirely, are the one intervention with results that are not ambiguous — they remove requests instead of accelerating them. Their spread has been slow because they require a payer to give up the review, which is the part of the process the payer values.
The measurement problem underneath is familiar from the effort to make hospital price data usable. Both sides can now report impressive numbers about their own performance. Neither number is the one the patient experiences, and nobody is accountable for that one.
Topics healthinsuranceautomation
