Post-market surveillance breaks at the point where volume outgrows attention. Most teams are already past it and working harder to hide the gap.
Nearly 2.9 million device adverse-event reports reached the FDA in 2025, about 7,900 a day. For any single manufacturer the number is smaller, but it climbs with every product code you add and every similar device your procedure includes. The standard response is to read faster, hire another analyst or quietly narrow what counts as "in scope." None of those fixes the method. They just delay the moment it fails.
The work behind a single signal
Picture the actual workflow. An analyst searches MAUDE for a product code. They export the results, paste them into a spreadsheet, then open each report and read the narrative. Medical-device narratives are dense and inconsistent. One describes a lead dislodgement in three clauses, another buries a patient death in a paragraph of device-history boilerplate.
For each one the analyst decides the failure mode, classifies severity, notes the patient outcome and flags anything that might escalate. Then they do it again. And again. A few hundred times a month, across products that don't fail in convenient batches.
This is skilled work done in a way that wastes the skill. The judgment matters. The copy-paste and the re-reading do not.
Where it actually fails
Three failures show up, and they're predictable.
The first is the missed event. Dense clinical narratives compete for limited attention. A record that changes the pattern can read much like a routine one. Volume hides the signal in plain sight.
The second is latency. By the time a quarterly review surfaces a trend, the trend is a quarter old. A competitor's recall reaches you through a news alert instead of the data. You respond defensively because you found out late, not because the information wasn't available.
The third is the PSUR crunch. The data exists, but it's scattered across spreadsheets, PDFs and database exports. Assembling it into a Periodic Safety Update Report means rebuilding the source trail from fragments every cycle before qualified review can begin.
None of these are competence problems. A good team with a manual method still hits all three, because the method assumes a human can hold the line against unbounded volume. They can't.
The honest counterargument
There's a real case for keeping humans on every report. Adverse-event narratives are ambiguous. Severity is a judgment call. A classifier that's wrong in a confident-sounding way is worse than slow-and-right, because it launders a mistake into a record.
That case is correct, and it's exactly why the answer isn't to remove the human. It's to remove the parts of the job that aren't judgment.
What scales instead
Automate the pipeline, not the decision.
Ingestion scales cleanly. A new product receives an initial MAUDE history within minutes, then synchronizes against openFDA's weekly refresh. Extraction scales too. A model can read a narrative and pull the failure mode, severity, patient outcome and FDA codes in seconds, with a confidence score and an explicit flag when the data is thin.
What stays human is the part that should: the review. Every matched summary lands in a queue where someone acknowledges, flags or rejects it, with a signed decision when the procedure requires one. The analyst reads a structured summary and a confidence score beside the source narrative. They spend their attention on the calls that need it, not on transcription.
The difference isn't speed for its own sake. It's that the bottleneck moves from "how many narratives can one person read" to "how many decisions need a human," which is a much smaller and more honest number.
The trade you're actually making
Every manual surveillance program is making a bet that volume won't outpace the team. That bet gets worse each quarter. More products, more competitors, more reports, same number of hours.
Automating the mechanical layer doesn't replace judgment. It gives judgment more room. The team stops racing the inbox and starts working the signals that matter, with an audit trail that builds itself as they go.
DeviceWatch was built for that trade. The reading, classifying and assembling run automatically. The deciding stays yours.