Originally published on Medium.
In the off-chance you need to recall product, would you even know where to start?
In September 2026 Ford recalled 11,405 Rangers over potentially counterfeit capacitors, and could name every truck. In the same month a contract manufacturer in Haridwar recalled every tube of toothpaste still in date it had shipped in two years, because it could not name one.
Neither company had a worse problem than the other. One of them had better records.
That difference has a name nobody uses, and it is the number that decides what a withdrawal costs: recall width. Not whether you have a problem. How much product you have to pull when you do.
The two cases, side by side
Ford’s recall is the one worth studying, because it is the only genuinely good outcome in a year of them.
A multi-layer ceramic capacitor on a centre-stack display board, from a tier-two supplier, failed both engineering and material specifications. A cracked capacitor shorts, blows the display fuse, and takes out the screen and the rearview camera feed with it. That is a safety defect, and the parts were described in the filing as potentially counterfeit.
What bounded it was a chain: the suspect capacitor lot connected to the printed circuit board’s serial number, the board to the centre stack assembly serial number, and the assembly to a vehicle. Lot to board, board to assembly, assembly to truck. The result was a recall with a build window, a truck count and a mailing list, and an estimate that about 47 per cent of those trucks actually carried the defect.
Now the other one. FDA’s warning letter of 27 October 2026 to Pritam International Private Limited, a contract manufacturer of over-the-counter toothpaste for the American market, records that investigators saw visible microbial growth on media plates in an incubator, and that the following day “there was no visible microbiological growth on plates labeled as the same samples observed the day before.” The firm’s own response described “a deliberate but isolated act by the microbiologist to circumvent your procedure.”
The letter also records that across its shipments of sorbitol, a material carrying a diethylene glycol contamination risk, the firm had one certificate of analysis showing the required identity results. FDA’s position is flat: “Identity testing, including DEG and EG, is required for each lot that you receive.”
On 10 September 2026 the company recalled all toothpaste still within expiry that it had shipped between September 2024 and August 2026.
Two years. Not because two years of product was bad, but because when the laboratory records cannot be relied on and incoming materials were not identity tested, there is no narrower answer available. A recall is only ever as precise as the evidence behind it, and the evidence was the problem.
Width is set years before the recall
Here is the part that matters to anyone reading this from inside a manufacturer.
On the day a regulator calls, your recall width is already fixed. You cannot improve it in the meeting. It was set by decisions taken quarters or years earlier, by people optimising for something else: whether to accept a supplier certificate instead of testing, whether to carry component serials through assembly, whether to hold a contract manufacturer’s batch records or just their certificates.
None of those decisions feels like a recall decision at the time. Every one of them is.
And the cost difference is not marginal. Ford pulled a defined population with an estimated defect rate under half. Pritam pulled everything it had shipped in two years, almost all of which was presumably fine. Same category of failure, radically different bill, and the variable was documentation.
The four widths
In practice there are four answers to “what would you have to pull”, and every manufacturer sits in one of them.
Unit level. You can identify individual affected units. Rare, and it is the only position where the recall matches the fault.
Lot level. You can bound it to batches. This is where most regulated manufacturers are, and it is a working position. The exposure is that a lot is a lot of units.
Window level. You would pull a production window, because nothing in your records separates good from suspect inside it. Expect months.
Everything in date. No narrower answer exists. If a supplier’s records are discredited, you withdraw all of it. This is where Pritam ended up, and this is where the eight-figure numbers live.
Most companies have never worked out which band they are in, which is odd given that it is a cost figure and finance would act on it immediately if anyone produced it.
So we built the ten questions
This is the part where an article usually tells you to think carefully about traceability and then ends.
Instead: we built the assessment, and it is free and takes three minutes. Ten questions, each one drawn from a documented case in the last three years, answered about your own operation. It computes which of the four bands you are in and shows you the three answers that put you there.
Nothing is collected. No email, no sign-up, no form. The answers stay in your browser and the result renders as text you can copy into a meeting. We built it that way deliberately, because a lead form in front of a diagnostic is a tax on the people most likely to need it.
A few of the questions, so you know the shape:
- For your highest-risk incoming material, how many of your last twelve lots did you identity-test yourself, rather than accept on a supplier’s certificate?
- Can you link a finished unit back to the specific lot of every critical component in it, without a manual file search?
- If a regulator published an alert naming your product tomorrow, could you state what is actually in the units within 48 hours?
- Have you confirmed in writing that your insurance covers a withdrawal compelled by a regulator?
That last one catches people. A US court held in 2023 that a recall forced by the FDA was not an “accident” within the meaning of a general liability policy, so the policy was never triggered. The claimant had a $13 million consent judgment and recovered nothing on the insurance. If you have assumed your general liability covers a compelled withdrawal, that assumption is the single most expensive unchecked item on this list.
What the assessment will probably tell you
We have run the ten questions against every case in our own files from the last three years, and the pattern is consistent enough to predict.
Companies do well on customer traceability, because distribution is a commercial system and somebody owns it. They do badly on incoming material, because testing costs money and a supplier certificate is free. They do worst on the question of who physically made each component, as opposed to who invoiced them, which is the question contract manufacturing makes hard and nobody asks until a regulator does.
If you score badly on incoming material and well on everything else, you are in the window band and you probably think you are in the lot band. That gap is the one worth closing, and it is cheaper to close than most people expect, because it is usually three or four materials that carry the real risk rather than the whole bill of materials.
What we are not saying
We are not saying good records prevent failures. Ford’s capacitors still reached an assembly line. Records decide the size of the response, not whether there is one.
We are not saying Pritam is unusual. It is in the public record because FDA published a letter, which most comparable situations never produce.
We are not saying the assessment is a product pitch with a quiz on top. It gives you a result and a list of your own weakest answers. If it tells you that you are in good shape, that is the result, and you owe us nothing.
And we are not saying that what we sell closes all four bands. It does not. Unit-level identity narrows the width of what you can find and name in the field. It does nothing about whether the sorbitol was tested, and we have said so in this series repeatedly, including about cases that ended in criminal convictions.
Who this is for
The quality lead who could answer eight of the ten questions today and would rather know which two before somebody else asks.
The supply chain director who has a traceability project competing with four other projects and needs a number to defend it.
The CFO who has never seen recall exposure expressed as a population rather than a provision.
And anyone who read the Ford recall this year and assumed that level of precision was normal. It is not. It is the exception, and it was bought years in advance.
Three minutes
The assessment is here, it is free, and nothing is collected: How wide would your recall be?
If the result is uncomfortable, the three weakest answers it gives you are the shortest list of things worth fixing that we know how to produce. If you want a second pair of eyes on those three, that is what we do.
Aeroz, Making Authenticity Checkable. Visit aeroz.io to learn more & get in contact with our team via [email protected].
© 2026 Aeroz. All rights reserved.