Measure traceability by the hours it takes to answer a recall question

Ask an aerospace supplier whether its parts are traceable and the answer will almost always be yes. It will be true in the sense that matters to an auditor reading a procedure: every part has a serial number, every operation is signed, every certificate is filed. My position is that this is the wrong test. Traceability should be measured as a performance, by how long it takes, how many people it needs and how many corrections follow when a customer asks a hard question about a suspect lot. A supplier that cannot state those three figures does not know how traceable its parts are.

I hold this view because of an autumn that went badly. I have described the case elsewhere on this site, but the lesson deserves its own argument.

Twenty-three days and thirty-one people

In September 2022 an engine customer asked Tessandier Aerospace which parts we had made since 2016 from fourteen heat lots of titanium bar whose mill certificates were under investigation. Every one of those parts was traceable in the procedural sense. The heat lot was on the material certificate, the certificate was referenced on the traveller, the traveller was signed at each operation and the serial number was recorded at final inspection.

The answer nevertheless took twenty-three days and thirty-one people in four plants. The data sat in three different execution systems, in scanned inspection files named by whoever scanned them and in paper travellers stored in boxes at two sites. Bar had been cut into billets at one plant and machined at another, and the link between bar heat and billet was in a spreadsheet maintained by one stores supervisor. Our first answer listed 1,162 parts. Eleven days later we corrected it to 1,140 because reworked parts had been counted twice.

Nothing in our quality system had failed an audit. Every record existed. What failed was the assembly of those records into an answer under time pressure, and that is the only form in which a customer ever experiences traceability. Nobody calls a supplier to ask whether its procedures are adequate. They call to ask about particular parts, usually on a Friday, usually with an aircraft or an engine shop waiting.

Design the record backwards from the question

Most traceability programmes are designed forwards. They start with the manufacturing process, decide what to record at each step and trust that the records, once complete, can answer any question. I now design backwards. We wrote down the questions customers, regulators and our own quality engineers actually ask, and then built the record that answers them fastest.

There are fewer of these questions than one might expect. Which parts came from this heat lot, this bar, this forging supplier? Which parts passed through this furnace load, this penetrant line on this shift or this machine during the week its probe was out of calibration? Where are those parts now: in our stores, at the customer or installed on an engine? Which inspector accepted them, against which revision of the drawing? Nearly every escape investigation I have seen is a combination of those questions.

Answering them quickly requires links rather than documents. Our serial genealogy record connects material certificate, heat lot, work order, each operation, each special-process run, each inspection and each shipment for a serial number, across all six plants, without replacing the plants' own systems. It is narrower than a full digital thread, and deliberately so. It also requires validity at source. A heat lot entered in the wrong format at a saw in Longueuil cannot be repaired by an analyst in head office three years later, so the operation now refuses to close until the field is valid. Heat-lot completeness rose from 91.4% to 99.7%, and that single change probably did more for our answer times than any software we bought.

Designing backwards also forces honesty about the past. Our record covers parts made since January 2020, plus digitised travellers for older parts still on active programmes. For parts made before 2016, an answer still takes days. I would rather say so on the first page of a board paper than let a director infer coverage from a programme title.

There are good objections to this approach. A genealogy layer above inconsistent plant systems leaves the inconsistency in place, and a single execution system across all plants would have given us cleaner data at the source. I agree, and I would choose that route at a company with fewer plants or a newer estate. At ours it would have cost nearly four times as much and taken years longer, while the question from our customer would not wait. It is also fair to say that measuring answer time can encourage a team to rehearse rather than improve. We vary the questions to make rehearsal hard.

A drill the board can watch

Each quarter our quality department writes a genealogy question without telling my team in advance. It might concern a heat lot, a furnace load or a gauge found out of calibration. We record when the question arrives, who works on it, when the answer goes back and whether anything in the answer is later corrected. The result goes to the audit and risk committee as three figures with a short note on what went wrong.

The drill in March 2026 covered 3,800 parts across four plants and took three hours and ten minutes, with no corrections afterwards. Earlier drills were less flattering. One in 2024 took nine hours and found two parts whose rework operations had been recorded against the wrong serial number. That result was more useful to the committee than any number of compliant audit findings, because it showed where the record was weak and what we were doing about it.

I think boards of manufacturers whose customers are regulated should ask for this kind of evidence. It is not a regulatory requirement, and I would not claim that any authority demands a particular answer time. Customer contracts, quality management standards for aviation suppliers and accreditation audits for special processes all expect complete and retrievable records; none tells a board how quickly its company can assemble them. That is a question for directors, because the consequence of a slow answer is commercial and reputational before it is ever regulatory. A customer that waits three weeks for an answer remembers it when the next programme is sourced.

The test I would offer another supplier's board is simple. Ask management to answer a real genealogy question, set by someone outside the data team, and report the hours, the people and the corrections. Ask what proportion of shipped parts the fastest answer covers. Then ask what it would cost to cover the rest. If the answers are uncomfortable, the company has learnt something its audit history could not tell it.