An OOS investigation on incoming material is a decision process, not a verdict on the batch. This guide walks the buyer’s path: freeze the lot, separate laboratory error from material failure, retest within the rules, and know when a third-party result is worth paying for.
The day an incoming test result lands outside the acceptance criteria, most buyers do one of two things. They reject the lot on the spot, or they quietly ask the laboratory to run it again and hope the second number is friendlier. Both moves feel decisive. Both cost money, and neither is what the written procedure actually expects of you.
An OOS investigation is a sequence of checkpoints, and each checkpoint exists because skipping it has burned somebody before. The traps are rarely technical. They are judgement errors made in the first days, while the batch sits in quarantine and the supplier is already on the phone asking for a decision.
This guide follows the path a buyer should walk, in order, and points out the traps at each step. It assumes your incoming testing is already in place, and it borrows the same logic regulators apply to finished products, because the FDA applies its OOS guidance to raw materials too.
An OOS Result Is A Decision Point, Not A Verdict
The first trap lives in the name. “Out of specification” sounds final, so buyers treat the result as a verdict and skip straight to disposition. Reject too early and you destroy your own leverage: the retained sample is still the best evidence you will ever have, and once the lot leaves your control, it is gone. Release too early, on the strength of a passing retest, and you have converted a quality signal into a downstream problem you will meet again in your finished product.
The correct first move is neither. Quarantine the lot, open an investigation file, and treat the OOS as a question that has not been answered yet. The question is always the same: does this result describe the material, or does it describe the test? Those two possibilities need different responses, and only the investigation tells you which one you are holding.
There is a third trap that hides inside the second. Some buyers “investigate” by re-weighing the sample, re-dissolving it, and injecting again until a number falls inside the limit. That is not an investigation. Regulators have a name for it, testing into compliance, and an investigation file built this way collapses the first time anyone reads the raw data.
The First 72 Hours: Three Questions Before Anything Moves
Before any retesting is planned, answer three questions in writing. They cost an hour. Skipping them is how a laboratory problem quietly becomes a supplier problem.
First, has this manufacturer lot been on site before? Companies assign internal lot numbers on receipt, so the same manufacturer batch can arrive twice under two different internal numbers. If a previous receipt passed your the inspection routine at the receiving dock, the impact assessment now has to cover material already consumed in production, not just the quarantined drums.
Second, was this lot on a reduced testing program? If your incoming plan skips certain tests based on supplier history, a confirmed OOS on the one parameter you did test puts every untested lot back in scope. That single sentence changes the size of the investigation from one batch to a season of receipts.
Third, is the sample history intact? Container condition on arrival, storage log, who opened it, how long it sat after opening. Gaps in that history do not tell you the result is wrong, but they tell you the result is unprovable, and unprovable is its own problem.
What Phase One Actually Settles
The first phase of an OOS investigation is a laboratory audit of itself. The FDA’s guidance describes it as hypothesis testing: analyst training, instrument performance, integration settings, calculation steps, reagent age. What it is not is a quiet second run of the same preparation. Every hypothesis gets tested against records, not against hope.
Two findings end Phase One. Either the laboratory finds an assignable cause, a wrong integration window, an expired reagent, a transcription slip, and the result is invalidated with the evidence documented, or it finds nothing, and the OOS stands as a true material signal. Both are clean outcomes. The trap is the middle ground: findings that are really retractions dressed as root causes. “Analyst error” with no training record attached is the classic example. If the paperwork would not survive an inspector reading it aloud, it is not a conclusion.
One more check belongs in this phase, and buyers forget it. The method itself has to suit the sample as received. A procedure validated on fresh material may misbehave on a lot that travelled three weeks through a humid port. That question sits next door to whether the method suits the sample, and it is the cheapest explanation to rule out before anyone blames the chemistry.
Keep the Phase One file narrow and honest. It is tempting to pad it with every instrument log in the laboratory, but an investigation that documents everything documents nothing. Five records that close the hypotheses beat fifty that merely decorate the folder.
When Your Result And The Supplier’s Certificate Disagree
Now the harder case. Your laboratory says the lot fails; the supplier’s certificate said it passed. Both laboratories can be competent and both numbers can be honest. Before the dispute escalates, check the three places where honest numbers disagree.
Methods first. Different versions, different calculation bases, different reporting conventions, results that only look comparable. Then the certificate itself: tools that test whether the certificate itself is genuine earn their keep here, because a fabricated document is rarer than a careless one but far more expensive. And then the basics: what each line of the certificate actually records, because half of the disputes trace to a line that was read as something it was not.
Where the disagreement survives all three checks, the acceptance criteria themselves become the question, and it helps to go back to where the acceptance criteria came from. A limit set years ago for a different route may be testing a property the current process does not control. That is not an excuse for the result; it is context for the decision.
| Situation | First signal | Buyer action | Typical end point |
|---|
| Method or basis difference | Different method versions or calculation bases between the two laboratories | Exchange method sheets and calculation basis; run the comparison on one retained sample | Results reconcile; no material failure |
| Sampling or handling gap | Unclear sample history, damaged container, long storage after opening | Reconstruct the sample record; resample under the agreed procedure if justified | Split sampling with witnesses; fresh results |
| Genuine material difference | Same method, same basis, and the result still fails | Freeze the lot; request the supplier’s Phase One report; plan third-party testing | Disposition under the quality agreement |
Retesting Rules The Guidance Actually Sets
Retesting is allowed. What the rules constrain is how. The sample tested must be the original preparation or a freshly prepared portion of the original sample, not a new drum pulled from quarantine. The number of retests is fixed before the work starts, written down, and justified, because a limit invented after the third attempt is not a limit. And the original OOS result stays in the file no matter how clean the retests look; a passing number next to a failing one is evidence, not a replacement.
Averaging is the trap that catches the most buyers. The FDA guidance is explicit that averaging an OOS result with retest results is inappropriate, because an average hides the very variability the investigation is supposed to explain. Two numbers, one failing and one passing, do not average into compliance. They average into an investigation finding.
One distinction belongs in your vocabulary before the supplier call. An OOS breaks a limit; an out-of-trend result stays inside the limits but wanders off the pattern your previous batches set. The trend failure is the quieter signal and the easier one to wave through, because every individual number passes. A lot that sits within specification but outside the historical range is telling you the process has moved, and the conversation is different from the one you have after a single hard failure.
If the laboratory phase closes with no assignable cause, the investigation moves to production: the supplier’s batch record, their deviations, their process history. From the buyer’s side, this is where you formally request the supplier’s own investigation and compare it against the history of previous batches. A lot that fails in isolation is a batch problem; a trend across lots is a process problem, and the two need different conversations. Where the trail ends in rejection, the paperwork follows the same discipline as how a rejection is documented and closed, so the decision is auditable long after the drums have gone.
The full reasoning lives in the FDA’s guidance document, Investigating Out-of-Specification (OOS) Test Results for Pharmaceutical Production, issued in May 2022. It applies to APIs and raw materials, and it is worth reading once rather than relying on summaries, because most of its value is in what it refuses to allow.
When A Retained Sample Is Worth Sending Out
Third-party testing is not a default escalation. It is a specific tool for a specific situation: the two laboratories disagree, both files look competent, and the decision is big enough that a third number changes what happens next. Before spending the money, run the same pre-agreement logic that applies to any arbitration. The method and its version are fixed in writing, the calculation basis is agreed, the sample is the retained portion with an unbroken custody chain, and both parties accept in advance what the result will mean for the lot.
Sent without those agreements, a third-party result is just another opinion to argue about, which is precisely the outcome you were paying to escape. The mechanics of sending the dispute to a third-party laboratory, and the limits of what such a result can prove, deserve their own read before the sample is boxed. What a third party cannot do is rebuild a broken custody chain or fix a method that was never agreed.
Expect the supplier to run their own investigation in parallel and to send you the conclusions. Read the investigation summary your supplier sends back against the same standards you apply to your own: hypotheses tested against records, retests numbered in advance, no conclusions that would embarrass anyone in front of an inspector. A supplier whose Phase One reads like an acquittal deserves the same scrutiny as a result that reads like a failure.
What An OOS Event Changes In Your Supplier File
The investigation closes, the lot is dispositioned, and the temptation is to file everything and move on. The last trap is treating the OOS as a closed event rather than an input. It belongs in the performance file you keep on suppliers, not as a grudge but as a data point: confirmed material failures count differently from laboratory errors that were properly found and corrected, and both count differently from disputes that were never resolved.
It should also travel back into your agreements. If this OOS took weeks to reach a decision because the notification obligations were vague, tighten the notification clauses you agreed to before the next batch ships, not after the next failure. And if the investigation surfaced a process step the supplier never mentioned, cross-check it against the change notices on file, because an undisclosed modification has a cousin called the change notice nobody sent.
Handled this way, an OOS investigation costs what it costs and buys two things back: a defensible disposition for this lot, and a sharper picture of the supplier you are betting future batches on. Skipped at any checkpoint, it costs the same amount and buys neither.
