Quality control in a medical courier business: how owners check the work

What to take away

Audit the evidence, not the impression. Pull actual run files at random and score them against your own written standard.
Four sources tell you different things: the file, the ride along, the site conversation, and the exception log. Any one of them alone will mislead you.
Score for completeness and repeatability rather than for effort. A record that another person can read and reconstruct is the whole test.
Do it monthly, on the same day, and act on one finding. An audit that produces no change is a report, not a control.

Nothing here states what any protocol requires. Score against the standard your customer's protocol and your own written procedure set, and confirm those with the bodies that issue them.

The monthly sample

Pull a fixed number of run files at random, across different drivers, sites and days. Randomly means genuinely randomly: choosing the runs you remember produces a flattering and useless result.

For each file, score six things as present or absent, with no partial credit.

Custody entries complete: who released, who received, times, signatures.
Every site on the route sheet accounted for, including sites with nothing collected.
Condition record present where the protocol requires one, and matched to the correct run.
Exceptions recorded where the log shows one occurred, and absent where it does not.
Legible and retrievable without asking the driver what it means.
Filed within the period your procedure states.

Six binary answers per file, tracked month over month. That trend is the most honest quality measure a courier business has, and it costs an hour.

Why binary scoring

Partial credit destroys the measure. A custody record with a missing signature is not eighty percent complete: it is a record that will fail when it matters. Scoring it as a pass with a note lets the same gap persist for a year.

The corollary is that a poor early score is expected and useful. Owners who introduce this and score badly in month one are looking at the truth for the first time.

The ride along

Once a quarter, ride the route. Not to check the driving, which is not the issue, but to see the four things files never capture:

What actually happens at the slow site. Who is asked, how long it takes, whether the material is genuinely not ready or simply not handed over.
Whether the packaging inspection happens. In files it always did. In practice it is the step most quietly dropped under time pressure.
How the driver behaves in clinical space. Waiting quietly, or filling silence. Staff notice, and they tell the laboratory.
What the driver has worked around. Every experienced driver has a workaround for something your procedure gets wrong. Ask, and treat the answer as information rather than as a fault.

That fourth point produces more useful change than the other three combined.

The site conversation

Ask two or three site contacts the same short question each quarter: what would you change about how we collect from you?

Do not ask whether they are satisfied. Satisfaction questions produce polite answers. A request for one change produces the specific thing that has been irritating somebody for months, and it is usually cheap to fix.

Record what you hear and, more importantly, tell them what you changed. A site that sees a complaint acted on becomes an ally at the moment the account is reviewed.

Reading the exception log

The log is the fourth source and the one most often filed and forgotten. Read it as a pattern rather than as a list.

Pattern	Likely cause	First action
One site, repeatedly	That site's internal process	A conversation about readiness at a fixed time
One driver, many sites	Training or route load	Ride along, then check the route design
One cause, many drivers	Your procedure	Rewrite the step, then retrain
Rising exception count overall	Route has outgrown its window	Re-sequence or split, do not push harder
Zero exceptions for weeks	Reporting has stopped	Ask why, and make reporting safe again

The last row is the one to take seriously. Nobody has a perfect month, and a log that goes quiet is telling you that reporting has become uncomfortable rather than unnecessary.

Turn findings into one change

An audit that produces a list produces nothing. Each month, pick the single finding with the widest effect and fix it properly:

Name the change, the owner and the date.
Update the written procedure, with a version date.
Retrain against the new version rather than sending an announcement.
Re-sample the same item next month to confirm it moved.

One change a month, verified, compounds. Twelve findings a month, unverified, do not.

The things worth scoring beyond the file

Route timing against plan, by site, so a slipping route is caught before a customer raises it. The design method is in designing routes around clinic closing times.
Vehicle and containment condition, since a worn closure is a quality failure waiting to happen. The checking regime is in the equipment and setup guide.
Cover readiness, tested rather than assumed. Once a quarter, have the substitute driver run the route.
Packaging inspection rate, observed rather than reported. This is what the ride along is for, and the classification and packaging expectations behind it belong with PHMSA's overview of transporting infectious substances and its guidance on transporting infectious substances safely, alongside the laboratory's own instructions.

Keep the audit results

File the scores, the findings and the changes together, dated. Two audiences will ask: a laboratory quality reviewer, who will find a self audit history far more convincing than assurances, and you, when a customer asks whether a problem is new. The IRS guidance on records a business should keep covers the retention baseline for the business records these sit alongside, and the laboratory's protocol may set a longer period for the operational ones.

The operating structure that all of this measures is described in how a courier operating day is built. If quality problems turn out to be caused by a territory that was never dense enough to serve properly, that is a different question, addressed in which local markets support a courier route.

Common questions

How many files should I sample?

Enough that a single unusual run does not swing the result, and few enough that you actually do it every month. Consistency in the number matters more than the number itself, because the trend is the output.

Should drivers know which runs are sampled?

They should know that sampling happens and not which runs. Announced audits measure preparation; unannounced ones measure practice.

What if the customer audits me first?

Then your own audit history is what you produce, and it changes the tone entirely. An operator who found and fixed a problem before the customer did is treated very differently from one who is hearing about it.

Is a perfect score a good sign?

Not on its own. Check that the standard is demanding enough to fail something, and check the exception log. A perfect file score alongside a silent exception log usually means the recording, not the work, is what is perfect.