Three Days Between 'Lot Complete' and 'Lot Shipped' Is Almost All DHR Review. That's the Process to Measure First.

Sixty-two hours. That is the gap I keep finding at small device plants between the last operator signature on the floor and the shipping label going on the box. The product is built. It passed final functional test. It sits on a blue pallet in the finished-goods cage under a yellow QUARANTINE tag for two and a half more days while somebody reads paperwork.

Ask the quality manager why and you get an honest answer: the device history record review takes as long as it takes, and longest on the lots where somebody left a field blank.

Two and a Half Days of Finished Goods in the Cage

A device history record review is the last human gate before a lot ships: one qualified person reads every entry made on the floor and decides whether the paperwork proves the product was built the way the device master record says it must be. Not whether the product is good — the final test said that. Whether the record proves it.

Under the Quality Management System Regulation, which took effect on 2 February 2026 and pulled ISO 13485:2016 into what used to be Part 820, that record has to establish lot identity, quantities manufactured and released, dates, acceptance records, the primary identification label and labelling used, and the unique device identifier. Miss any of it and the lot cannot ship, however good the parts are.

This is the process to measure first for four reasons: high volume, identical repetition, hard start and stop times already recorded in your system, and countable failures. You do not need a study to find your baseline. It is sitting in your eQMS timestamps.

What the Reviewer Is Actually Looking For, Line by Line

Watch a quality engineer do this and it is not analysis, it is proofreading with legal consequences. Blank in-process check boxes. A missing second-check initial. An entry corrected with a scribble instead of a single line, initial and date. A torque driver whose calibration due date fell inside the build window. A traveler printed at revision C when the change order made revision D effective two days before the build started. An out-of-tolerance in-process measurement with no disposition next to it.

None of those means the device is bad. All of them stop the shipment. The fix, in most shops, is a kickback: the traveler goes back to the line, the lead finds the operator who worked that station on second shift, and the correction happens tomorrow, because that operator is not in until 3 p.m. That is the workaround everyone pretends is fine, and it is why the cage is full on Thursday and empty on Monday.

84% Report a Positive Return, and the Pattern Behind It Is Deliberately Boring

Deloitte's State of AI in the Enterprise 2026 found 84% of organisations investing in these tools report a positive return. The interesting part is not the percentage but how consistent the winning projects look: pick one messy, repetitive process, keep a human reviewing the output, and prove time saved or errors reduced before widening scope.

That argues for the device history record review over everything else on the shortlist. Demand forecasting is interesting and hard to prove. Predictive maintenance on the presses is real — 64% of manufacturers report using it — but the payback shows up over quarters and gets argued about. Record review pays back in hours, this month, and the argument ends because timestamps are not opinions.

flowchart TD
  A["Last operator signature on the traveler"] --> B["Traveler scanned into the eDHR"]
  B --> C["Assistant checks every field against the DMR"]
  C --> D{"Blank field, wrong rev, or expired calibration?"}
  D -->|Yes| E["Line lead corrects: strikethrough, initial, date"]
  E --> B
  D -->|No| F["Quality engineer reviews exceptions only"]
  F --> G["Lot released, label printed, invoice dated same day"]

The Baseline You Have to Capture Before You Switch Anything On

Run this for eight weeks before you change anything. Skip it and you will win the project but lose the argument, because six months from now the controller will ask what it bought and nobody will be able to say.

Capture five things per lot, and only five:

• Release cycle time in hours — last floor signature to release signature. Record median and 90th percentile, not the average. The tail is where the money is.
• Kickback rate — what fraction of lots go back to the floor for a documentation-only correction.
• Kickback reason, five buckets — blank field, missing second check, improper correction, wrong revision, calibration lapse. No more than five, or nobody will fill it in.
• Reviewer hours per lot. Have them log it honestly for eight weeks. Tell them why.
• Lots that missed the customer's promise date for documentation reasons alone — and what the expedited freight cost.

The number that settles the argument later is median release cycle time in hours, per lot, with the kickback rate beside it. Everything else is supporting evidence.

A Tuesday in the New Version

Second shift finishes lot 24601 at 11:40 p.m. and the traveler is scanned. Before the quality engineer arrives at 7 a.m., the assistant has read every field against the device master record and produced a one-page exception sheet: two blanks at station 4, a torque driver whose calibration expired eleven days into the build, and a flag that the traveler was printed at revision C while ECO-1188 made revision D effective on day two of the run.

She does not read 34 pages. She reads three exceptions. The calibration flag is the real one and becomes a nonconformance report and possibly a small extent-of-condition check. The blanks go back to the line lead at the 7:15 stand-up, while the second-shift operator is still reachable by text, instead of tomorrow afternoon. The revision flag is a false alarm — the change order had a 30-day use-up clause — and she writes that into the record once, where the next reviewer will see it.

Lot ships Tuesday. The invoice is dated Tuesday. That last sentence is the entire financial case.

The Arithmetic, in Hours and in Cash

Assumptions, illustrative: 34 lots shipped a month, average invoice value $18,500, reviewer loaded cost $64 an hour, terms Net 45 from invoice date, working capital at 9.5%. Reviewer time 3.1 hours per lot, kickback rate 31%, of which 40% break the promise date and cost about $220 in expedited freight.

Note which number is largest. It is not the cash-flow one. Owners reach for the working-capital story because it sounds financial; the reviewer hours are four times bigger and far easier to defend, because you measured them first.

What the Reviewer Still Has to Do Personally

The release signature does not move. A designated individual releases the lot, whether the exception sheet came from a person or a piece of software. Write it that way in the procedure and do not get clever.

Second, judgement stays human. "Is this an acceptable deviation?" is a judgement. "Is this field blank?" is a comparison. The tool does comparisons. The moment someone asks it to disposition a nonconformance you have crossed a line a notified body will find at your next surveillance audit.

Third — the one that actually bites — watch for rubber-stamping. If exception sheets come back clean for six weeks, reviewers stop reading them. Salt the system: once a month have the quality manager put a known documentation defect into a training copy and see whether it gets caught. Log the result. It is a legitimate effectiveness check and the only defence against the thing that quietly ruins this.

Frequently asked questions

Is the FDA going to have a problem with software reviewing a device history record?

Not with the review aid itself, provided two things are true. The software is validated for its intended use under ISO 13485 clause 4.1.6, with a record you can hand an investigator. And the release decision, with its electronic signature under 21 CFR Part 11, is still made by a qualified named person. Describe it in your procedure as a review aid that generates an exception list, because that is what it is. Trouble starts when a shop lets the tool close out lots and cannot produce a validation record.

How long should the baseline run before the number means anything?

Eight weeks minimum, spanning a month-end. Device shops push volume in the last week of the quarter, reviewers get squeezed, and the kickback rate rises exactly when you can least afford it. A four-week baseline taken in a quiet stretch understates your problem and makes the result look like noise later.

What if we still run paper travelers?

You can, but the first month of work is scanning discipline, not software. Fix the printer, standardise the scan settings, get operators used to writing inside the box. Plenty of shops find this step alone drops the kickback rate before anything is switched on — annoying, but a real measurable win you should take credit for.

Will this let me cut a quality headcount?

Probably not, and do not sell it internally that way. In every shop I have seen do this, the returned hours went into deferred work: overdue supplier audits, CAPA effectiveness checks that were being closed on paperwork alone, internal audits against the new clause numbering. Frame it as headcount reduction and your quality engineers will find reasons the exception sheets cannot be trusted — and they will be right often enough to win.

To start: open your eQMS, export the last 120 lots with two columns — last floor signature and release signature — and subtract. You will have your baseline median in twenty minutes, and you will probably be unpleasantly surprised. That number goes in front of whoever signs the purchase order.

One knock-on effect worth planning for: when lots start releasing same-day, customers notice, and the "where is my lot" calls come earlier and more often from planners chasing their own build schedules. CallSphere builds voice and chat agents that answer those calls on the first ring, take the lot and purchase-order number accurately, and route anything that sounds like a complaint to the quality on-call rather than a voicemail box. It does not review your records. It keeps the phone from undoing the days you just saved.