Falsified inspection data begins with a specification nobody could meet
Japanese manufacturing has been through many data scandals, and each is usually explained as a failure of individual ethics or compliance training. In my experience the cause sits earlier, in a commercial promise the process could never keep.
Over the past decade a long series of Japanese manufacturers, in steel, aluminium, rubber, automotive parts, electrical equipment and chemicals, have disclosed that inspection data given to customers had been altered, invented or reported without the testing it claimed. I will not name any of them. The pattern matters more than the names, and I am in no position to judge others: in November 2022 my own company disclosed that certificates of analysis for two grades of laminate resin at our Ichihara plant had been altered for eight years.
The standard explanation for these cases is ethical. Employees under pressure cut corners, supervisors looked away, compliance training failed. The remedies follow from the explanation: more training, a stronger reporting line, a new code of conduct and a statement from the president. All of those have their place. We did most of them. But after reading our own investigation report and many others, I believe the ethical explanation describes the last step of the failure and misses the first.
How a certificate becomes fiction
At Ichihara the first step was taken in 2012, when a customer asked for tighter limits on gel time and epoxy equivalent for two resin grades. Our sales team agreed, because the customer was important and a competitor was said to meet the new limits. The plant's engineers were consulted. Some of them doubted that the process could meet the limits consistently, but the qualification batches passed, and nobody wrote down the doubt.
The process could not meet the limits consistently. Most lots fell inside them; some did not. The plant had three choices for a lot that failed: hold it and ask the customer for a waiver, rework it, or ship it. Asking for a waiver meant telling the customer that the promise made by our sales team two years earlier had been wrong, and in the plant's culture that was not a conversation an inspection supervisor could start. Reworking cost time the production plan did not allow. So, beginning in 2014, some values on certificates were adjusted to fall inside the limits, and some measurements that the specification required were not taken. Over eight years this affected 2,140 lots shipped to 37 customers.
No one at Ichihara set out to deceive. They were caught between a commercial promise and a physical process, and they resolved the conflict in the only place they had any control, which was the certificate. That does not excuse what happened. It does explain why ethics training alone would not have prevented it. The people involved knew the rules. They did not believe they had permission to tell a customer that the company could not do what it had said.
Process capability belongs in the board pack
The question I now ask about any specification is not whether the plant passed the qualification batches but what the process can achieve on its worst reasonable day. Quality engineers express this as a capability index, which compares the natural variation of a process with the width of the specification. An index of 1.33 is a widely used minimum for a process expected to meet a limit with confidence. Below 1.0, the process will produce out-of-specification material as a matter of routine.
Before 2022 our board had never seen a capability index. It saw sales, margins, safety statistics and customer complaints. A board looking at those numbers would have had no idea that a meaningful share of our accepted specifications sat at or below 1.0, which is what our 2023 review found. Twenty-three specifications across our plants were in that position. Not all involved altered data; most were managed honestly through holds and rework. But each one was a place where a promise and a process disagreed, and where a plant would sooner or later face the choice Ichihara faced.
We now report process capability against accepted specifications every month in a capability review that I chair, and the audit and supervisory committee sees a summary every quarter. Quality assurance staff in every plant report to a chief quality officer who reports to me, not to the plant manager, so that a release decision cannot be overruled by the person responsible for output. Certificates are generated directly from laboratory instruments, and manual entry has been removed in all nine of our plants. These are mechanical changes, and mechanical changes are underrated. They remove the opportunity, not only the temptation.
The customer has to be told the truth about tolerance
The hardest change is commercial. If a specification cannot be met reliably, someone must tell the customer. In 2023 and 2024 we went back to customers on all 23 specifications. For most we agreed new limits that the customer's own engineers accepted once they saw our capability data; in several cases they discovered that their original limits had been set with more margin than their own process needed. For a few we invested to improve our process. Two customers declined to change their limits and moved their business to other suppliers, taking about ¥3.6 billion of annual sales.
The conversations were not easy for our sales teams, who had spent years assuring customers that every limit was comfortably met. Several customers asked, reasonably, why we had accepted limits we could not keep. The honest answer was that nobody in our company had been responsible for checking, and that answer embarrassed us more than any data we showed them.
I accept that loss. A supplier that cannot tell its customers the truth about what its plants can do will eventually tell them something false, and in our sector that day tends to arrive through a whistleblower and a timely disclosure rather than a negotiation. Since 2023 our sales teams have been required to attach a capability statement, signed by the plant's quality lead, to every new specification they propose to accept. Some of them dislike it, because it slows quotations. It is meant to.
For directors of other manufacturers, I would offer three questions. How many of our accepted customer specifications does the process meet with a capability index below 1.33? Who in each plant can stop a shipment, and to whom do they report? And when did a salesperson last go back to a customer to say that a limit we had agreed was wrong? If the answers are unknown, nobody, and never, the board has learnt where its next disclosure may come from.